The method of insulator recognition based on deep learning

The insulator is an import part of transmission line, and the defects detection of insulator rely deeply on the insulators' position. Traditional methods about insulator recognition task are depend on color features and geometric features, those methods would be influenced by lots of factors, such as illumination and background in result getting poor generalization ability. In this paper, we propose a method to recognize insulator based on deep learning algorithm. Firstly, we construct the training dataset which includes insulator, background and tower three categories. Secondly, we initialize the convolution neural networks as a six-level network, and adjust training parameters to train the model. Lastly, the trained model is used to predict the candidate insulator position. With the help of non-maximum suppression algorithm and line fitting method, we can get the exactly location of insulator. The experiment results on UAV dataset show the proposed method can effective localize the insulator and improve generalization ability significantly.