Research on infrared image recognition method for substation equipment based on YOLO-CFA

To address the shortcomings of traditional object detection algorithms, such as insufficient recognition accuracy, slow detection speed, and high model redundancy, this paper proposes an infrared image object detection algorithm for substation equipment based on YOLO CFA. Firstly, a lightweight C2-PC (C2-Partial Conv) convolution module is designed to improve the backbone network of YOLOv8n, reducing computational complexity while enhancing the model's feature extraction capability. Secondly, the FWA-RepVGG (Feature Weight Aggregation Network-RepVGG) attention mechanism module is introduced into the neck network to improve the model’s ability to focus on key target regions and strengthen the expression of target edge features. Finally, an AGFPN (Attention Global Feature Pyramid Network) feature fusion module is added to the neck network to effectively integrate high-level semantic features with low-level detailed information, thereby enhancing the model’s multi-scale object recognition capability. Experimental results show that the proposed model achieves a recognition accuracy of 91.2 % on a dataset containing infrared images of 13 classes of substation equipment, representing a 3.1 % improvement in detection accuracy compared to the original algorithm. The model has only 2.95 M parameters and achieves a detection speed of 416.66 frames per second. This demonstrates a favorable balance between high detection accuracy and model lightweighting, providing strong support for deploying the algorithm on edge devices.