GCW-YOLO: an improved infrared image detection algorithm for substation electrical equipment based on YOLOv8n

In intelligent substation inspection systems, thermal faults in electrical equipment can be diagnosed by examining infrared images. However, due to the complex background of the substations' environment and the close arrangement and occlusion of equipment, traditional algorithms face difficulties in extracting features from infrared images and achieving high detection accuracy. To address these problems, a network called GD-CReToNeXt-WIoUv3-YOLO (GCW-YOLO) is proposed to analyse infrared images of electrical equipment. Firstly, the model integrates an improved gather-and-distribute (GD) mechanism within the neck network, improving the extraction of multi-level features and enhancing information interaction. Secondly, the introduction of the CReToNeXt module improves the performance of the model in detecting overlapping targets. Additionally, Wise-IoUv3 (WIoUv3) has been adopted, reducing the false positive rate of the model. Detection experiments on a substation infrared image dataset of six types of electrical equipment demonstrate that the proposed model achieves high detection precision and recall. In particular, it can give the highest mAP@0.5 and mAP@0.5:0.95 results compared to other models, reaching 90.88% and 70.26%, respectively. Furthermore, experiments show that the proposed GCW-YOLO model also possesses noise resistance capabilities. Taken together, this improved algorithm can accurately recognise infrared images of electrical equipment in substations.