Abstract: The thermal insulation, waterproofing, isolation, vapor barrier, and leveling layers are critical structural and functional components of substation rooftops.Defects on these layers can significantly impact the performance, lifespan, and personnel safety of substations.This study proposes an improved target detection solution for these surface defects using a you only look once, YOLOv5s algorithm enhanced with the optimal transport assignment（OTA）.The OTA algorithm refines label assignment, providing a more accurate match than traditional threshold methods and balancing the learning between positive and negative samples.Experimental results demonstrate that the OTA-optimized YOLOv5s algorithm can comprehensively utilize image information and learn geometric features, thereby reducing localization loss, object loss, and classification loss.Furthermore, the OTA enhances precision, recall, and mean average precision（MAP）,indicating improved predictive accuracy, integrity, and overall performance of the model.Therefore, the application of the YOLOv5s algorithm combined with OTA optimization for rooftop defect detection in substations holds significant practical value.