Since the transformer oil works in a high temperature and high pressure transformer, the phenomenon of deterioration and discoloration will occur, so the image of transformer oil obtained by the visual system of the transformer internal fault detection robot has a problem of color distortion and low contrast. This paper describes a transformer oil image fusion enhancement algorithm to improve the accuracy of internal fault point detection of transformers. First, the image is white balanced by the perfect reflection algorithm to eliminate the impact on uneven oil light, the color of the image is made more balanced. Then, conduct adaptive gamma correction on the color correction image to improve the contrast to the image. Finally, a multi-scale fusion strategy is used to fuse the image after the color correction and the image after the adaptive gamma correction to obtain a clear image of transformer oil. The experimental results show that after the algorithm is treated, the color of the image of the transformer oil is more vivid, more detailed. Compared with the original image, the image quality evaluation indicator UCIE, feature point matching, information entropy is significantly improved under test, which provides a clear data for the internal fault detection of the transformer.