温室复杂环境下茄子及果梗快速检测方法  

Abstract ID:1003-188X(2025)07-0018-EARapid Detection of Eggplant and Fruit Stem inComplex Greenhouse Environment

作　　者：魏喜安 白龙[1] 闫涛 郭嘉褀 Wei Xi′an;Bai Long;Yan Tao;Guo Jiaqi(School of Mechanical and Electrical Engineering,Beijing Information Science and Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学机电工程学院,北京100192

出　　处：《农机化研究》2025年第7期18-24,34,共8页Journal of Agricultural Mechanization Research

基　　金：国家自然科学基金项目(11802035);北京市科技计划一般项目(KM201911232022);北京信息科技大学“勤信英才”项目(5112111110)。

摘　　要：为了满足茄子采摘机器人对温室不同光照、遮挡环境下茄子与果梗的快速准确检测,该研究提出了一种轻量化YOLOv5s茄子检测模型。首先,使用MobileNetV3替换YOLOv5s特征提取网络;然后,将轻量级CBAM注意力机制嵌入MobileNetV3骨干网络中,增强特征提取能力,在保证具有良好精度的前提下,减少模型的参数量;最后,试验确定使用WIoU替换CIoU作为边界回归损失函数。基于自制不同光照下茄子数据集,对改进后的YOLOv5s算法模型进行测试。试验结果表明:改进后的模型相较于原模型参数量减少49.7%,计算量减少61%;使用该模型对不同光照条件的茄子及果梗的平均检测精度为95.2%,相较于原模型提高1.2%;在GPU下的检测速度为55.6帧/s, CPU下的检测速度为10.4帧/s。研究表明:改进后算法可以满足茄子采摘机器人对茄子采摘的实时检测要求。In order to meet the fast and accurate detection of eggplant and fruit stems under different illumination and shading environments of greenhouse by eggplant picking robot,this study proposed a lightweight YOLOv5s eggplant detection model.Firstly,MobileNetV3 was used to replace the YOLOv5s feature extraction network,and then the lightweight CBAM attention mechanism was embedded into the MobileNetV3 backbone network to enhance the feature extraction capability and reduce the number of parameters of the model while ensuring with good accuracy.The final experiment determined the use of WIoU instead of CIoU as the boundary regression loss function.The improved YOLOv5s algorithm model was tested based on the homemade eggplant dataset under different illumination.The test results showed that the improved model reduced the amount of parameters by 49.7%and the computation by 61%compared with the original model,and the average detection accuracy of eggplant and fruit stems under different illumination conditions using this model was 95.2%,which was 1.2%better compared with the original model.The detection speed was 55.6 frames/s under GPU and 10.4 frames/s under CPU,which showed that the improved algorithm can meet the requirements of eggplant picking robot for real-time detection of eggplant picking.

关 键 词：茄子 果梗 YOLOv5s 轻量化 光照变化 

分 类 号：S223.14[农业科学—农业机械化工程]