LOCATION RECOGNITION AND LOSS DETECTION BASED ON IMPROVED YOLOV4 SHOCKPROOF HAMMER

A new algorithm model based on improved YOLOv4 is presented to identify, locate and detect the loss of shockproof hammer in high voltage line inspection.Purposeful data enhancements were made based on the collected patrol images to expand the dataset. The idea of transfer learning was incorporated, and pre-weights and freeze training were used during model training. The YOLOv4 original trunk feature extraction network was replaced by the lightweight network MobileNet V2, and the deep detachable convolution was applied to the network, which greatly reduced the amount of parameters. By comparing and analyzing the experimental results, the improved algorithm model performs well and meets the requirements of patrol inspection.