Detection Method for Insulators in Power Transmission Lines Based on Improved RetinaNet Algorithm

Insulator detection in aerial images plays a crucial role in the inspection of transmission lines, as it is vital for maintaining the safety and sustainability of the power system. This paper presents an enhanced algorithm for detecting insulators in transmission lines using the RetinaNet model. Our approach first utilizes ResNet101 instead of ResNet50 to extract more comprehensive feature information. Additionally, we replaced the original L1Loss with the Distance-IoU (DIoU) loss function to improve the accuracy of bounding boxes. Experimental results demonstrate that our method achieves a 1.2% increase in mean average precision (mAP) and a 1.7% increase in average recall (AR) compared to the original RetinaNet algorithm. These findings indicate that our enhanced approach effectively improves the performance of insulator detection.