Aiming at the identification of infrared dim targets from complex background,a multi-source information fusion recognition method for medium-wave and long-wave infrared images is proposed. Firstly,during the preprocessing of infrared images with low signal-to-noise ratio,an improved method for decorrelation and block threshold segmentation is proposed to suppress the nonlinear influence of the temperature field,extract the target and its characteristics effectively. Then the target relativity based on the grey theory is calculated and assigned the basic probability and Dempster-Shafer evidence theory based combination rule is used to fuse the confidence of multiple observation samples. The final recognition results are obtained from the simultaneous data of the multi-sensors. Experiments show that the method has high recognition efficiency for the dim small targets under the complex background.