The invention designs a low-small slow target photoelectric identification tracking method based on machine learning. Firstly, the direction of the target is determined, and then the orientation angleand the pitch angle of the camera are adjusted so that the target is located in the field of view of the camera. Then the camera reads the image frame by frame, and the on-line detection of the target recognition is performed, the read image is taken as the input of the neural network. After machine learning, the trained network is complete to obtain the output of the network, including the classification of the target and the binding frame of the position. If the output classification belongs to the low small slow target, then go to the next step, otherwise skip the next step, directly readinto the next frame image, target tracking. While ensuring real-time performance, the accuracy of automatic recognition is improved, and the robustness to illumination, target posture and other factors is enhanced. The invention can be used for multi-band fusion imaging equipment, expands the application range of a single identification and tracking algorithm, and improves the adaptability of thealgorithm.