Infrared Thermal Image Fault Detection based on YOLOV3-L
Si Gao, Yiting Ruan, Qiancheng Hong, Decai Yin · 2022 IEEE International Conference on Artificial Intelligence and Computer Applications (ICAICA) · 2022
In the substation scenario, since the substation equipment is exposed to nature for a long time, it is easy to generate heat and cause the failure of the substation equipment. In order to effectively detect potential faults, a thermal infrared image detection method based on improved YOLOV3 (You Only Look Once) is proposed. This method is based on the overall framework of YOLOV3, and uses another lighter convolution structure to replace the standard convolution in Darknet-53, thereby reducing model parameters, improving the detection speed of thermal infrared images, and optimizing the use of The latter NMS algorithm optimizes the detection accuracy. The experimental results show that the proposed YOLOV3-L method has almost a 100% increase in FPS compared with YOLOV3, while the accuracy is almost unchanged, reaching a MAP value of 95.4%.