Infrared Target Recognition Method of Substation Equipment Under Complex Background Based on Improved YOLOv8

Equipment target recognition is an important step in the intelligent inspection technology of substation equipment infrared images. It addresses common challenges such as complex backgrounds, multiple equipment types with mutual occlusions, and poor quality of infrared images. By introducing the C2f _ SimAM module to replace the C2f module in the feature extraction stage, the feature extraction ability of the model is significantly improved. Next, a small target detection head is added, and the SCConv module is incorporated to reduce redundant information during feature fusion, thereby enhancing the spatial and channel expression ability of the input features. Then, the model's loss function CIoU is replaced with WIoU_v3 to suppress the damage caused by low-quality labeled samples. Finally, Soft_NMS is introduced to improve the detection performance of the model in densely overlapping equipment environments. Experimental validation shows that the mean average precision (mAP0.5) of the model is increased to 93.9%, the number of model parameters is reduced to 3.1M, and the Frames Per Second reaches 135.