Improved Faster RCNN-Based Metal Particle Detection on the Inner Surface of GIS Equipment

Abstract

To focus on the problem of low detection rate of metal particles on the inner surface of GIS equipment, an improved Faster RCNN GIS equipment inner surface metal particle target detection network is proposed.The residual network is introduced as the feature extraction network based on the feature of small metal particles on the inner surface, and the feature pyramid network and receptive field block are combined with the residual network,three feature extraction networks, VGG16, ResNet50,ResNet50+RFB-FPN, are used to experimentally compare 2000 metal particle maps.The results show that the accuracy and recall of ResNet50+RFB-FPN feature extraction network are 97.3% and 83.4%, which are improved compared with both VGG16 network and ResNet50 network. It is concluded that the improved Faster RCNN-based metal particle identification on the inner surface of GIS equipment can meet the requirements of detection accuracy and speed.