Image Classification Algorithm for Transmission Line Defects Based on Dual-Channel Feature Fusion

The power system is of great significance to the normal production of society and the daily life of the people, so regular inspection of transmission lines is essential. However, transmission lines are usually exposed to the outdoors, and the surrounding terrain and environment are complex, which may lead to problems such as structural aging and mechanical strength reduction, which in turn may lead to large area power outages and cause huge economic losses. In this paper, a two-channel feature fusion classification method is proposed to address the transmission line image classification problem. Using a two-channel parallel network structure, a neural network model is constructed to fuse the overall and local feature information, and then determine whether there are defects in the transmission line images. The experimental results show that the classification accuracy of the two-channel parallel convolutional neural network based on ResNet32 is 82.24% and 77.87% for the bird’s nest defect and insulator burst defect, respectively, on the actual transmission line image dataset, which exceeds the classification accuracy of other CNN models. This indicates that the classification accuracy can be effectively improved by fusing feature information.