Abstract: The accurate recognition and detection of infrared thermal images of substation equipment is a prerequisite for intelligent analysis of its thermal status. To address the interference posed by complex backgrounds, this paper presents an improved light weight YOLOv7 method aimed at enhancing the recognition performance of substation equipment under intricate infrared conditions. The proposed approach introduces a high-resolution P2 detection head to improve small target detection, the parameter-free attention module SimAM effectively extracts target features of various substation equipment under the complex infrared backgrounds. Additionally, the CARAFE module minimizes the loss of feature information during the upsampling process, further bolstering the algorithm's robustness. Experimental results demonstrates that the proposed model surpasses the original YOLOv7-tiny by 2.6% in detection accuracy, achieving a real-time inference speed of 101 FPS. It is proved that the proposed model has advantages in infrared image target recognition of substation equipment, especially small substation equipment, and the proposed model has higher recognition and detection accuracy than other lightweight models.

    Key words: substation equipment; infrared image; target recognition and detection; computer vision; deep learning