Detection of spontaneous explosion of transmission line insulators based on improved Mask-RCNN algorithm

Abstract:
Aiming at the high missed detection rate and slow detection speed of transmission line insulator self-explosion defect detection, a deep learning algorithm based on improved Mask-RCNN is proposed for the detection of insulator selfexplosion defect. First, a fusion factor is introduced to the FPN network of the Mask-RCNN feature extraction and feature fusion link to improve the detection ability of small target self-explosion defects and reduce the missed detection rate. Secondly, an improved NMS algorithm is used in the Mask-RCNN network to replace the classic NMS (nonmaximum suppression) algorithm to improve the speed of real-time detection of self-explosive defects. Finally, the experimental results show that compared with the Mask-RCNN algorithm, the improved algorithm significantly improves the detection speed and accuracy, mAP reaches 96.9%, and has strong recognition robustness.