Based on the imaging characteristics of micro CCD cameras, this paper analyzes the problem of recognizing small targets at long distances under complex environmental conditions. Using deep learning YoloV4 tiny as the recognition algorithm, and combining the accuracy and speed of recognition, the YOLOv4 tiny network is optimized and improved. On the basis of the original two-layer feature pyramid, a three-layer feature pyramid YOLOv4 tiny network architecture is constructed. The Nvidia Jetson Nano hardware platform is used for design and development. The experimental results show that the optimized algorithm improves the recognition accuracy of targets and ensures the processing speed of the algorithm.