李尚平,郑创锐,文春明,李凯华,甘伟光,李洋.基于改进YOLO v5s的甘蔗切种茎节特征识别定位技术[J].农业机械学报,2023,54(10):234-245，293.
LI Shangping,ZHENG Chuangrui,WEN Chunming,LI Kaihua,GAN Weiguang,LI Yang.Stem Node Feature Recognition and Positioning Technology for Transverse Cutting of Sugarcane Based on Improved YOLO v5s[J].Transactions of the Chinese Society for Agricultural Machinery,2023,54(10):234-245，293.
|
摘要点击次数: 247
全文下载次数: 105
|
| 基于改进YOLO v5s的甘蔗切种茎节特征识别定位技术 [下载全文] |
| Stem Node Feature Recognition and Positioning Technology for Transverse Cutting of Sugarcane Based on Improved YOLO v5s [Download Pdf][in English] |
| 投稿时间：2023-05-30 |
| DOI：10.6041/j.issn.1000-1298.2023.10.023 |
| 中文关键词: 甘蔗智能横向切种工作站 茎节识别与定位 YOLO v5s 边缘端部署 |
| 基金项目:国家自然科学基金项目（52165009） |
|
| 中文摘要:为了实现甘蔗智能横向切种工作站的精准、高效的自动化切种，针对工厂化切种任务的特点，提出了一种基于改进YOLO v5s的甘蔗茎节特征边缘端识别与定位方法。首先，利用张正友相机标定法对摄像头进行畸变矫正；然后对甘蔗茎节数据集进行数据增强，利用原始的YOLO v5s模型进行训练和测试，结果显示数据增强能一定程度上提高检测精度。针对茎节特征目标小以及模型体积大导致检测精度低、部署难度高等问题，对YOLO v5s的骨干网络进行改进，在SPPF特征融合模块前引入坐标注意力（Coordinate attention，CA）模块和Ghost轻量化结构，在Head网络中剔除P5大目标检测头，得到了改进后甘蔗茎节检测模型YOLO v5s-CA-BackboneGhost-p34，测试结果表明该模型优于其他主流算法和原始模型，具有高精度、小体积等优势。其中，平均精度均值1和平均精度均值2分别提高5.2、16.5个百分点，模型浮点数计算量和内存占用量分别降低42%和51%。最后，为了提高检测速度和实时性，将模型部署于边缘端，利用TensorRT技术加快检测速度，并在传送速度为0.15m/s的甘蔗智能横向切种工作站上完成实际切种实验。实验结果表明，加速后茎节检测速度达到95f/s，实时检测定位平均误差约为 2.4mm，切种合格率为100%，漏检率0.4%，说明本文提出的模型具有高度可靠性和实用性，可以为甘蔗横向切种工作站的工厂化、智能化以及标准化应用提供有效的技术支持。 |
| LI Shangping ZHENG Chuangrui WEN Chunming LI Kaihua GAN Weiguang LI Yang |
| Guangxi Minzu University |
| Key Words:intelligent transverse sugarcane cutting workstation stem node identification and positioning YOLO v5s edge-end deployment |
| Abstract:In order to achieve accurate and efficient automated seed cutting in sugarcane intelligent transverse seed cutting workstation, a method based on improved YOLO v5s for identifying and locating the edge end of sugarcane stem node features was proposed for the characteristics of factory seed cutting tasks. Firstly, the camera was corrected for distortion by using the ZHANG Zhengyou camera calibration method, then the sugarcane stem node dataset was enhanced and the original YOLO v5s model was used for training and testing, and the results showed that the data enhancement can improve the detection accuracy to some extent. Then, to address the problems of low accuracy and high model complexity caused by small stem node feature targets, the backbone network of YOLO v5s was improved by introducing the coordinate attention module and Ghost lightweight structure before the SPPF module, and removing the P5 large target detection head in the Head network to obtain the improved sugarcane stem node detection model YOLO v5s-CA-BackboneGhost-p34. The test results showed that the model outperformed other mainstream algorithms and the original model with high accuracy and small size. Among them, mAP@0.5 and mAP@0.5∶0.95 were improved by 5.2 and 16.5 percentage points, respectively, and the model computation and size were reduced by 42% and 51%, respectively. Finally, in order to improve the detection speed and real-time performance, the model was deployed at the edge end, and the detection speed was accelerated by using TensorRT technology, and the model was completed on a sugarcane with transmission speed of 0.15m/s. The actual seed cutting test were completed on the smart transverse seed cutting workstation with transmission speed of 0.15m/s. The test results showed that the accelerated stem node detection speed reached 95f/s, the average error of real-time detection and positioning was about 2.4mm, the seed cutting qualification rate was 100%, and the leakage rate was 0.4%, which indicated that the model proposed was highly reliable and practical, and can provide effective technical support for the industrialization, intelligence and standardization of sugarcane transverse seed cutting workstation. |
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 |
|
|
|
