Abstract: Infrared imaging of power transmission and transformation equipment in substations is often influenced by varying shooting scenes and light intensities, which degrades data quality. This has a significant impact on the positioning and identification of power transmission and transformation equipment in infrared images. To address this problem, an object detection algorithm based on CA-YOLOv8 is proposed, which enhances feature extraction in the backbone network by integrating an improved Coordinate Attention module into the YOLOv8 architecture. Furthermore, by employing SIoU as the regression loss, the algorithm overcomes the gradient vanishing issue of IoU and reduces degrees of freedom, thereby accelerating model convergence. Grad-CAM++ was applied to visualize and verify the model performance on the measured dataset. The algorithm demonstrates higher accuracy compared to mainstream YOLO series algorithms. Based on the loss training process and actual prediction results, the algorithm enables rapid and accurate identification and localization of power transmission and transformation equipment.