Abstract:Remote sensing is of unique advantages in quickly obtaining and analyzing information such as crop types, planting areas, and yields duo to its rapid, macroscopic, non-destructive and objective observing characteristics. The crop spatial distribution map, planting area, and yield information extracted or interpreted by remote sensing can serve many agricultural applications such as resource supervision, information census, insurance and investment, and precision agriculture. The research status, problems and future potential research directions of crop type identification and yield estimation using remote sensing were summarized. Firstly, the research status of crop type identification was summarized from aspects of identification features and classification models. In view of the core problem of the lack of crop-wised identification feature knowledge, deep learning methods were proposed to be used to collaboratively learn the feature of “temporal-spatial-spectrum” in the process of crop growth, and finally a knowledge graph for crop remote sensing identification was constructed, so as to solve the problems, identification accuracy and identification efficiency, that affected current crop type identification using remotely sensed imagery. Secondly, by summarizing characteristics of three types of crop yield estimation models (i.e., empirical statistical model, remote sensing photosynthesis model and crop growth model), highly integrating crop growth model and deep learning methods were proposed to forecast crop yield which may be a valuable potential solution in the future, under the circumstance of the popularization of high spatial, high spectral, and high temporal-resolution data and the development of deep learning technology. Because crop growth model was of strong mechanism and deep learning methods were capable of learning complex problems. In the future, crop growth models can be used for point-scale simulation to drive deep learning methods to build yield forecasting model in complex scenarios, and finally a yield estimation model was achieved which used growth mechanism as constraints and deep learning model as spatial extrapolation.

Abstract:In the earlier stage, a stepped vibrating seed sowing mechanism in the seed supply link of the precutting cassava seeder was designed. In order to further realize the precision seeding of cassava seed stems, a seed sowing mechanism was designed behind the stepped vibrating seed sowing mechanism to complete reliable seed filling and precision seeding. However, it was blind to take a single kind of stem from the cassava seed stem group that had been adjusted and sequenced through the seed metering mechanism scoop, which was easy to cause problems such as missing filling and reseeding. The spoon chain seed metering mechanism was further designed to solve the problems of seed filling difficulty and low seeding qualification index of the current precutting cassava precision seeder. The working principle of the seeder and the parameters related to the design of the spoon chain metering mechanism were described. Based on the theory of the fastest descent line, the parameters of the seed scoops of the metering mechanism were designed. The forces on the cassava seed stems and the motion state during the seed filling and seed feeding process of the mechanism were theoretically analyzed. It was determined that the significant factors affecting the seed filling performance were the type of seed scoops, the number of seed scoops, the motion speed of the conveyor chain, and the seed filling angle. Single factor simulation was carried out by using EDEM software, and the influence rules of different test factors on seed filling performance were obtained. The response surface BBD simulation test was carried out to determine the optimal factor parameter combination. The prototype seeder was developed for bench and field tests. The results showed that when the seed filling angle was 37°, the number of seed scoops was 12, and the moving speed of the conveyor chain was 0.63m/s, the qualified index of seed filling was 93.8%, and the missed filling index was 1.9%. The field experiment showed that the spoon chain seed metering mechanism of the precut seed vibrating seed feeding cassava seeder had better performance and it can meet the agronomic requirements of precision cassava sowing.

Abstract:Fertilization stability is an important index to evaluate the performance of variable rate fertilization system. In order to solve the problem that the precision of fertilization is reduced due to obvious fluctuation during fertilizer discharge with external trough wheel type fertilizer drainer. A piecewise PID control method using fertilizer flow rate as feedback value was proposed. And a high-precision fertilizer flow control system of fertilizer planter was designed. The system regarded the real-time fertilizer flow value collected by the fertilizer flow detection module as the feedback value of the fertilizer flow controller in the vehicle terminal. Controller calculated output based on target and real-time fertilizer flow, and sent to fertilizer motor through USB to CAN module. Accurate control of fertilizer flow rate was realized. Fertilizer application test stand was built. Mathematical model between fertilizer and accumulative value of capacitance was established and validated. The results showed that the maximum measurement error of sensor was 1.20% when using this model, which met the requirement of fertilizer flow detection. Indoor bench test on response of fertilizer flow rate change and fertilization accuracy was carried out. The results showed that the response time of the fertilizer flow control system was 1.42s, mean value was 0.98s, maximum overshoot value was 3.49%, mean value was 2.82%, maximum steady-state error was 0.89%, mean value was 0.64%, minimum fertilization accuracy was 97.83%, and mean value was 98.14%. Under different test conditions, the accuracy of fertilizer flow control and fertilization of fertilizer flow control system was better than that of constant speed system. Field tests showed that at vehicle speeds of 4km/h, 6km/h and 8km/h, the accuracy of fertilizer application rate of fertilizer flow control system was 97.84%, 97.78% and 97.82% respectively, and the average accuracy of fertilizer application rate was 97.81%, and the standard deviation was 0.28%, which met the requirements of fertilizer application accuracy of fertilizer system.

Abstract:In order to improve the utilization efficiency of fertilizers, reduce the amount of chemical fertilizers applied, and realize fertilization in the root zone of rapeseed, combined with agronomic requirements for rapeseed sowing and fertilization, the side deep hole fertilization process for rapeseed was proposed, and a kind of mechanical hole fertilization device was designed. The working process of the hole fertilization device was explained, the basic parameters of the device were determined through theoretical analysis, the mechanical models of fertilizer particle group in fertilizer filling and feeding zone were established, and the main influencing factors affecting its fertilizer distribution performance were determined. The discrete element software EDEM was used to carry out a simulation analysis on the fertilizer distribution performance of the fertilizer point-applied device, and the influence of rotation speed of fertilizer feeding unit, length of fertilizer hole and fertilizer tube on the error of hole fertilizer amount and the fertilizer distribution long axis was studied. The optimized result showed that when the fertilizer feeding unit speed, the length of the fertilizer hole and the fertilizer tube were 60r/min, 18mm and ABS fertilizer tube, respectively, the error of hole fertilizer amount and the fertilizer distribution long axis were 7.05% and 62.45mm, respectively. The bench test showed that under the conditions of rotation speed of fertilizer feeding unit of 30~90r/min, the error of the hole fertilizer amount was 4.56%~15.69%, the fertilizer distribution long axis was 76.32~91.50mm, the stability coefficient of fertilizer distribution long axis was 4.53%~9.78% and the error of fertilizer pointed distance was 3.24%~7.31%, respectively. Field test results showed that when the fertilizer feeding unit was 30~90r/min, the error of the hole fertilizer amount was 4.73%~16.07%, the fertilizer distribution long axis was 85.21~101.65mm, the coefficient of variation of the fertilizer distribution long axis stability was 4.82%~10.63%, the error of the fertilizer pointed distance was 3.36%~7.58% and the stability coefficient of fertilization depth was 6.43%~10.85%, respectively. The cavitation performance was well, and it met the requirements of hole fertilization.

Abstract:In order to improve the stability and uniformity of paddy field side deep fertilizing and discharging apparatus, enhance the ability of fertilizer regulation, and ensure the efficiency and quality of paddy field side deep fertilizing apparatus, a kind of conical-disc and push plate double-row fertilizer apparatus was designed according to the agronomic requirements of paddy field fertilizing in Heilongjiang Province. The working principle of the fertilizer apparatus was described, the mechanical models of different stages of fertilizer were constructed, and the structural parameters and critical speed were determined. The influence of the number of push plates on the fertilizer filling capacity and fertilizer discharge performance was simulated and analyzed by using the discrete element software EDEM. It was concluded that when the number of push plates was 8, the fertilizer discharge apparatus had the best fertilizer discharge performance. The full factor test method was used to carry out the bench test of the fertilizer discharge capacity and performance of the fertilizer apparatus under the condition that the rotating speed of the conicaldisc ranged from 15r/min to 45r/min and the opening of the fertilizer discharge port ranged from 5mm to 25mm. The results showed that the range of fertilizer discharge was 122~934kg/hm2, which had a high linear correlation with the rotating speed of conical-disc and the opening of fertilizer discharge port, and had the highest correlation with the rotating speed of conicaldisc. The variation coefficients of consistency of double row fertilizer discharge, stability of total fertilizer discharge and uniformity of fertilizer discharge ranged from 1.01% to 3.88%, 1.05% to 3.81% and 6.64% to 15.79%, respectively. The maximum variation coefficients of consistency of double row fertilizer discharge under the inclined state of fertilizer apparatus was 6.17%. The experimental results met the requirements of paddy field side deep fertilization performance. The research result may provide a reference for the implementation of paddy field side deep fertilization technology and the design of disc fertilizer apparatus.

Abstract:Pesticide spraying in orchard is an important inspection content of fruit quality and safety, and the reliable record of pesticide spraying behavior is an important part of fruit traceability system. Aiming at solving the problem that it was difficult to accurately grasp the real situation of pesticide application in the farmer professional cooperatives during the fruit planting in China, monitoring of the spraying behavior in orchard based on the interaction of human posture estimation and scenes was proposed. Firstly, the fine tuned YOLO v5 model was used to complete the precise detection of sprayers and fruit tree targets, and the features of scene interaction were extracted. Then, the OpenPose model was used to recognize human skeleton and extract human posture features. Finally, the distance and angle of the above features were calculated respectively, and fused into 11244 sets of feature vectors, which were trained by the SVM model to complete the detection of orchard spraying behavior. In order to verify the effectiveness of the algorithm, totally 92 videos with different illuminations, different distances, different numbers of people and different occlusion degrees were tested. The results showed that the ACC of the algorithm was 85.66%, the MAE was 42.53%, the RMSE was 44.59%, the RMSEP was 44.34% and the RPD was 1.56. Simultaneously, the effectiveness of spraying behavior recognition in orchard was validated under different illuminations, occlusions, distance change and single spraying among multiple people. Experimental results showed that it was feasible to apply the model to the detection of orchard spraying behavior. The research result could provide technical reference for the standardization and reliability of orchard management in the fruit traceability system.

Abstract:The level of mechanized harvesting of wheat in China has reached over 97%, and the impurity rate is one of the important indicators of mechanized wheat harvesting. In order to realize the online detection of the impurity rate in the wheat mechanized harvesting process, an online detection method of the wheat machine harvesting impurity rate was proposed based on the improved U-Net model combined with attention. Based on the wheat sample images collected by machine, the Labelme was used to manually label the images, and the images were enhanced by random rotation, scaling, shearing, and horizontal mirroring to construct a basic image dataset; an improved U-Net model combined with attention was designed. The model was classified and identified, and the offline training of the model was implemented under the torch 1.2.0 deep learning framework; the optimal offline model was transplanted to the Nvidia jetson tx2 development kit, and a quantification model of impurity rate was designed based on image information, so as to realize wheat on-line detection of impurity content in mechanized harvesting. The experimental results showed that the comprehensive evaluation index F1 of the improved U-Net model combined with attention was 76.64% and 85.70%, respectively, which were 10.33 percentage points and 2.86 percentage points higher than that of the standard U-Net, and 10.22 percentage points and 11.62 percentage points higher than that of DeepLabV3, which was 18.40 percentage points and 14.67 percentage points higher than that of PSPNet. Quantitative analysis of the detection results of impurity rate showed that in the bench test and field test, the average online detection of impurity rate of the device was 1.69% and 1.48%, respectively, which was higher than the manual detection by 0.26 percentage points and 0.13 percentage points. Qualitative analysis of the test results of impurity rate showed that whether it was a bench test or a field test, the test results of the device and the labor were all less than 2%. It was judged that the operation performance of the combine harvester during the test process met the national standards, and the test results were consistent. Therefore, the online detection method of wheat impurity rate proposed can provide technical support for the online quality control of wheat combined harvesting operations.

Abstract:During rapeseed direct seeding operation in rice-rapeseed rotation area of mid-lower Yangtze River, because the soil is sticky and hardened, the stubble of the fore crop of rice on the surface is high, and the large amount of straw is retained, which leads to the production problems of easy winding of rotary tillage part, low straw burying rate, easy hanging of grass and blockage of deep fertilization shovel, difficult to realize deep fertilization operation, a burying stubble and anti-blocking deep fertilization composite operation device for rapeseed direct planting was developed to adapt high stubble and heavy soil rice stubble field. The structural parameters of the deep rotary curved blade, the shallow rotary curved blade, and anti-blocking straight rotary blade of the burying stubble and anti-blocking blade roller, and the deep fertilization shovel were determined. Meanwhile, the arrangement and installation mode of the rotary blade and the deep fertilizing shovel were defined. EDEM simulation was used to analyze the straw burying and spatial distribution, the distribution depth after deep application of granular fertilizer, and the mixing results of soil particles in the plough layer soil after machine operation. The test results showed that when the working speed was 2.5km/h, the tillage depth was 150mm, and the burying stubble and anti-blocking blade roller rotating speed was 345r/min, the straw burying rate was 86.53% and the fertilization depth was 83~106mm. The validation test of four field working conditions of the rotary tillage and deep fertilization device was carried out. The field test results showed that the deep fertilization part had good anti-blockage performance in the field. When the rotary tillage and deep fertilization device operated on the sticky surface with high stubble, the fertilization depth was 87.4~109.5mm, the straw burying rate was 86.69%~90.35%, and the uniformity of the ridge was 16.48~22.65mm, the broken rate of soil was 81.24%~92.13%. which could meet the test requirements of fertilization depth and seedbed preparing during rapeseed direct seeding.

Abstract:Hilly and mountainous areas were the main potato planting areas in China. In view of the low efficiency of mechanized harvesting and separation and picking up of potatoes in small plots, a crawler type self-propelled sorting potato harvester was designed in combination with the requirements of potato planting agronomy and harvesting. The overall structure and key components of the prototype were designed, which were mainly composed of crawler drive chassis, excavation device, automatic alignment device, separation device and sorting device. The prototype had technical advantages such as crawler driven walking, high-frequency low amplitude vibration clod crushing, automatic row excavation, manual auxiliary sorting and hydraulic drive mode. On the basis of describing the overall structure and working principle, combined with the potato kinematics model and collision characteristics analysis, the structural parameters and operating parameters of the key components were determined to achieve the goal of high efficiency and low loss harvest. The separating screen was inclined at 30°, the distance between two adjacent gear rods was 210mm, the height of the stop bar was 25mm, the maximum rotating speed of separator driving wheel was 120r/min, the maximum line speed of the sorting screen was 0.65m/s, the drop height between the end of the separating screen and the beginning of the sorting screen was 120mm. As the method of manually assisted potato sorting and gathering was adopted, the number of drops and rolls of potato blocks was reduced, the separation stroke of potatoes was shortened, and the collision frequency and damage of potatoes in unit time were reduced, which was helpful to realize the loss reduction harvest. The prototype was tested in the field. The results of the field test showed that when the operating speeds of the prototype were 1.0km/h and 1.2km/h, the running linear speeds of the separating screen were 0.61m/s and 0.72m/s, the running linear speeds of the sorting screen were 0.42m/s and 0.50m/s, and the productivity was 0.10hm2/h and 0.12hm2/h, respectively. The average value of the three peak impact acceleration of electronic potato was 51.02g and 51.85g, the peak impact acceleration of potato was less than the critical damage threshold. There was no missing inspection and skin damage of potatoes, and the effect of the harvest was good. All performance indexes met the requirements of relevant standards.

Abstract:Aiming at the problems of high potato damage rate, low soil removal rate, single structure and inconvenient adjustment of the separation device of the traditional potato harvester, a left and right spiral symmetrical soil removal roller and adjustable type made of polyurethane material were designed. The smooth rollers were alternately arranged and combined in the potato harvester roller group conveying and separating device. Through the dynamic analysis of the machine body structure, the coupling mechanism analysis of potato-soil separation and the discrete analysis of the collision between potatoes in the soil removal process, the key factors affecting the potato damage rate and soil removal rate of the roller-type conveying and separating device of the potato harvester were determined. And it was tested, taking the potato damage rate and soil removal rate as the test index, taking the distance between the soil removal roller and the smooth roller, the rotation speed and the inclination angle of the conveying separation device as the experimental factors, a mathematical regression model was established for the orthogonal test results, and a response surface analysis was carried out. Through analysis and parametric analysis, it was determined that when the distance between the soil removal roller and the smooth roller was 16.5mm, the speed of the soil removal roller was 100r/min, the speed of the smooth roller was 100r/min, and the inclination angle of the separation device was 8°, the damage potato rate was 0.64%, and soil removal rate was 97.1%. Compared with the traditional potato harvester separation device, the potato damage rate was decreased by 0.12 percentage points, and the soil removal rate was increased by 2.6 percentage points. The device can meet the requirements of conveying separation.

Abstract:Directing at the problems that domestic sugarcane harvester header height adjustment and nonautomatic control, a servo control system for sugarcane header height was designed. The servo control system mainly consisted of a self-weight swinging profiling mechanism, STM32 controller, displacement sensor, upper computer module, keypad module, solenoid valve drive module, etc. The self-weight swinging profiling mechanism consisted of a ground contact, a connection sleeve, a left fixed connection plate, a frame, a right fixed connection plate, an angle sensor, etc. To address the problem that the harvester may cause impact or destroy the profiling mechanism when performing the reversing operation, ADAMS dynamics simulation software was utilized to obtain the vertical height and force variation of the profiling mechanism, and complete the optimization design of the tail end of the profiling mechanism. The simulation tests showed that when the radius of the tail end was 105mm, the maximum collision force of the ground on the profiling mechanism was equal to 1976N less than the permissible bending strength of 45 steel, which met the design requirements of the profiling mechanism. It was also verified that the profiling mechanism can adhere to the ground. To study the relationship model between the height of the harvester's cutting table and the signals collected by the profiling mechanism, and design a PID algorithm for cutting table height control. The PID control algorithm was optimized by using Matlab/Simulink software. After tuning the calculation optimization, when the proportional coefficient Kp was 0.41, the integral coefficient Ki was 0.76 and the differential coefficient Kd was 0.009, the PID controller met the requirements of the servo control system. The profiling mechanism sent the detected terrain height data to the STM32 control unit, and after analysis and processing, the hydraulic actuators were driven to control the lifting and lowering of the cutting table. After completing the design of the cutting table servocontrol system, it was installed on a 4GZQ130-A sugarcane harvester for functional tests. The driver started the machine, activated the cutting table servo control system for harvesting, set the cutting height, harvested three monopoles, observed the stubble height at each harvest, recorded the test data, measured the distance from the ground to the cut point of the cane and calculated the head breakage rate. The test results showed that after the 4GZQ130-A sugarcane harvester was fitted with a cutting table follower control system, the stubble height deviated from the preset stubble height within 20mm, an average head breakage rate was 21%. In comparison with the manually controlled harvesting trials, the average head breakage rate was reduced by 18.5 percenage points. The harvesters performance was further improved and the overall performance of the cutting table follower control system met the design and use requirements.

Abstract:Aiming at the problems of slow harvesting, an improved rapidly-exploring random trees with visual servoing (VS-IRRT) algorithm was proposed to solve the problems of slow path planning, high path cost and picking failure caused by visual positioning error and joint position error of manipulator in harvesting process. By using the sampling method based on super ellipsoid gravity bias and density reduction strategy, the purpose of tree expansion was increased, the sampling density of tree was reduced and the efficiency of path planning was improved. The greedy idea and B-spline curve were introduced to eliminate unnecessary nodes, and the remaining polyline were smoothed to optimize the implementation effect of the path on the manipulator. Combined with visual servoing control based on translation controller, the influence of positioning error on harvesting process was reduced. Matlab was used to simulate the improved RRT algorithm and the visual servo based on translation controller in two-dimensional and three-dimensional space. The results showed that the number of sampling points of the improved RRT algorithm was reduced by 92.9% compared with that of RRT*-connect algorithm, the planning time was reduced by 86.1% compared with that of RRT*-connect algorithm, and the path cost was reduced by 35.2% compared with that of RRT algorithm. Using six degrees of freedom manipulator for harvesting test, the harvesting speed of VS-IRRT algorithm was increased by 48.36% compared with that of RRT*-connect algorithm, the path cost was reduced by 17.14% compared with that of RRT algorithm, and the harvesting success rate was increased by 2.1 percentage points, therefore, in specific application scenarios, especially in agricultural harvesting scenarios, VS-IRRT algorithm can better improve the comprehensive performance of manipulator harvesting.

Abstract:In the study of production and processing technologies such as mechanical shelling, sowing and planting of Camellia oleifera seeds, due to the lack of accurate discrete element simulation models and parameters, the simulation and actual errors of design equipment are large. Reverse engineering techniques were used to establish a discrete element model of Camellia oleifera seeds in EDEM software. 〖JP2〗Through physical tests, the angle of repose (AOR) of Camellia oleifera seeds was measured to be (27.93±1.46)°. The parameter intervals of density, collision recovery coefficient and static friction coefficient between camellia seed and plate were measured. The discrete model parameters of Camellia oleifera seeds were filtered by using the Plackett-Burman Design to obtain the parameters that had a significant impact on the AOR. The path of steepest ascent method was carried out to determine the optimal value range of the parameters. The central composite design (CCD) response surface method (RSM) and machine learning were used to establish the regression models involving the AOR and the significant parameters. The results showed that the predictive ability and stability of BP artificial neural network based on genetic algorithm (GA) were better than that of random forest, support vector regression and BP artificial neural network. GA optimization was used to obtain the static friction coefficient between seeds, which was 0.443, the static friction coefficient between seeds and steel plates was 0.319, and the rolling friction coefficient between seeds was 0.063. The simulated AOR was measured to be 27.63°, and the relative error from the actual AOR was 1.09%. RSM optimization was used to obtain the static friction coefficient between seeds, which was 0.383, the static friction coefficient between seeds and steel plates was 0.335, and the rolling friction coefficient between seeds was 0.064. The simulated AOR was measured to be 26.99°, and the relative error from the actual AOR was 3.33%. The results showed that GA-BP-GA had better parameter optimization effect than RSM in the parameter calibration of Camellia oleifera seeds. Moreover, the built model and parameter calibration results of Camellia oleifera seeds can be used for discrete element simulation research.

Abstract:Axial flow pumps are widely used in agricultural area because of their high flow rate and low head. Accurate prediction of boundary layer transitions is important to improve the accuracy of internal flow calculations in axial pumps. The applicability of the SST γ-Reθt transition model at different Reynolds numbers was explored with a hydrofoil. It was found that the prediction accuracy of the SST γ-Reθt transition model was close to the experimental value under the low Reynolds number condition (ReL was less than 1.6×106); under the high Reynolds number condition, the boundary layer transition position predicted by the SST γ-Reθt transition model was gradually moved forward compared with the experimental value as the Reynolds number was increased. This indicated that the SST γ-Reθttransition model was not effective in determining the occurrence of boundary layer transitions in high Reynolds number hydrofoils. Based on this, the transport equation in the SST γ-Reθt transition model was modified by using the ambient source term method, introducing the parameters of environmental turbulent kinetic energy and environmental turbulent specific dissipation rate, and establishing the relationship between turbulent specific dissipation rate and Reynolds number to obtain the modified SST γ-Reθt transition model. The model was validated in the high Reynolds number flow of Donaldson trailing edge hydrofoil and NACA0016 hydrofoil. The prediction accuracy of typical flow characteristics such as wake vortex shedding frequency under the condition of high Reynolds number of Donaldson modified trailing edge hydrofoil was improved by about 8% compared with the original transition model. Compared with the original transition model, the prediction accuracy of the relative thickness of the boundary layer and the coefficient of friction in the transition region in the middle of the hydrofoil of NACA0016 was improved by more than three times.

Abstract:Current remote sensing technology can quickly and accurately obtain the spatial distribution information of crops. In order to explore the spatial distribution information of winter wheat in the Huang-Huai-Hai Plain in 2021, based on the Google Earth Engine (GEE) cloud platform. Sentinel-1 SAR radar image and Sentienl-2 optical remote sensing image were used as data sources, the spatial distribution information of winter wheat in the study area was extracted by computing polarization characteristics, spectral characteristics and texture characteristics, using four machine learning methods and deep learning network model. The classification accuracy of each classifier and network architecture was compared. The results showed that the total area of winter wheat in the Huang-Huai-Hai Plain was 16226667hm2, accounting for 49.17% of total area of the study area. The winter wheat planting area was the largest in Henan Province, accounting for 4647334hm2. The winter wheat planting distribution in the study area showed a decreasing trend from east to west and from south to north. Random forest was the classifier with the highest recognition accuracy among the four machine learning methods, with an overall classification accuracy of 94.30%. In the random forest algorithm, the overall accuracy of only using Sentinel-1 radar data was 87.38%, and the overall accuracy of only using Sentinel-2 optical data was 93.95%, while the overall accuracy of the fusion sequence Sentinel active and passive remote sensing data was 94.30%. In a wide range of winter wheat classification, the generalization of deep learning model was higher than that of machine learning.

Abstract:The construction of coastal aquaculture ponds has huge economic benefits and is of great significance to ensuring the supply of seafood and enriching the food diversity of residents. The rapid expansion of aquaculture ponds will also bring about a huge environmental crisis, so effectively revealing the spatial and temporal distribution characteristics of aquaculture ponds is crucial for orderly management of coastal aquaculture ponds. However, aquaculture ponds are mostly distributed on the side of tidal flats with tortuous coastlines and close to the ocean. It is challenging to identify aquaculture ponds effectively and with high precision. In response to this problem, an aquaculture ponds identification method that combined Google Earth Engine cloud platform and ArcGIS local classification post-processing was proposed. Based on water body frequency, object characteristics and fine processing, the spatial distribution of coastal aquaculture ponds in Zhejiang Province from 2016 to 2021 with high precision was obtained. The results showed that the overall accuracy of the aquaculture ponds was greater than 93%, and the Kappa coefficient was greater than 82%, indicating that the research method showed good applicability. The area of coastal aquaculture ponds in Zhejiang Province tended to decrease in 2016, 2019 and 2021, which was 30360.60hm2, 24375.35hm2 and 21700.02hm2, respectively. The prefecture-level cities of aquaculture ponds were concentrated in Ningbo, Taizhou, Shaoxing and Hangzhou, and the counties were concentrated in Cixi, Ninghai, Sanmen, Xiaoshan, Shangyu and Xiangshan. The agglomeration of aquaculture ponds in Zhejiang Province was decreased, and they were concentrated in bays, estuaries, coastal plains and tidal flats, such as Hangzhou Bay, Xiangshan Port, Sanmen Bay, Puba Port and Yueqing Bay. The spatial differences of aquaculture ponds were prominent, that in the sea side was larger than that in the land side, and that in the north was larger than that in the south. 

Abstract:The identification and location of wheat planting rows in the field environment is of great significance to the navigation operation of agricultural machinery such as pesticide spraying and weeding in the field. A method for detecting wheat planting row at multiple growth stages was proposed based on visible light remote sensing images of winter wheat at tillering stage and jointing stage obtained by unmanned aerial vehicle, combining with deep semantic segmentation and Hough transform linear detection. Firstly, wheat planting regions were extracted by SegNet deep semantic segmentation to overcome the sensitivity of traditional detection methods to light and improve detection accuracy. Secondly, based on the pre-detection results of wheat planting rows by Hough transform, dichotomy k-means clustering was proposed to further refine the detection results to identify the center line of wheat planting rows. Respectively, for winter wheat images at tillering stage and jointing stage, the mean absolute values of straight position deviation of planting row were 0.55cm and 0.11cm, and the mean absolute values of angle deviation were 0.0011 rad and 0.00037 rad. It was superior to the traditional method in detecting accuracy and line missing rate. The research results can provide a method for detecting the direction of crop planting in the navigation operation of intelligent agricultural machinery.

Abstract:Nitrogen (N) deficiency can directly reflect the degree of crop N nutrient deficiency, and it is important to obtain the information of rice N deficiency quickly and in a large area to achieve accurate fertilization of rice. Most of the existing studies focused on the use of UAV remote sensing to monitor rice N nutrition, and less research was conducted on N deficiency itself. Based on the canopy spectral data obtained by UAV hyperspectral remote sensing and rice agronomic data obtained by field sampling, the method of constructing the critical nitrogen concentration curve of northeastern rice was studied, and the nitrogen deficit of rice on this basis was determined; the spectrum in the state of nitrogen deficit approximately equal to 0 was used as the standard spectrum, and ratio, difference and normalized difference transformations on the spectral reflectance data were carried out respectively, and then the competitive adaptive re-weighting sampling method was used to the inversion models of rice nitrogen deficit based on the multivariable linear regression (MLR), extreme learning machine（ELM）and the bat algorithm optimized extreme learning machine（BA-ELM） were constructed by taking the extracted feature bands as input variables and the nitrogen deficit as output variables. The results showed that the equation coefficients a and b of the critical nitrogen concentration curve of northeastern rice were 2.026 and -0.4603, respectively, based on field data, which were consistent with previous studies; compared with other transformation methods, the normalized difference transformation and feature band extraction of the rice canopy spectrum significantly improved the correlation between the canopy spectral reflectance and rice nitrogen deficit, and also improved the inversion of the subsequent inversion model. The BA-ELM inversion model with normalized difference spectra as input predicted significantly better than the rest of the models, with the validation set R2 of 0.8306，RMSE of 0.8141kg/hm2, which had better estimation of N deficit.

Abstract:Aiming at the problems that it is difficult to identify the content of experimental image resources of the product data management (PDM) system of no-tillage seeding equipment in the process of storage and query, and it is difficult to guarantee the demand of users to obtain relevant resources, applying VB in VS (Microsoft Visual Studio) environment Net language with SQL Server database to develop an interactive resource management system, marking the content of experimental image resources with multiple information and assign weights, and applying ADO.Net (Microsoft ActiveX Data Objects. Net) technology to realize the editing and storage of multiple information of image resources, based on multiple information weights to create a recommendation query method, joint browsing and selection, and realize the acquisition and application of image resources. The test results showed that the system can add, delete, modify and query based on multiple information. When the input field did not exactly match the local database, recommendation data can be obtained, which realized the effective management of multiple information of image resources. Multivariate information can uniquely and accurately identify image resources and serve as the basis for resource management. The recommendation method based on multivariate information weight design can effectively solve the problem that user input fields did not exactly match local data tables.

Abstract:It is of great significance to realize the intelligent cognitive decision-making ability of robots in the agricultural field and help the further development of smart agriculture that researchers use human cognitive experience and objective knowledge to assist computers and robots in object cognition and behavioral decision-making under the small sample data situation. On the prerequisites of the ability to recognize and judge basic attribute information such as image color and image shape by using methods such as statistical counting and support vector machine(SVM), tools such as Protégé was firstly used to build a professional knowledge base for fruit recognition and classification based on human cognitive experience and objective knowledge in object recognition. Then, under the rules set by artificial experience, the color information and shape information obtained from the image were used as the input of the knowledge base, and the classification results of the items in the image were obtained through matching reasoning. The experiments selected and used 2091 images from the Fruit360 public data set for the first part experiment，which included multiple fruit images of grapes, bananas, and cherries. The research firstly selected 30 images of grapes, bananas and cherries as the training set and validation set for the computers image attribute ability learning, and then the image classification performance was tested on the data set of the first part experiment. The experimental results showed that the image classification accuracy of grapes and cherries was 100%, and that of bananas was 93.30%. Subsequently, totally 984 yellow peach images in the Fruit360 public data set were selected as the data set for the second part experiment. By only adding the knowledge of yellow peach to the professional knowledge base built with ontology technology, the classification accuracy of the images can reach 97.05%. All experimental results showed that the proposed method can effectively accomplish the task of image classification decision-making and the method had good process interpretability, ability sharing and scalability.

Abstract:Stomata are the important structure for plant leaves to exchange gas and water with environment. In order to solve the problem that traditional analysis methods of stomatal traits adopt manual observation and measurement, which causes tedious process, low efficiency and prone to human error, you only look once (YOLO) deep learning model was adopted to complete automatic identification and automatic measurement of stomata in maize (Zea mays L.) leaves. Combined with the characteristics of stomata data set, the YOLO deep learning model was improved to effectively improve the precision of stomata identification and measurement. The prediction end in YOLO deep learning model was optimized, which reduced the false detection rate. At the same time, the 16-fold and 32-fold down-sampling layers were simplified according to the characteristics of stomata, which improved the recognition efficiency. Experimental results showed that the identification precision of the improved YOLO deep learning model reached 95% on the maize leaves stomatal data set, and the average accuracy of parameter measurement was above 90%. The proposed method can automatically complete the identification, counting and measurement of stomata of maize, which solved the low efficiency of traditional stomatal analysis methods, and it can help agricultural scientists and botanists to conduct the analysis and research related to plant stomata.

Abstract:In order to achieve accurate, non-destructive and rapid classification of tea leaf species, the tea leaf species classification was realized through convolutional neural network by taking the images of tea leaves of six varieties under complex background as the research object. The classic lightweight convolutional neural network SqueezeNet was selected, and by adding batch normalization processing in the Fire module, the network parameters were not significantly increased to greatly improve the accuracy of the classification of multi-variety tea leaves. The 3×3 standard convolution kernel was replaced with a depthwise separable convolution, which further reduced the network model and reduced the networks requirements for hardware resources; by introducing an attention mechanism into each Fire module, the networks extraction of important feature information was enhanced. The test results showed that the original SqueezeNet model had an accuracy rate of 82.8% for the classification of multi-variety tea leaves, and the model after adding batch normalization had an accuracy rate of 86.0% in the test set, and the number of parameters was only 7.31×105, compared with the parameters before improvement. The amount of calculation was only increased by 0.8%, and the amount of calculation was basically the same as that before the improvement; the model after replacing the 3×3 standard convolution kernel in the Fire module with a depthwise separable convolution model had an accuracy rate of 86.8% in the test set, and the accuracy rate was increased by 0.8 percentage points, the amount of parameters were decreased to 2.46×105, the model parameters were decreased by 66.3%, and the amount of computation was decreased by 60.4%; the classification accuracy of the model test set after the introduction of the attention mechanism reached 90.5%, which was increased by 3.7 percentage points, while the amount of parameters was only increased by 1.23×105, and the amount of operations was only increased by 2×106. The improved model was further compared with the classic models AlexNet, ResNet18 and the lightweight networks MobilenetV3_Small and ShuffleNetv2. The results showed that the improved model had the best comprehensive performance in the classification of multi-variety tea leaves, and the three indicators of model scale, classification accuracy and classification speed were well balanced.

Abstract:Automatic and accurate estimation of disease severity is critical for disease management and yield loss prediction. Traditional disease severity estimation steps are complicated and inefficient, which makes it challenging to achieve accurate estimation in practical scenarios. A disease severity estimation method was proposed based on mixed dilated convolution and attention mechanism to improve UNet (MA-UNet). Firstly, to solve the problem of different sizes and irregular shapes of lesions, the mixed dilation convolution block (MDCB) was proposed to increase the receptive field and maintain the continuity of lesion information to improve the accuracy of lesion segmentation. Secondly, to overcome the influence of complex background, the attention mechanism (AM) was used to model the correlation between the spatial and channel dimensions. It can obtain the response within each pixel class and the dependency between channels to alleviate the backgrounds influence on network learning. Finally, the ratio of diseased lesion pixels to leaf pixels in the disease segmentation map was calculated to obtain the severity. It was validated based on cucumber downy mildew and powdery mildew images collected under field conditions and compared with fully convolutional network (FCN), SegNet, UNet, PSPNet, FPN, and DeepLabV3+. MA-UNet can meet the segmentation requirements of leaves and lesions in complex environments, with a mean intersection over union 84.97% and a value of frequency weighted intersection over union value of 93.95%. Moreover, it can accurately estimate the severity of cucumber leaf diseases, the correlation coefficient was 0.9654, and the RMSE was 1.0837%. The results showed that MA-UNet outperformed the comparison methods in refining lesion segmentation and accurately estimating disease severity. The research result can provide a reference for artificial intelligence to estimate and control disease severity in agriculture rapidly.

Abstract:In the process of pig body temperature detection based on thermal infrared video, the head posture of pigs in the nursery period changes greatly, and the ear base area was small, resulting in low positioning accuracy of the head and ear base area, which affected the accurate detection of pig ear base temperature. In view of the above problems, an improved YOLO v4 (Mish Dense YOLO v4, MD-YOLO v4) method for detecting the temperature of pig ears was proposed and a detection model for key parts of pigs was built. Firstly, in the CSPDarknet-53 backbone network, dense connection blocks were added to optimize feature transfer and reuse, and the spatial pyramid pooling (SPP) module was integrated into the backbone network to further increase the backbone network receptive field; secondly, an improved path aggregation network (PANet) was introduced in the neck to shorten the high and low fusion paths of the multi-scale feature pyramid graph; finally, the Mish activation function was used in the backbone and neck of the network to further improve the detection accuracy of the method. The test results showed that the mAP of the model for the detection of key parts of live pigs was 95.71%, which was 5.39 percentage points and 6.43 percentage points higher than that of YOLO v5 and YOLO v4, respectively, and the detection speed was 60.21f/s, which can meet the requirements of real-time detection. The average absolute errors of the left and right ear root temperature extraction of pigs in the thermal infrared video were 0.26℃ and 0.21℃, respectively, and the average relative errors were 0.68% and 0.55%, respectively. The results showed that the pig ear root temperature detection method based on the improved YOLO v4 proposed can be applied to the accurate positioning of the key parts of pigs in thermal infrared video, thereby realizing the accurate detection of pig ear root temperature.

Abstract:At present, the computer vision-based pig body measurement shows a high dependence on pig posture and low measurement efficiency. To solve these problems, a rapid and non-contact pig body size measurement method based on DeepLabCut was proposed. The top view RGB-D images of landrace pigs were captured by a RealSense L515 camera. The training effects of 10 backbone networks of ReNet, MobileNet-V2, and EfficientNet series were compared and analyzed, and then the EfficientNet-b6 model was selected as the optimal backbone network of DeepLabCut algorithm for feature point detection of pig body size. In order to achieve accurate calculation of pig body size data, SVM model was used to identify the standing stance of pigs and screen the natural standing stance of pigs. Based on this, the depthvalued proximity region replacement algorithm was used to optimize the outlier feature points and calculate the five body size indexes of pig body length, body width, body height, rump width and rump height by Euclidean distance. This method was tested on 140 groups of standing images of pigs, and it was found that the algorithm could achieve real-time and accurate measurement of body size in the natural standing posture of pigs, with maximum root mean square error of 1.79cm and computation time of 0.27s per frame.

Abstract:To solve the problem that the closed-set pig face recognition model cannot recognize pig individuals that have not appeared in the training set, an open-set pig face recognition method that integrated attention mechanism was proposed, which can realize open-set pig face image recognition and recognize pig individuals that the model had never seen. Firstly, a lightweight feature extraction module (GCDSC) was constructed based on a global attention mechanism, inverted residual structure, and depth separable convolution. Secondly, C3ECAGhost module was designed based on efficient attention mechanism, Ghost convolution, and residual network to extract high-level semantic features of pig face images. Finally, based on the MobileFaceNet network, incorporating GCDSC module, C3ECAGhost module, SphereFace loss function, and Euclidean distance measurement method, the model PigFaceNet was constructed to realize open-set pig face recognition. The experimental results showed that the GCDSC module can improve the accuracy of pig face recognition by 1.05 percentage points, and the C3ECAGhost module can further improve the accuracy of the model by 0.56 percentage points. The accuracy of the PigFaceNet model in open-set pig face recognition verification can reach 94.28%, which was 1.61 percentage points higher than that before modification. The model proposed was a lightweight model with 5.44MB parameters, which can improve the accuracy and provide a reference for intelligent breeding of pig farms.

Abstract:In order to realize the real time detection of Urechis unicinctus burrows in the actual aquaculture pond scene, and provide support for the automatic harvesting and yield prediction of Urechis unicinctus, a deep learning based identification method of Urechis unicinctus burrows was proposed. In view of the limited storage space of the embedded equipment of harvesting vessel and high real time requirements for target detection, the YOLO v4 model had a large number of parameters and a slow detection speed. By replacing the backbone network CSPDarkNet53 of YOLO v4 with a lightweight Mobilenet v2 to reduce the amount of network model parameters and improve the detection speed. On this basis, depthwise separable convolution blocks were used instead of the normal convolution blocks in the Neck and Detection Head parts of the original network to further reduce the number of model parameters. For the poor quality of underwater images, the multi-scale retinex with color restoration (MSRCR) algorithm was selected for image enhancement. Finally, for the original anchor box obtained by clustering the COCO dataset, which was not suitable for small target recognition, the K-means++ algorithm was used to recluster the dataset and optimize the linear scaling of the obtained new anchor box size to obtain the most suitable anchor box for the dataset in order to improve the target detection effect. To simulate the automatic capture scene of Urechis unicinctus, a set of image acquisition equipment with an unmanned ship as the main body was built, and an image data set was established through the collected video to conduct experiments. The trained model deployed on the embedded device Jetson AGX Xavier can detect mean average precision (mAP) of underwater Urechis unicinctus burrows up to 92.26% with detection speed of 36f/s and model size of only 22.2MB. Experiments showed that the method achieved a better balance of detection speed and accuracy and can meet the demand of practical application scenarios where the model was deployed in the embedded equipment of the Urechis unicinctus harvesting vessel. It provided a reference for the subsequent automatic harvesting of Urechis unicinctus and yield prediction in breeding ponds.

Abstract:In order to solve the problem that the traditional field obstacle recognition methods rely on manual feature extraction, long calculation time, and it's difficult to achieve real-time recognition in unstructured field environment, an optimized unstructured field obstacle instance segmentation method based on Mask R-CNN model was proposed. Firstly, an unstructured field obstacle dataset was constructed by aerial photography and network search. And then based on the ResNet-50 residual network, the spatial attention was introduced to focus on the significant apparent features of the tracking target, and the influence of useless features such as noise was suppressed. In addition, the deformable convolution was introduced into the structure of the ResNet-50 to add the offset, increase the receptive field and improve the robustness of the model. Comparative analysis was made by adding spatial attention and deformable convolution to different stages in the structure of ResNet-50. The results showed that compared with the original Mask R-CNN model, the mAP values of Bbox and Mask in Mask R-CNN improved by adding spatial attention and deformable convolution in Stage 2, Stage 3 and Stage 5 of the ResNet-50 were increased from 64.5% and 56.9% to 71.3% and 62.3%, respectively. The improved Mask R-CNN can well realize field obstacle detection and provide technical support for plant protection UAV to work safely and efficiently in unstructured field environment.

Abstract:Aiming at the problem of camera shake and relative motion of objects leading to blurred detection images during target detection in orchards, a D2-YOLO one-stage deblurring recognition deep network that combined the DeblurGAN-v2 deblurring network and the YOLOv5s target detection network was proposed. It was used to detect and identify obstacles in orchard blurred scene images. To reduce the number of parameters of the fusion model and improve the detection speed, firstly the standard convolution used in the YOLOv5s backbone network with a deep separable convolution was replaced, then CIoU_Loss was used as the bounding box regression loss function of prediction. The fusion network used the improved CSPDarknet as the backbone for feature extraction. After recovering the original natural information of the blurred image, it combined multi-scale features for model prediction. To verify the effectiveness of the proposed method, seven common obstacles in the real orchard settings were selected as the target detection objects, based on the chassis of the crawler mobile robot, the BUNKER was equipped with portable computers, cameras and other equipment to form a mobile platform for image acquisition, and the model training and testing were carried out on the Pytorch deep learning framework. The precision and recall rates of the proposed D2-YOLO deblurring detection network were 91.33% and 89.12%, respectively, which were 1.36 percentage points and 2.7 percentage points higher than that of the step-by-step training DeblurGAN-v2+YOLOv5s. Compared with YOLOv5s, there was an increase of 9.54 percentage points and 9.99 percentage points in precision and recall rates, which can meet the accuracy and realtime requirements of orchard robot obstacle deblurring recognition. The research result can provide a reference for obstacle detection of agricultural robots in orchard in the later stage.

Abstract:To solve the problems of high labor intensity and long working time of artificial overthrowing of feed in the pasture, an autonomous navigation system based on LiDAR for synchronous positioning and map building was designed to realize robot navigation and grass turning in pasture environment. The autonomous navigation system platform perceived the pasture environment through LiDAR, a ranch environment map was constructed by using Cartographer algorithm loaded with odometer information, the AMCL algorithm was used which did not load the odometer information to achieve robot positioning, and Dijkstra algorithm was used to plan the robot to overthrow the grass work path. The experiment showed that when constructing the ranch environment map, the maximum deviation of the robot loading odometer information was lower than that of the map when the odometer information was loaded, which was 0.02m and 0.14m, respectively, and the maximum value of the horizontal and vertical deviation and the maximum heading declination angle were less than 0.04m, 0.10m and 11° when the positioning and navigation of the robot were realized, and the navigation accuracy was higher than the value when loading the odometer information. All the results showed that the navigation accuracy can meet the requirements of overthrowing grass operations in a pasture environment.

Abstract:The development of wetting zone in soil under film hole irrigation is a crucial parameter for designing an effective irrigation system. HYDRUS-2D/3D numerical simulation was used to study the effects of soil bulk density and film hole diameter on the soil infiltration characteristics under 12 soil textures.Based on 180 simulated results, the parameters of the wetting front movement model were improved. The accuracy of the model was validated against experimental data published in the literature and measured by authors. The results showed that the migration process of soil wetting front was increased with the increase of film hole diameter and irrigation time, however it was decreased with the increase of soil bulk density. The migration distance of the wetting front had a power function relation with the steady infiltration rate and time,whereby the model of wetted volume was established for different soil texture types. For a given soil texture, the steady infiltration rate had a good power function relationship with soil bulk weight and film hole diameter with power function exponents of -6.3 and 1.1, and the power function coefficients was deduced from only one set of cylinder infiltrometer field tests. The model simulation values were in good agreement with the measured values from 12 groups experiments, the RMSE was between 0.020cm and 0.170cm, and the NSE was between 0.995 and 0.999, which realized the practical application of the empirical model for predicting the wetting pattern dimensions under film hole irrigation in the field with simple experimental design, easy operation and quick field evaluation.

Abstract:Gansu Province is located in an ecologically fragile area, with complex climate conditions, high probability and wide range of drought. In order to better study the temporal and spatial variation characteristics of drought in Gansu Province, it was divided into four climate zones according to the climate type and geographical characteristics: the continental climate area of Hexi (Region Ⅰ), the monsoon climate area in the northern of Longzhong (Region Ⅱ), the monsoon climate area of Longnan and the south of Longzhong (Region Ⅲ) and the alpine climate area of Gannan (Region Ⅳ). The meteorological data of 26 national meteorological stations in Gansu Province were used to calculate the standardized precipitation evapotranspiration index on monthly, quarterly and annual scales（SPEI-1, SPEI-3 and SPEI-12）in the past 60 years (1960—2019). The spatiotemporal evolution characteristics of drought in Gansu Province in recent 60 years were discussed by means of climate tendency rate, Mann-Kendall〖JP〗 mutation test and spatial interpolation. The results showed that the SPEI values of different time scales showed a decreasing trend from the perspective of time process, and the fluctuation range of SPEI values was smaller as the time scale was increased. For seasonal factor, in spring, summer and autumn, the SPEI values showed a downward trend in fluctuations in all climate regions of Gansu Province. And the downward trend was obvious, indicating a significant drying trend. In winter, the SPEI values showed a rising trend in fluctuations in various climate regions, indicating a trend of wetting. Spatially, it showed a humid trend in Region Ⅰ of Gansu Province, but it was dire in Region Ⅱ, Ⅲ and Ⅳ of Gansu Province. And in spring, the drought in various climate regions of Gansu Province had an obvious trend of aggravation, followed by summer and autumn, while in winter, the drought was basically slowed down. The frequency of occurrence of different levels of drought in different climate regions of Gansu Province was very different and uneven, and the drought frequency from smallest to largest was extreme drought, moderate drought, heavy drought and mild drought.

Abstract:Soil microbial respiration and its entropy (soil metabolic entropy and microbial entropy) are important parameters indicating soil carbon metabolic activity and sensitivity indicators of soil quality, which can reveal the impact of environmental or biological factor changes on soil earlier. Ridge-furrow mulching system is an efficient water-saving cultivation mode widely used in dry farmland in China. In order to prove the influence of limited supplementary irrigation on soil quality under the condition of wheat field, three rainfall conditions (high flow year was 275mm, normal flow year was 200mm, and dry year was 125mm) and four irrigation treatments (150mm, 75mm, 37.5mm, and 0mm) under the ridge-furrow mulching system (RF) were set up during the growth period. Traditional flat planting (TF) was used as the control. The soil microbial respiration, microbial biomass carbon, and its entropy (soil metabolic entropy and microbial entropy) were determined in different soil layers under RF and TF. The results obtained after three years (October 2017 to June 2020) showed that RF had more significant effects on the soil microbial respiration and microbial entropy in the upper soil layer compared with those in the deep soil. Under the same amount of rainfall and supplementary irrigation during the growth period of winter wheat, soil microbial respiration levels under RF in the upper and deep soil layers were 2.47%~21.67% and 3.28%~24.59% higher compared with TF, respectively, and the difference was significant in the years with low flow year (125mm). The microbial entropy was increased by 9.09%~27.05% and 11.9%~24.76% in the upper and deep soil layers, respectively, under RF. These results provided a scientific basis for predicting the soil quality and planning irrigation management for fields under RF by clarifying its effects on the sustainable development of land.

Abstract:Drip irrigation with film mulching has been widely used in crop cultivation. Film mulching can change surface optical properties, including reflectance, absorption, and emissivity, which will affect the energy distribution in the farmland. It is necessary to study the water and heat transport and its influence factors in drip irrigated field with film mulching. The quantification of water and heat transfer process of the farmland is of great significance for agricultural water management and irrigation schedule formulation. Based on the measured data with Bowen ratio system and meteorological station, the water and heat flux variation and its response to environmental factors were studied in drip irrigated purple potato field with film mulching. The results indicated that latent heat flux was the main part of energy expenditure in drip irrigated purple potato field with film mulching during the whole growth period. Sensible heat flux and soil heat flux accounted for a relatively small proportion. During the whole growth period, the proportion of latent heat flux, sensible heat flux and soil heat flux was 69.12%, 25.14% and 6.57%, respectively. Under different weather conditions, the magnitude and range of sensible heat flux were smaller than that of latent heat flux. Latent heat flux had the most significant response to rainfall and irrigation. The influence of rainfall on latent heat flux was greater than that of irrigation. Net radiation and air temperature had great impact on latent heat flux, while the effects of surface soil temperature and wind speed had low impact. Various environmental factors affected the latent heat flux directly or indirectly. The research results can deepen the understanding of water and heat transfer in drip irrigated purple potato farmland under film mulching, and provide theoretical basis for efficient water use of crops.

Abstract:In order to clarify the significance of suitable fertilization straw measures for summer maize farmland in Guanzhong region in the future climate conditions to control ammonia and stabilize yield and cope with climate change, based on the two-year field experiment conducted in 2019—2020, the impact of different nitrogen fertilizer types and different straw returning modes on soil ammonia volatilization and crop yield in farmland was studied. The DNDC model was calibrated and validated according to the field measured data, and the validated model was used to simulate the effects of different fertilization-straw measures on summer maize yield and soil ammonia volatilization accumulation under future climatic conditions. Taking into account the yield and the cumulative amount of soil ammonia volatilization of maize per production unit, the optimal ammonia control and stable yield fertilization-straw measures for summer maize farmland in Guanzhong area under future climatic conditions were finally put forward. The results showed that the corrected DNDC model could well simulate summer maize growth and soil ammonia volatilization accumulation under different fertilization-straw measures. Under future climatic conditions, straw returning would significantly increase summer maize yield and reduce soil ammonia volatilization accumulation per unit yield of maize. Under the RCP4.5 emission scenario, in the future from 2030 to 2090, the soil ammonia volatilization accumulation amount per unit yield of maize was lower and the yield was higher when the full amount of straw was returned to the field and 180kg/hm2stable nitrogen fertilizer was applied to the field. Under the RCP8.5 emission scenario, in the future of 2030—2050 and 2070—2090, the full amount of straw was returned to the field with 180kg/hm2 of stable nitrogen fertilizer and the full amount of straw was returned to the field with 162kg/hm2 of stable nitrogen fertilizer production unit, the soil ammonia volatilization accumulation in yield maize was lower and yield was higher. Therefore, under the RCP4.5 emission scenario, the full amount of straw returned to the field and the application of 180kg/hm2 stable nitrogen fertilizer was more optimal fertilization-straw measure for controlling ammonia production and stabilizing production in Guanzhong area from 2030 to 2090. Under the RCP8.5 emission scenario, the combination of 180kg/hm2 stable nitrogen fertilizer and 162kg/hm2 stable nitrogen fertilizer combined with the full amount of straw returning to the field were the optimal fertilization-straw measures for 2030—2050 and 2070—2090 in Guanzhong area, respectively. The results can provide a reference for the realization of sustainable agricultural development and stable yield and emission reduction in Guanzhong area.

Abstract:The accurate prediction of greenhouse environment variation based on the constructed prediction model is helpful to precisely regulate the crop environment, and promote the growth of fruits and vegetables. Due to the coexistence of multiple parameters, complex coupling with each other, temporality and nonlinearity of greenhouse microclimate environment, the accurate prediction model is difficult to establish. Based on above issues, a greenhouse environment prediction model was proposed based on the sparrow search algorithm (SSA) optimized-long short term memory (LSTM) neural network method, so as to realize the prediction of greenhouse environment data sequence with the Internet of things (IoT) collecting accurate multipoint environment data. The experimental results showed that the automatic parametric optimization process by SSA could deal with the time consuming problem of manual parameter selection for the LSTM model. The proposed SSA-LSTM method could lower the model training time, and the optimal parameters selection could make sure the model worked with the optimum capability. The trained SSA-LSTM model was used to predict six kinds of greenhouse environment data, including the air temperature, air humidity, soil temperature, soil humidity, CO2 concentration, and the illumination intensity. The proposed SSA-LSTM could realize a 97.6% average prediction fit index, compared with the back-propagation network, the gated recurrent unit neural network and the LSTM, the prediction fit index was elevated by 8.1 percentage points, 4.1 percentage points and 4.3 percentage points. Therefore, the prediction accuracy of SSA-LSTM was obviously improved. The research result could provide reference for the development of greenhouse environment control strategy and deal with the lag problem of environment control.

Abstract:Compost fermentation is a very important technical means of transforming agricultural waste into substrate and fertilizer. In the process of agricultural waste compost, temperature affects the fermentation rate and quality of the compost. Therefore, it is very important to study the temperature distribution inside the compost for the accurate control of compost. However, for fermentation piles with different stacking methods, their volume and height are different, which makes it difficult for the traditional probe sensor whose length of probe and the position of monitoring point are fixed to measure flexibly and conveniently which brings great inconvenience to the accurate control of the fermentation process of agricultural waste. In order to solve the above problems, a combined multi-storey temperature monitoring system was designed which can run in the state of low power consumption. The detection rod of the system was in the form of modularization, and the modules can be spliced freely according to the monitoring requirements. Adaptive identification of monitoring module based on CAN bus, configuration information synchronization and adaptive matching based on JSON text and relational database, as well as human-machine interface adaptive generation of system hardware, server, and WeChat applet were realized. After function test, operation power consumption test and longterm operation test on site, the results showed that the monitoring system can realize real-time data monitoring and low-power operation in the continuous monitoring process, maintain good stability in the whole operation process, and realize the synchronous configuration and information display of system hardware, server side and WeChat applet side, which can meet the needs of multi-storey long-term temperature monitoring in the composting and fermentation process of agricultural waste.

Abstract:The soybean milk was obtained by coldpressing from commercially available soybeans, the obtained milk was then subjected to high-pressure homogenization pretreatment assisted by different proteases (alkaline protease, papain, bromelain) to hydrolyze soybean milk and assess the effect on its structure and quality. The results regarding physical properties revealed that with the increase of homogenization pressure the solubility and hydrolysis degree of soybean milk reached 91.9% and 8.24%, respectively. Meanwhile, there was a substantial improvement in stability, negative surface charge and uniform particle size distribution. SDS-PAGE electrophoresis, infrared spectroscopy and fluorescence spectroscopy confirmed the changes in protein structure of soybean milk under different homogeneous pressures assisted enzymatic hydrolysis, showing a decreased band depth anti-nutritional protein factors. There was reduction in secondary structure and the contents of α-helix, β-rotation. The anti-nutritional factors revealed the relationship between protein structure and function. When the homogenization pressure was set at 100MPa, the inhibitory rates of the three enzymes on soybean globulin, β-concomitant soybean globulin, phytic acid, soybean lectin, trypsin inhibitor and lipoxygenase were up to 51.28%, 57.83%, 72.31%, 71.4%, 89.55% and 82.96%, respectively. The results showed that high pressure homogeneous pretreatment effectively improved the hydrolysis of protease anti-nutritional factor components. The obtained results might be fruitful for the preparation of nutritious and healthy soybean milk.

Abstract:With the aim to solve the problem of manual sampling and sensory identification of feed raw material entering the silo in the feed production process, and realize automatic identification of raw material type, taking bulk feed raw material such as corn, bran, wheat, soybean meal and fish meal as the research object, a multi-channel automatic identification device for feed raw material type was designed and built independently, feed raw material image dataset was collected, and data augmentation methods were used to increase sample diversity. Based on ResNet18 convolution neural network, CAM-ResNet18 network model for feed raw material type identification was constructed by adding the channel attention mechanism, adding the Dropout method, adopting the Adam optimizer and embedding the cosine annealing method，while the migration learning was introduced to train the model. The average accuracy of the CAM-ResNet18 network model for feed raw material type reached 99.1% in the validation set, with a recognition time of 2.58ms. Compared with the ResNet18, ResNet34, AlexNet and VGG16 network models, the validation accuracy was improved by 0.6, 0.2, 3.7 and 1.1 percentage points, respectively. For the result analysis of confusion matrix, the average accuracy of test set recognition was 99.4%, which had high accuracy and recall. The results showed that CAM-ResNet18 network model had higher accuracy rate and faster detection speed in the identification of feed raw material, providing a theoretical method and technical support for the identification of feed raw material entering the silo in the actual production.

Abstract:The physiological characteristics of Fuji apples change during the post-ripening process of storage. If the storage time is too short, the best edible quality cannot be achieved. Excessive storage will seriously reduce the quality, then affects the quality of out-of-warehouse and the selling price. In order to make the fruits during the storage period with better quality for sale, the study on the quality prediction model of apple during storage was carried out, and on this basis, the out-of-warehouse quality of apple was evaluated and predicted. The near-infrared spectrum and quality indexes (soluble solid content (SSC), hardness and weight loss rate) of apple at different times during the whole storage period were collected. The variation rule of fruit diffuse reflectance spectrum and quality index during storage was analyzed. Partial least squares (PLS) and nonlinear autoregressive with external input (NARX) prediction model for apple quality during storage was established based on the diffuse reflectance spectrum in the wavelength range of 1000~2400nm, combined with pretreatment and feature wavelength extraction. According to apple industry standards, the judgment basis of apple out-of-warehouse quality was determined, and the TOPSIS method based on entropy weight was used to comprehensively evaluate the fruit out-of-warehouse quality, and realize the prediction of the quality score by PLS and the prediction of multiple quality indexes by NARX. The results showed that when predicting SSC, hardness and weight loss rate, the optimal models were CARS-SPA-PLS, CARS-NARX and SPA-NARX, respectively, the correlation coefficients were 0.914, 0.796 and 0.918, and the root mean square errors were 0.511°Brix, 0.475kg/cm and 0.682%. When predicting quality scores, the correlation coefficient and root mean square error of the PLS model were 0.896 and 0.0434, respectively, the correlation coefficient of the NARX multi-output model were 0.794, 0.785 and 0.905, and the root mean square errors were 0.308°Brix, 0.492kg/cm2 and 0.714%. The application of near-infrared spectroscopy technology can realize the detection of fruit storage quality and the screening of quality of out-of-warehouse, and the research result can provide a method for efficient storage management technology.

Abstract:Calibration transfer can solve the problem that multivariate calibration models cannot be shared among different near-infrared spectrometers. Taking edible oil as the research object, transfer analysis of its acid value and peroxide value model was conducted. The partial least squares multivariate correction model was established on the master spectrometers, and the calibration transfer was realized by using the parameter-free and efficient calibration enhancement (PFCE) calibration transfer algorithm in NS-PFCE without standard sample transfer and FS-PFCE with standard sample transfer, and the dependence of calibration transfer on the number of standardization samples was explored. In addition, it was compared with three calibration transfer algorithms with standard sample, which were slope/bias (S/B), direct standardization (DS) and piecewise direct standardization (PDS), and two calibration transfer algorithms without standard sample, which were finite impulse response (FIR) and stability competitive adaptive reweighted sampling (SCARS). The results suggested that after the NS-PFCE without standard sample algorithm was transferred, the root mean square error of prediction (RMSEP) of the acid value and peroxide value was decreased from 0.613mg/g and 16.153mmol/kg to 0.275mg/g and 9.523mmol/kg,respectively. Furthermore, after the FS-PFCE with standard sample algorithm was transferred, the root mean square error of prediction (RMSEP) of the acid value and peroxide value was dropped to 0.274mg/g and 8.945mmol/kg, respectively. Specifically, the increase of the number of standardized samples, the root mean square error of prediction (RMSEP) was lower. The parameter-free and efficient calibration enhancement (PFCE) algorithm combined a single transfer method without a standard sample and a standard sample, which enhanced the adaptability and inclusiveness of the transfer model. And PFCE algorithm effectively reduced the difference between the master spectrum and the slave spectrum, and also realized the calibration model sharing between different spectrometers.

Abstract:With the improvement of consumption level, the internal quality of fruit has become an important factor to attract consumers. However, the traditional methods used to measure soluble solids content (SSC) and firmness are destructive. To realize rapid non-destructive detection on the internal quality of multi-fruits, a handheld non-destructive detector for internal quality of multi-fruits with replaceable probe was developed. The hardware system of the detector consisted of a host and a multi-spectral acquisition probe. The host included a microprocessor, a power management module, a voltage regulator drive module and an input and output module. The multi-spectral acquisition probe included 12 light emitting diodes (LEDs) at different wavelengths and a digital optoelectronic sensor. The software system of the detector was developed in C language in the development environment at MDK 5.0. The diffuse reflectance multi-spectral of “Huayou” kiwifruit and “Xue” pear were collected by the developed detector, and the prediction models for SSC and firmness were established based on the partial-least-squares regression. After downloading the model into the detector, the detection performance of the detector was tested. The results showed that the root mean square errors of SSC and firmness prediction for kiwifruit were 1.51% and 5.13N, and the root mean square errors of SSC and firmness prediction for pear were 0.52% and 4.57N. Moreover, the measurement could be realized in 2s. The measurement on the internal quality of multifruits was realized by replacing the probe of the detector.

Abstract:To eliminate the deviation between the actual pressure and the expected pressure of the shifting clutch of hydraulic mechanical continuously variable transmission (HMCVT) during the pressure tracking control process, a global terminal sliding mode control algorithm based on extended observer was proposed to achieve the high-precision tracking control of the pressure. By establishing the nonlinear mathematical model of wet clutch with uncertain disturbance, the state space equation of pressure control system was derived. The mismatched disturbance was estimated by extended observer, the linear term was introduced to accelerate the global convergence of terminal sliding mode control. Then a global terminal sliding mode pressure tracking controller based on extended observer was designed for HMCVT shifting clutch pressure system in real time. Finally, the effect of the controller was simulated and verified by bench test. The simulation and test results showed that the uncertain disturbance can be accurately observed by the extended observer. Compared with the traditional TSMC, the dynamic response time was only 0.13s, the overshoot was only 0.08MPa, and no chattering phenomenon occurred for the proposed algorithm. In addition, the control algorithm had good performance anti-interference capability, which was reflected by the smallest jerk (reduced by 12.7% at most) and sliding friction work (reduced by 10.2% at most). The results proved that pressure tracking control algorithm proposed had good robustness, and it can be applied to pressure tracking control of HMCVT shifting clutch.

Abstract:According to the theory and method of topological structure design of parallel mechanism(PM) based on position and orientation characteristic (POC) equation, a 3-DOF asymmetric two-translation and one-rotation (2T1R) PM with zero coupling-degree and partial motion decoupling was firstly designed and analyzed, including the topological design process of the PM, and the main topological characteristics such as the degree-of-freedom and coupling degree κ. Secondly, according to the kinematic modeling method based on topological characteristics, the forward and inverse solutions of symbolic positions were found, from which the working space calculation of the PM was carried out based on the forward solution. At the same time, the position inverse equation was derived to obtain the Jacobian matrix of the PM, from which the velocity, acceleration and singularity of the PM were derived. Thirdly, the sequential single-chain method based on the principle of virtual work was used to carry out reverse dynamic modeling, and the driving force change curve of the actuated pair of the PM was obtained, the supporting reaction force of the kinematic pair at the sub-kinematic chain (SKC) connection were also obtained, which were then verified by ADAMS dynamic simulation. Finally, the potential application scenarios of this mechanism used as automatic material transfer and unloading device between conveyor belts with different heights were conceptually designed and analyzed. The research result can provide a theoretical basis for the efficient kinematics, dynamic modeling and analysis, performance optimization and prototype development of the twobranch parallel mechanism with larger rotation space and partial motion decoupling.

Abstract:Parallel mechanism has the advantages of high bearing capacity, high precision and high stiffness, and is widely used in all walks of life in the industrial field. In order to reduce the difficulty of measuring and compensating the mechanical errors of parallel mechanisms and realize the accurate control of the end of the mechanism, a method of end error compensation was proposed based on the combination of Jacobian and RBF neural networks. Taking a 3-PTT parallel mechanism as the research object, the forward and inverse kinematics of the mechanism were analyzed using geometric method, and the correctness of the mathematical model was verified by Matlab/GUI. Jacobian was solved according to kinematics model, and constraint singularity and motion singularity of mechanism are analyzed. In order to verify the effectiveness of the mechanism end error compensation method, two experimental conditions were set up, namely, whether there was a return error compensation of the lead screw and whether the end was subjected to different loads, and the end position was measured by the laser tracker. By training the compensation model through the collected data, the error compensation is completed. The experimental results show that the axial (x axis) and radial (y axis) position errors of the end of the mechanism are reduced by more than 90%, and the vertical (z axis) position errors are reduced by more than 80% after using the error compensation method. In this paper, the error compensation effect is good, the precision of the end of the mechanism is obviously improved, and the proposed method is effective.

Abstract:A simple and effective control method for robot trajectory tracking with uncertain dynamics was presented. The core idea of the proposed method was to modify the reference trajectory in real-time. The main operation of this method was to accumulate the generated tracking errors and compensate them feedforward in real-time to the points to be tracked on the reference trajectory. Firstly, the control block diagram for the proposed method was showed. Then, the equation for the relationship between the tracking error and the command error was derived from the control block diagram. The equation showed that the control algorithm in the controller only needed to ensure that the velocity error was stable and the tracking error would converge. The increase in compensation gain can also accelerate the convergence of the error. Subsequently, the convergence condition that PD control law can satisfy the proposed method was analyzed. At the same time, the adjustment scheme of parameters in the proposed method was given. Finally, the effectiveness of the proposed method was verified by simulation and physical experiment. In the physical experiment, the absolute value of error obtained from tracking trajectory 1 of each joint was no more than 0.0087rad; the absolute value of error obtained from tracking trajectory 2 of each joint was no more than 0.0059rad.

Abstract:为使伺服压力机具有良好的拉深性能，需要对其轨迹进行合理规划。针对六连杆大型伺服压力机，构建一种主传动机构轨迹规划方法。基于库伦-粘性摩擦模型，建立主传动机构的高精度动力学模型。考虑冲压过程中的工艺约束和要求，提出基于复合三角函数的伺服电机加减速控制的改进模型。针对主传动机构的能耗和效率进行分析，以滑块运动的周期能耗和生产节拍为优化指标，通过线性加权的方式构造多目标优化函数，引入机械手送料时间、滑块最大速度、主动件角速度、伺服电机动态限和热极限等约束，采用遗传算法完成了多目标优化设计。结果显示：优化后，伺服压力机主传动机构在一个周期的能效提升4.54%，生产节拍时间减小3.23%，实现了良好的拉深工艺模式。