摘要
针对棉花采摘机器人自动采摘棉花过程中需要实时检测并定位棉花的问题,对基于双目视觉的棉花采摘检测与定位技术进行了研究。在YOLOv5s模型中,首先利用Mosaic数据增强技术把CSP结构融入Darknet53,减少模型计算量和内存占用量,然后将模型Neck中的SPP替换为SPPF,解决目标多尺度问题,提高模型计算速度,最后使用搭建的棉花检测与定位系统对不同数量和不同朝向的棉花进行检测并定位。结果表明:该系统能够快速准确定位不同数量和不同朝向的棉花,检测精度达99.4%,基本可以完成棉花检测与定位任务。
Regarding the problem of real-time detection and positioning of cotton during automatic cotton picking by cotton picking robot,this paper studies the detection and positioning technology of cotton picking based on binocular vision.In the YOLOv5s model,data enhancement was performed using Mosaic.Integration of CSP structure into Darknet53 reduces model computing and memory usage.The SPP in the model Neck is replaced with SPPF to solve the multi-scale problem of the target and improve the calculation speed of the model.Finally,the established cotton detection and positioning system is used to detect and position cotton of different quantities and directions.The results show that the cotton positioning system can quickly and accurately locate different amounts and orientations of cotton,the detection accuracy reaches 99.4%,and can basically complete the cotton detection and positioning task.
作者
吕俊
韦海诚
LU Jun;WEI Haicheng(College of Automation,Zhejiang Polytechnic University of Mechanical and Electrical Engineering,Hangzhou 310000,China;School of Mechanical and Electrical Engineering,Aksu Vocational and Technical College,Aksu 843000,China)
出处
《河南工程学院学报(自然科学版)》
2025年第2期64-70,共7页
Journal of Henan University of Engineering:Natural Science Edition
基金
浙江省教育厅一般科研项目(Y202147344)。
