To address the limitations of traditional target detection and recognition methods in complex battlefield environments, a deep convolutional neural network (CNN)-based method for extracting image features and localizing targets is proposed in this paper. Traditional shape and texture features as well as the hotspot information in infrared images are comprehensively considered, and the accuracy of target detection and identification is improved through the training of large-scale labeled datasets and the optimization of the back propagation algorithm. Compared with traditional methods, the method can automatically learn the feature representations in images without relying on manually designed features and classifiers. In order to verify the effectiveness of the algorithm, this paper selects the Hathi Hi3559AV100 as the core processing chip to design the hardware platform, and by porting the algorithm to this platform, the collected data samples are analyzed and tested, and the experimental results show that the system exhibits relatively stable performance in complex background environments, and is able to reliably perform target detection and recognition.