This paper addresses the issues of blurred details, low contrast, and feature degradation in insulator images under harsh meteorological conditions, as well as the challenges of high computational complexity and insufficient real-time performance when deploying existing deep learning models on edge devices. It proposes a lightweight insulator defect detection method that integrates an improved image enhancement algorithm. The method introduces Mahalanobis distance-based modulation weight optimization for scene depth estimation and improves the color decay prior model to effectively enhance foggy insulator images. It further designs a lightweight detection network integrating region-aware routing attention mechanisms, utilizing multi-scale feature fusion strategies to achieve precise insulator identification and localization. Experimental results demonstrate that the proposed method significantly enhances inference speed while maintaining detection accuracy, effectively adapting to edge computing devices. This provides a viable technical solution for real-time deployment in intelligent transmission line inspection systems.
Zheng et al. (Tue,) studied this question.