Abstract: Aiming at the contradiction between high compression ratio and high fidelity of the return image in man-in-the-loop (MITL) of infrared imaging guidance weapons, a hierarchical infrared image compression algorithm based on visual attention model was proposed. On the basis of salient regions of infrared image extracted by visual attention model, salient regions, background regions and transition regions were divided and compressed by different compression methods according to the importance of information of different image regions, in this way the compression of the whole image was realized. The experimental result shows that the proposed algorithm can not only preserve the important information of typical targets, but also decrease the data size in image transmission, and it is more suitable for MITL operation mode than the traditional full image compression method.