Object detection of transmission line visual images based on deep convolutional neural network

A deep convolutional neural network based method is adopted to detect objects such as tower, glass insulator and composite insulator in visible images of transmission lines. About 600 visible images of 19 different transmission lines are captured by manned helicopter with high-definition camera. All of the images are then annotated manually and segmented into blocks with 4 different labels: background, tower, glass insulator and composite insulator. These blocks are then augmented to around 150 000 training samples which comprise the transmission line image dataset. A five-layer deep convolutional neural network is designed and pre-trained by using Cifar-100 dataset, the trained network is then fine-tuned by using transmission line image dataset. The experimental results show that when detection true positive rate is 90 %, the false alarm rate is less than 10 %, which is obviously superior to the traditional methods. It can be used for the detection of tower, glass insulator and composite insulator in visible images of transmission lines. The detection result can be used as reference for diagnosis or state analysis of transmission lines. This method can be used to detect tower and insulator in visible images of transmission lines, and can be extended to detect other typical objects.