•  
  •  
 

Keywords

GIS equipment, SF6 gas leak detection, Real-ESRGAN, improved YOLO 11 model, GAM attention mechanism, fusion detection head

Abstract

An SF6 gas leakage detection method for GIS equipment based on an improved YOLO 11 model is proposed to address the problem of internal SF6 gas leakage in outdoor substation gas insulated switchgear （GIS） equipment caused by environmental factors and equipment aging.The Real-ESRGAN super-resolution generative adversarial network is adopted for denoising and enhancement，and then the YOLO 11 model is improved to solve the problems of missed detection of small gas plumes，false detection due to cloud interference，and inability to perform multi-task detection in the original YOLO 11 model.The improved model increases the number of detection heads to four and incorporates a GAM attention mechanism based on the YOLO 11 model，replacing the original detection heads with fused detection heads to simultaneously accomplish segmentation detection and keypoint detection tasks.Finally，simulation experiments as well as simulation and field experiments carried out by a self-developed SF6 gas leakage inspection robot are performed to verify the feasibility and effectiveness of the proposed method.Simulation experiments show that the improved model achieves ImAP 50 and ImAP 50-95 values of 85.2% and 54.4% for the segmentation task，and 96.3% and 96.5% for the keypoint task，respectively，outperforming the YOLO 11 model.Simulation and field experiments using a self-developed SF6 gas leakage inspection robot further verify the leakage detection effectiveness of the improved model，demonstrating that the proposed method can replace manual labor in gas leakage detection.

DOI

10.19781/j.issn.1673-9140.2026.04.027

First Page

327

Last Page

339

Download

Share

COinS