Abstract:To address the issues of insufficient accuracy in spot target recognition and poor tracking stability in air-to-ground laser communication systems, this paper investigates intelligent recognition and stable tracking methods for weak spots under complex environmental conditions. A real-time detection model for weak spots based on the YOLOv8 target detection network was constructed. During training, a SAM optimization strategy was introduced to constrain the model to converge to a flat optimal solution, thereby improving its generalization ability and robustness under atmospheric turbulence, platform jitter, and background light interference. Combined with the ByteTrack multi-target tracking algorithm, cross-frame target matching was achieved using a joint association mechanism of high-confidence and low-confidence detection boxes, thus maintaining trajectory continuity even with fluctuations in spot brightness or a decrease in detection confidence. Image acquisition and algorithm verification experiments were conducted on an air-to-ground laser communication experimental platform. Under noise interference conditions, the model's mAP50 improved from 0.607 to 0.711, and mAP50-95 improved from 0.286 to 0.339 after introducing the SAM optimization strategy. Combined with the ByteTrack algorithm, stable and continuous tracking of spot targets was achieved under atmospheric turbulence disturbance conditions, effectively reducing the probability of target loss. Experimental results show that this method can improve the stability of spot recognition and tracking in air-to-ground laser communication systems, meeting the application requirements for stable tracking in complex environments.