A Human Body Infrared Image Recognition Approach via DCA-Net Deep Learning Models

With the continuous exploitation of coal resources, human safety has been seriously threatened during the mining process. Therefore, it is of great significance to establish an efficient human infrared image recognition system. In this paper, three classes of infrared image data of the human body are collected by a thermal imager, namely Human, Human others and None. According to the characteristics of downhole infrared images, a distributed channel feature extraction module (DCFE) is designed, and DCA-Net is proposed based on this module. The experimental results show that the recognition rate of the network reaches 98%. Compared with other networks, this network has better recognition performance. Among them, the recognition rate of DCA-Net50 reaches 98.214%, the amount of parameters and calculations are relatively small, and the cost-effectiveness is the highest. It is suitable for the human body infrared image recognition system that requires high accuracy and high real-time performance.