Multi-spectral image can be simultaneously obtained from the spatial and spectral dimensions of the detected object information, the acquired information contains both two-dimensional spatial information, but also contains the spectral information with the object wavelength change, so multi-spectral image fusion technology is widely used in the field of power detection. In this paper, multispectral image fusion technology is applied to the defect detection of composite insulators in ultra-high voltage transmission lines, and the defect detection method of composite insulators is innovated. Research based on improved ORB algorithm to realize multi-spectral image alignment of composite insulator defects. Propose non-uniform texture complex surface defects saliency fusion algorithm, so as to realize multi-spectral image fusion of composite insulators. Least squares method, pixel integral projection method and other methods are used to detect the defects such as self-detonation, cracks and fouling of composite insulators. The robustness of the improved ORB algorithm used in this paper is better than that of the traditional ORB algorithm in rotational transform, scale transform and rotational plus scale transform image alignment, and the saliency image fusion method shows obvious superiority in the test. The detection rate of the second channel for the three types of defects of composite insulators is above 94.80%, which fully demonstrates that the defect detection method for composite insulators applied in this paper is practicable.