SMA-YOLO: A Defect Detection Algorithm for Self-Explosion of Insulators Under Complex Backgrounds

Abstract

Insulators are essential components in power transmission systems, and identifying faults in them promptly through UAV imaging is vital for maintaining grid reliability and operational safety. In order to reduce the rate of missed and false detection of insulator self-explosion defects under the interference of complex background, we propose an advanced model named SMA-YOLO, based on YOLOv11. We propose the SALSK module. It improves the detection precision for objects of varying sizes. Furthermore, we redesign the core module C3k2 in YOLOv11 by incorporating a multi-scale feature extraction mechanism, which can aid in better distinguishing between defects and background noise and reduces the impact of complex backgrounds. Additionally, we propose the ASSPPF module. The module integrates detailed information across various scales, strengthening the detection capability for small objects. Experimental results demonstrate the superior performance of the proposed model across two distinct datasets. The model achieves mAP@0.5 values of 95.5% and 99.2%, and mAP@0.5:0.95 values of 84.3% and 84.4%, respectively. Compared to the baseline model, mAP@0.5:0.95 improved by 2.5% and 2.0%, validating the robust generalization and resilience of the model. These findings confirm that SMA-YOLO holds high practical value and application potential in the insulator inspection tasks.