岳学军,凌康杰,王林惠,岑振钊,卢杨,刘永鑫.基于高光谱和深度迁移学习的柑橘叶片钾含量反演[J].农业机械学报,2019,50(3):186-195.
YUE Xuejun,LING Kangjie,WANG Linhui,CEN Zhenzhao,LU Yang,LIU Yongxin.Inversion of Potassium Content for Citrus Leaves Based on Hyperspectral and Deep Transfer Learning[J].Transactions of the Chinese Society for Agricultural Machinery,2019,50(3):186-195.
摘要点击次数: 1778
全文下载次数: 850
基于高光谱和深度迁移学习的柑橘叶片钾含量反演   [下载全文]
Inversion of Potassium Content for Citrus Leaves Based on Hyperspectral and Deep Transfer Learning   [Download Pdf][in English]
投稿时间:2018-09-17  
DOI:10.6041/j.issn.1000-1298.2019.03.020
中文关键词:  柑橘叶片; 钾含量  深度迁移学习  堆栈稀疏自动编码机  高光谱  支持向量回归
基金项目:国家自然科学基金项目(30871450)、广东省科技计划项目(2015A020224036、2014A020208109、2016A020210081)、广东省水利科技创新项目(2016-18)和广州市科技计划项目(201803020022)
作者单位
岳学军 华南农业大学 
凌康杰 华南农业大学 
王林惠 华南农业大学 
岑振钊 华南农业大学 
卢杨 华南农业大学 
刘永鑫 华南农业大学 
中文摘要:针对传统柑橘叶片钾含量检测方法耗时费力、操作繁琐且损伤叶片等弊端,引入高光谱信息探索柑橘叶片钾含量快速无损检测与预测模型,选用ASD FieldSpec 3光谱仪采集柑橘4个重要物候期(萌芽期、稳果期、壮果促梢期和采果期)的叶片反射光谱,同步采用火焰光度法测定叶片的钾含量;先用正交试验确定小波去噪的最佳去噪参数组合,再进行不同光谱形式变换,对不同物候期光谱进行基于堆栈稀疏编码机-深度学习网络(Stacked sparse autoencoder-deep learning networks,SSAE-DLNs)的特征提取迁移和融合多种特征,对比支持向量机回归、偏最小二乘法回归、广义神经网络、逐步多元线性回归等多种诊断模型,结果表明,模型SSAE-DLNs基于一阶微分光谱特征建立全生长期钾含量预测模型的性能最优,其校正集和验证集决定系数分别为0.8988、0.8771,均方根误差分别为0.5443、0.5528。试验表明,深度迁移学习网络可对柑橘叶片钾含量进行精确预测,为高光谱检测技术用于柑橘树长势监测和营养诊断提供了参考。
YUE Xuejun  LING Kangjie  WANG Linhui  CEN Zhenzhao  LU Yang  LIU Yongxin
South China Agricultural University,South China Agricultural University,South China Agricultural University,South China Agricultural University,South China Agricultural University and South China Agricultural University
Key Words:citrus leaves  potassium content  deep transfer learning  stacked sparse autoencoder  hyperspectral  support vector regression
Abstract:Traditional methods of obtaining potassium content of citrus leaves are time consuming procedures with complex operations which can be harmful to citrus trees. Moreover, traditional methods cannot meet the demand for rapid and non destructive monitoring of potassium content in large scale citrus orchards. Combined with the state of the art deep learning technology, a model based on stacked sparse autoencoder (SSAE) and deep learning networks (DLNs) using hyperspectral information for potassium content prediction in four growth stages was proposed. The experiments were conducted in the Crab Village of Luogang District, Guangzhou City, Guangdong Province, and the samples were 109 citrus trees planted. During four growth stages, i.e., germination, stability, bloom and picking stages, hyperspectral reflectance of citrus leaves was respectively measured by spectrometer (ASD FieldSpec 3), and at the same time, potassium content of citrus leaves was obtained by using traditional chemical method. All the collected samples constituted a large scale dataset with totally 436 tuples, 80% of which were utilized as the calibration set and remaining 20% as the validation set. The constructed model which relied on the calibration set and the validation set was evaluated respectively. Firstly, successive projection algorithm (SPA) was provided to deal with the high dimensional spectral vectors for dimension reduction and feature extraction. A prediction model of multiple linear regression (MLR) for potassium content of citrus leaves was established based on those extracted features. The result showed that the potassium spectrum contained a large number of complex nonlinear characteristics. Secondly, wavelet de noising was applied to reduce the high frequency noise in the original spectrum, and the optimized parameter combination of wavelet de noising through orthogonal test was as follows: “coif2” as wavelet basis function, the number of decomposition layer was 3, “sqtwolog” as the threshold, and “one” as noise estimation scheme, respectively. Thirdly, the features of SSAE in a specific stage were transferred and merged into baseline layer by layer to find out the best number of layers. The result showed that the best numbers of transferred layer were 3, 1, 4 and 3, and the corresponding values of determination coefficients for calibration set were 0.8999, 0.8598, 0.8869 and 0.8547 at germination, stability, bloom and picking stages, respectively, which were improved by 19.82%, 9.45%, 21.49% and 7.21%, respectively, compared with baseline. Then, the features of SSAE in the best layer were transferred and merged into baseline stage by stage to find out the best number of transferred stage. The experiment revealed that features of all four stages were transferred to its corresponding stage domain achieving the best performance. In this situation, the coefficients of determination for calibration set were 0.877 2, 0.8981, 0.9049 and 0.8894 at germination, stability, bloom and picking stages, respectively, which were improved by 16.80%, 14.32%, 23.96% and 11.56%, respectively, compared with baseline. Fourthly, after performing wavelet de noising and four kinds of spectrum transformation, i.e., the first derivative, second derivative, reciprocal and logarithm to the original spectrum, the layers’ features and stages’ features, which were obtained in SSAE previously, were transferred and merged into spectrum in four growth stages. When the first derivative spectrum was used as the input vector of the samples with wavelet de noising, the SSAE-DLNs model achieved the best result and the coefficients of determination for calibration set were 0.8992, 0.8899, 0.8838, 0.8727 and 0.8988, respectively, and the corresponding values of RMSE were 0.5425, 0.5496, 0.5509, 0.5539 and 0.5443, respectively, and the corresponding values of sparse proportion were 0.1411, 0.1633, 0.1189, 0.1856 and 0.2078, respectively, at germination, stability, bloom, picking stages and the whole growth period; and for the validation set, the coefficients of determination were 0.8651, 0.8704, 0.8551, 0.8580 and 0.8771, respectively, and the corresponding values of RMSE were 0.5693, 0.5674, 0.5786, 0.5722 and 0.5528, respectively. Comparing with traditional models such as support vector regression (SVR), partial least square regression (PLSR), general regression neural networks (GRNN) and stepwise multiple linear regression (SMLR), SSAE-DLNs model achieved the best performance, and the next was SVR, in which R2 of calibration and validation set were 0.8988 and 0.8771, respectively. Finally, the research result proved the feasibility of monitoring potassium content of citrus leaves, which may provide a theoretical basis for growth monitoring and nutritional diagnosis of citrus trees. 

Transactions of the Chinese Society for Agriculture Machinery (CSAM), in charged of China Association for Science and Technology (CAST), sponsored by CSAM and Chinese Academy of Agricultural Mechanization Science(CAAMS), started publication in 1957. It is the earliest interdisciplinary journal in Chinese which combines agricultural and engineering. It always closely grasps the development direction of agriculture engineering disciplines and the published papers represent the highest academic level of agriculture engineering in China. Currently, nearly 8,000 papers have been already published. There are around 3,000 papers contributed to the journal each year, but only around 600 of them will be accepted. Transactions of CSAM focuses on a wide range of agricultural machinery, irrigation, electronics, robotics, agro-products engineering, biological energy, agricultural structures and environment and more. Subjects in Transactions of the CSAM have been embodied by many internationally well-known index systems, such as: EI Compendex, CA, CSA, etc.

   下载PDF阅读器