Abstract: Aiming at the problem that insulator detection methods have poor effect in extracting detailed features of small targets, a new insulator fault identification scheme based on image fusion is proposed. First, conduct a series of preprocessing on the original image and use DSAGAN to fuse the infrared and visible light images of insulators, which retains the image details while improving the model stability. Then, the YOLOv8 object detection algorithm is utilized to conduct fault identification on the fused image. The experimental results show that the insulator images after DSAGAN fusion have multiple evaluation indicators superior to those of the other seven fusion strategies. For insulator fault detection, YOLOv8 achieves an average F1 score of 0.91 for the four types of faults at a confidence level of 0.410. It demonstrates enhanced capability in extracting features of small-target defects, thus holding certain practical application value.