基于机器学习的中耕期甘蔗幼苗识别定位  

Recognition and location of sugarcane seedlings in intertillage period based on machine learning

作　　者：李威 李尚平[1] 潘家枫 李凯华[1] 闫昱晓 Li Wei;Li Shangping;Pan Jiafeng;Li Kaihua;Yan Yuxiao(School of Artificial Intelligence,Guangxi University for Nationalities,Nanning,530006,China;School of Mechanical Engineering,Guangxi University,Nanning,530004,China)

机构地区：[1]广西民族大学人工智能学院,南宁市530006 [2]广西大学机械工程学院,南宁市530004

出　　处：《中国农机化学报》2021年第11期130-137,共8页Journal of Chinese Agricultural Mechanization

基　　金：亚热带农业生物资源保护与利用国家重点实验室科教结合创新基地课题(SKLCUSA—a202007)。

摘　　要：针对甘蔗中耕培土过程中,由于甘蔗植株种植垄间距的改变,导致中耕培土机难以兼顾左右两侧垄的甘蔗植株而引起甘蔗植株培土不到位、不充分的问题,即“火山口”现象,结合多功能甘蔗中耕培土机械的开发,提出一种基于机器学习的中耕期甘蔗幼苗识别定位和坐标分类计算方法。该方法通过YOLOv4网络建立识别网络模型,对中耕期甘蔗幼苗根部和土壤接触的局部区域进行识别定位和坐标获取,然后采用支持向量机将坐标数据分成两组,并分别对每组坐标数据进行实时计算处理,得到两组坐标数据的倾斜值,后续中耕设备将根据倾斜值的数值大小进行调整,改变供土量和供土方向。试验结果表明,采用基于YOLOv4识别网络模型对甘蔗幼苗的识别准确率可达95.50%,采用支持向量机对甘蔗幼苗坐标分类的准确率可达92.60%,实现了对中耕期甘蔗幼苗的实时动态识别及分类计算,为智能甘蔗中耕植保联合作业机械的开发提供数据基础。Due to the change of the ridge spacing of sugarcane plants,it is difficult for the conventional cultivator to control the quality of soil cultivation on both sides of the ridge,which causes inadequate soil cultivation,i.e.,the“crater”phenomenon.We thus developed a multi-functional sugarcane intertillage cultivator and proposed a machine learning-based method to locate and identify sugarcane seedlings and calculate the coordinate classification during the intertillage period.Based on the YOLOv4 network,the method established a recognition model,which identifies and locates the area between the roots and the soil and accesses the coordinates.Then,the Support Vector Machine divided the coordinate data into two groups and processed them separately in real-time to obtain the tilt value.Finally,the data were adjusted according to the tilt value to modify the quantity and direction of soil supply.The results of the present study indicate that in use of the YOLOv4 recognition model,the recognition accuracy can reach 95.50%,and the classification accuracy using the SVM is 92.60%,which realizes the real-time dynamic recognition and classification calculation of sugarcane seedlings in the intertillage period,and provided data foundation for the development of intelligent sugarcane protection and joint operation machinery.

关 键 词：甘蔗中耕培土 YOLOv4网络 识别定位 分类计算 支持向量机 倾斜值 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]