Abstract: To overcome the deficiency that the fixed filter with single-scale cannot effectively remove the clutter and improve the performance of dim target detection, a dim target detection method based on a multi-scale adaptive sparse dictionary was proposed in this paper. Firstly, an adaptive multi-scale sparse dictionary was learned based on the sparse coding theory, and the sparse coefficient of the original image at different scales was decomposed. Then exponential fitting function was adopted to fit the statistical sparse representation coefficient histogram at the small-scale atom. Finally, the differences in the exponential fitting function for the target and noise in the multi-scale adaptive sparse dictionary could be extracted and applied to detect the target. This sparse dictionary contained the atoms with different scale, the large-scale atom can describe the background of the image, and the small-scale atom can capture the subtle feature. The results show that this proposed method based multi-scale adaptive sparse dictionary could suppress the clutter more greatly and improve the performance of dim target detection more effectively compared to wavelet and Contourlet method.