Deep learning-based laser and infrared composite imaging for armor target identification and segmentation in complex battlefield environments

In this paper, a novel method of armor target recognition and segmentation in complex battlefield environments is proposed. First, distance image and infrared image data of armored targets are acquired using LIDAR and IR detectors, and the data are fused based on improved guided filtering to increase the diversity of input features. Secondly, the experimental dataset is expanded using improved GAN to generate simulated data onto different features. Then, effective recognition of armored targets is achieved by adding a CBAM attention mechanism to the YOLOv5 target detection algorithm. Finally, by improving the Deeplabv3 + semantic segmentation algorithm, which mainly lies in adding the CBAM attention mechanism and lightening the algorithm, the segmentation accuracy is improved while improving the real-time performance of detection. The experimental results show that the algorithm proposed in this paper has higher accuracy and real-time performance than the experimental comparison algorithm, and the mean average precision (mAP) can reach more than 99.64 %. And the algorithm can accurately segment the armored targets' combat area, and the mean pixel accuracy (MPA) can reach more than 96.57 %. It is concluded that the algorithm significantly improves the effective identification and precise strike capability of armored targets in complex battlefield environments.