With the rapid advancement of urbanization, the number of manhole covers has grown exponentially, while traditional manual inspection methods suffer from low recognition accuracy, high labor costs, and delayed identification of safety hazards. To address these challenges, a research team has developed an autonomous defect detection system that integrates drone-based aerial imaging with an enhanced YOLOv8 architecture. The system innovatively combines an anchor-free detection mechanism with a dynamic label assignment strategy, and employs multi-scale feature fusion to improve detection performance in complex urban scenarios, significantly enhancing test accuracy. A core innovation of the system is the introduction of a lightweight convolutional attention module (CBAM), which enhances small object recognition by performing dual recalibration across both channel and spatial dimensions. Experimental results show that, compared to traditional methods, the system achieves a significant improvement in operational efficiency and can operate continuously and stably for over eight hours. Under complex backgrounds and varying lighting conditions, the system achieved a recall rate of 93.23% for detecting anomalies such as damaged or missing manhole covers, greatly reducing the rate of missed detections. The optimized model demonstrates superior overall performance in terms of coverage density and response speed. The research offers a new paradigm for intelligent urban infrastructure management and provides a valuable technical reference for integrating edge computing with aerial inspection systems, revealing broad application potential in areas such as pipeline monitoring and bridge inspection.