In order to improve the accuracy of insulator target detection and shorten the model training time

the InsuNet insulator target identification method based on convolutional neural network is proposed. The InsuNet network used 56 convolutional layers of the backbone network. After each pooling layer of the feature extraction network

a layer of opening operation is added to filter out the interference features around the target features and enhance the robustness of the algorithm. The output feature image through the backbone network was fed into two branches

one was the RPN (region proposal networks)

the other was ROIAlign (region of interest align) processing. The first branch output the regions of interest

then the second one aligned each insulator area and produced class of the whole insulator area and the background area and bounding boxes of each insulator regions. The experimental results show that compared with the Mask R-CNN model

the accuracy of this method is 94.7% when identifying various types of insulators

and the single image processing time is about 0.18 s

which is shortened by about 40 ms.