Research on Insulator Defect Detection Based on Improved YOLOv7

Aiming at the problems such as large size change of insulator target, small size of defect target, complex background with more interference and imbalanced samples in the process of insulator equipment and defect image detection in UAV inspection of transmission lines, an improved YOLOv7 based insulator equipment and defect detection algorithm is proposed: To solve the small size problem of insulator defects, convolutional attention mechanism is introduced in the shallow layer of YOLOv7 feature extraction network to focus on defects; In view of the big problem of insulator target scale variation, receptive field enhancement module is introduced at the bottom of YOLOv7 feature extraction network to extract target features of different scales; To solve the background interference problem, improve the YOLOv7 feature fusion network, introduce the display vision center module at the top, and pay attention to the global information and local information at the same time; To solve the problem of unbalance of difficult and easy samples, PEIoU loss was used to replace YOLOv7 coordinate regression loss. On the combined public data set of insulator and its defect UAV images, the average multi-class target detection accuracy of the algorithm reaches 98.50%