Abstract: Aimed at the problems of loss of result information, low global contrast, slow model running efficiency in the field of infrared and visible image fusion, a fast perception optimization algorithm for infrared and visible image fusion is proposed. This method decomposed the source images into detail layers and basic layers for follow-up processing, on this basis, a detail optimization network was designed to further extract the features of the optimization detail layer. In addition, the network also used the lightweight Ghost Module instead of traditional convolution to improve the running efficiency of the model. The experimental results show that compared with the previous methods, the proposed method has excellent results in the subjective visual comparison and objective metric evaluation, and has the efficiency of image processing.