基于YOLOv5的旋转边界框电容器目标检测OA

To solve the problem that classical YOLOv5 object detection algorithm can only achieve object localization with horizontal rectangular bounding boxes,this paper designs an object detection method of rotated bounding boxes for capacitors based on the YOLOv5s model.This method transforms the angle prediction problem from a regression problem to a classification problem by using circular smooth labels,and describes the loss function of angle prediction using binary cross-entropy loss.Additionally,the original training data is expanded through replication,rotation transformation,and stitching to improve the accuracy and generalization ability of the model.Experimental results on a real capacitors dataset show that the improved object detection algorithm of rotated bounding boxes for capacitors achieves an average accuracy of 83.5%.Compared with the original YOLOv5 model,the predicted object bounding boxes are more consistent with the actual rectangular contours of the capacitors.