【摘要】 目前我国水果自动化缺陷检测技术不完善的问题亟待解决，针对农业中水果缺陷检测精度低、速度慢、分类少等问题，提出了一种改进的YOLOv5s检测模型，将其应用于脆桃表面缺陷检测。首先通过引入高效通道注意力机制(efficient channel attention, ECA),有效避免降维产生的低效通道权重且增强跨通道互动；其次在Neck部分中引入轻量级卷积结构GSConv以及一次性聚合VoV-GSCSP模块，降低模型参数量；最后使用EIoU损失函数提高模型的定位精度。改进后的模型平均精度均值(mAP)达到98.3%,较未改进之前提高了3.3%,并且参数量降低了4.65%,检测速度达到了106 fps,与其他不同检测算法相比，也具有显著的优势，对自动化分拣分级分类具有实际应用价值。更多还原

【Abstract】 At present, the problem of imperfect fruit automatic defect detection technology needs to be solved urgently in China. In this paper, aiming at the problems of low precision, slow speed, and few classifications of fruit defects detection in agriculture, an improved YOLOv5s detection model was proposed and applied to the surface defects detection of crisp peaches. Firstly, ECA is an efficient channel attention mechanism, which is introduced to effectively avoid the inefficient channel weight generated by dimensionality reduction and enhance cross-channel interaction; Secondly, the lightweight convolution structure GSConv and the one-time aggregation VoV-GSCSP module are introduced into the Neck part to reduce the model parameters; Finally, the EIoU loss function is used to improve the positioning accuracy of the model. The mAP of the improved model reaches 98.3%, which is 3.3% higher than that of the original parameter, quantity decreases by 4.65%, and the detection speed reaches 106 fps. Compared with other different detection algorithms, it also has significant advantages and has practical application value for automatic sorting and classification.更多还原