基于SlimYOLO的控制箱零件检测方法

Detection method of electrical cabinet parts based on SlimYOLO

控制箱零件检测是控制箱生产过程中的重要环节。采用机器视觉方法可自动识别控制箱内零件的类别及安装位置,及时检测控制箱装配缺陷。然而现有目标检测深度学习模型时效性较低,难以满足控制箱零件在线实时检测需求。本文对YOLOv4目标检测模型进行剪枝和优化,提出了轻量级的目标检测模型SlimYOLO。SlimYOLO改进了特征提取网络结构,压缩了冗余特征层,提高了模型推理速度。同时采用Kmeans++聚类算法对模型anchor框参数进行聚类分析,提升了模型对控制箱的检测效果。基于自主构建的控制箱零件数据集开展了多项对比实验研究,SlimYOLO的平均检测精度为98.08%,较YOLOv4提升0.58%,模型体积缩小9.8%,参数量减少了700万,推理速度提升了10%,为实际工业场景中控制箱零件的快速智能化检测奠定了基础。

The detection of electrical cabinet parts is an important part of the production of electrical cabinets.Machine vision is used to automatically identify the type and installation location of parts in the electrical cabinet,and to detect the assembly defects of the electrical cabinet in time.However,the existing object detection in-depth learning model has low timeliness,which makes it difficult to meet the online detection requirements of electrical cabinet parts.In this paper,the YOLOv4 object detection model is pruned and optimized,and a lightweight object detection model SlimYOLO is proposed.SlimYOLO improves the feature extraction network structure,compresses the redundant feature layer,and improves the detection speed of the model.At the same time,the Kmeans++clustering algorithm is used to cluster anchor box parameters,which improves the detection effect of the model for electrical cabinet parts.Based on the self-built data set of electrical cabinet parts,an experimental study was carried out.The average detection accuracy of SlimYOLO is 98.08%,which is 0.58%higher than YOLOv4,the model volume is reduced by 9.8%,the parameter is reduced by about 7 million,and the detection speed is increased by 10%,which lays a foundation for the fast and intelligent detection of electrical cabinet parts in the actual industrial scene.