Smart Grid Insulator Detection Network Improved based on YOLOv8

Abstract

Insulators are critical components of power transmission lines. Due to environmental changes, insulators may fail, making timely and effective detection of these failures a pressing issue. However, the detection of inclined insulators faces challenges, such as inadequate fitting of detection frames and excessive background noise within the target frames. To address this, this paper proposes an improved inclined insulator detection network (RCAS-YOLOv8). To resolve issues related to feature sparsity and effectiveness, a non-local module with row and column-level sharing is introduced by considering the correlations between feature points. Finally, the task of locating the four vertices of the insulator is completed by summing the predicted offsets of the target frames four vertices. Experimental results show that the proposed RCAS-YOLOv8 algorithm has achieved significant improvement in the detection of tilted targets in the Power Line Insulator Dataset (CPLID), with high detection accuracy, in which the APR index of our method reached 0.891.