To improve the performance of object detection and recognition in a real-world combat environment, an improved intelligent object recognition algorithm based on infrared saliency object guidance is proposed. It uses the object information in an infrared image to guide deep self-learning in vision images. The improved YOLO-V3 recognition network is based on the Darknet-53 network architecture, using DenseNet instead of the original transfer layer with lower resolution. Classification network pretraining, multiscale detection network training, and other measures are used to enhance feature propagatio n and reuse and fusion performance. Simulation results show that the proposed model can effectively improve the performance of existing object detection and recognition networks.