Based On Improved Yolov5 Mine Infrared Target Detection Algorithm

The environment under mine is dimly complex and has low visibility. The infrared detection equipment has a better target detection effect than visible light equipment in this environment. Therefore, the monitoring of personnel and vehicles by infrared thermal imaging monitoring equipment can accurately determine the vehicle-to-person position and generate an alarm through the edge end and embedded equipment, which can guarantee the safety of personnel. This paper addresses the problem of low detection accuracy in IR thermography, proposing to improve the detection of targets by the yolov5s network by adding a se-attention mechanism to the head of the neck network, improving the feature extraction capability of the network. Aiming at an embedded platform with limited resources, the lightweight convolution gsconv is introduced to replace the original convolution, being able to maintain accuracy while reducing the computational effort of the model. And to achieve mine car and human detection on the embedded side ends, the Yolov5s-gconv-se presented in this paper, with a map lift of 3.2% versus the original yolov5s and a frame rate of 2 frames per second, provides better accuracy for target detection.