Thermal Defect Detection in Power Equipment Based on YOLOv8s-SCINet-GAM Model with Low-Light Image Enhancement algorithm

In context of power vision, thermal defect detection in power equipment is critical. The detection effect of traditional deep learning models is often unsatisfactory in low-light environments, mainly because the image quality in low-light conditions decreases, which makes it difficult for the models to extract effective features. Aiming at the above difficulties, this paper designs a YOLOv8s-SCINet-GAM model based on a lowlight image enhancement algorithm for thermal defect detection in power equipment. For the problem that the detection accuracy of the traditional YOLOv8 model decreases under low-light conditions, we combine the SCINet network and the global attention mechanism GAM to improve the robustness of the model in complex environments. By enhancing the information such as contrast and details of low-light images, the improved model significantly increases the detection accuracy and detection speed of thermal defects in power equipment. Dealing with low-light thermal images, the mAP of the improved model reaches ${9 6. 8 \%}$, and FPS reached 161.29, which provides strong support for the intelligent maintenance of power equipment.