A Fire Detection Method for Power Transmission Lines Based on Improved YOLOv3 Algorithm

Fire is a great threat to transmission lines. It can save a lot of manpower to use computer visual technology for fire detection in the fire prevention and control of transmission lines, and has a very positive significance. However, it has very high requirements to the real-time, credibility and stability of the detection algorithm. Therefore, we propose a lightweight and reliable flame detection algorithm based on YOLOv3 algorithm. We developed the backbone network with the depthwise separable convolution, which makes the model more streamlined and improves the real-time performance. In addition, we added the attention mechanism and the fusion features detection head to make the features output by the backbone network more expressive, and also improved the efficiency of target prediction. Finally, we designed a discriminator based on Siamese neural network to confirm the detection results of the detection network, which improves the credibility of the algorithm. Experiments show that the proposed fire detection algorithm has high real-time and credibility.