基于YOLOv3不同场景辣椒采摘机器人识别定位研究OA
Research on Recognition and Location of Pepper Picking Robot Based on YOLOv3 in Different Scenarios
针对辣椒采摘受环境光、枝叶遮挡和果实重叠的影响问题,构建了基于 YOLOv3 模型和 realsense 深度相机的识别定位系统,研究不同补光位置、枝叶遮挡和果实重叠程度对辣椒识别和定位精度的影响规律.结果表明:模型召回率Recall 达 0.98,平均精度均值mAP 达 0.95,精确率precision 达 0.854;不同光照场景下,识别成功率由高到低依次为正向光、顶光、侧光和背光;轻度枝叶遮挡和轻微果实重叠时,模型识别成功率均保持在 96%左右,综合定位误差最大为 0.024m,满足辣椒采摘机器人识别和定位精度需求.
A recognition and location system based on YOLOv3 model and RealSense depth camera was constructed to study the effects of different light filling positions,shade of branches and leaves and overlapping degree of fruit on the recognition and location accuracy of pepper.The results show that the Recall,mean average precision(mAP)and preci-sion are 0.98,0.95 and 0.854 respectively.In different lighting scenes,the recognition success rate from high to low is forward light,top light,side light and backlight.The recognition success rate was about 96%when there was slight shade of branches and leaves and slight overlapping of fruits.The maximum comprehensive positioning error is 0.024m,which can meet the requirements of pepper picking identification and positioning accuracy.
刘思幸;李爽;缪宏;柴岩;陈福康;王健;董佩璇
扬州大学 机械工程学院, 江苏 扬州 225127扬州市蒋王都市农业观光园有限公司, 江苏 扬州 225127江苏亿科农业装备有限公司, 江苏 扬州 225231
农业工程
辣椒采摘机器人识别定位YOLOv3
pepperpicking robotrecognition and locationYOLOv3
《农机化研究》 2024 (002)
38-43 / 6
国家特色蔬菜产业技术体系岗位专家项目(CARS-24-D-03);江苏省农业科技自主创新资金项目(CX(20)1005);江苏省科技计划项目(BE2019348,BE2019345)
评论