Abstract: To address challenges in substation equipment defect detection—such as severe background interference, large variations in defect scales, and imbalanced sample distribution—this study focuses on a 500 kV substation and proposes an improved YOLOv8 model by integrating coordinate attention (CA) and an adaptive spatial attention gating unit. This enhancement optimizes feature extraction and multi-scale fusion strategies. Experimental results demonstrate that the proposed method significantly improves the simultaneous detection capability of both small-scale and large-scale defects under complex environmental conditions, providing effective technical support for intelligent operation and maintenance of power equipment.