Deep Learning to Detect Stone Burial Sites from Drone Images in the Southern Foothills of the Tianshan Mountains in Xinjiang, China

Stone burial sites (SBSs) are widespread across Eurasia, particularly in Xinjiang, China, but their spatial distribution and abundance remain poorly documented. We combined drone imagery and You Only Look Once (YOLO) models to develop an intelligent method for rapid SBS survey and inventory. Using drone imagery and field-verified sites, we constructed an augmented deep-learning dataset, compared several YOLO models, and proposed three optimization strategies. The results indicated that the YOLO family model has good identification ability for SBS targets, achieving a precision and recall above 0.84. The proposed strategies effectively enhanced the model’s accuracy in identifying SBS, particularly for small and less distinct targets, with precision, recall, and mean average precision (mAP_0.5) exceeding 0.90, and the improved model successfully identified SBS in three unexplored geographical regions. Future continued expansion of SBS sample data could potentially extend our method to establish inventories for all SBS present in other unknown regions.