Abstract: Image fusion is one of the important contents in the field of image processing. The traditional fusion algorithm fuses the source images and processes them according to certain rules. Although a good fusion effect can be achieved, the algorithm requires high image registration, then the fusion image also has the problem of loss of details, and the problem that the target is not obvious enough. To improve the above problems, the characteristics of the infrared image, visible image and the infrared target were analyzed, target detection was used in image fusion, and the DOG filter was used to extract the targets in the infrared image. The fusion coefficient matrix was obtained through multi-scale DOG image calculation, and then the fusion sub-map was calculated. Finally, a fusion image with obvious targets and good details was obtained,and the requirement for image registration was reduced. Five commonly used evaluation indicators, as well as the signal-to-clutter ratio and background similarity, were used to evaluate the fusion image. Experimental results show that the proposed algorithm is superior to the commonly used image fusion methods in both subjective vision and objective evaluation indicators.