The invention discloses a remote sensing image change detection method based on a sparse automatic code machine, mainly for solving the disadvantages of too many undetected and wrongly detected pixels of a change detection result and incapability of using implicit information of a difference graph during direct processing of the difference graph in the prior art. The method is realized by the following steps: 1, reading images; 2, constructing the difference graph; 3, training the sparse automatic code machine; 4, extracting characteristics; 5, carrying of fuzzy clustering on the characteristics; 6, performing deblurring processing; and 7, outputting a change detection result. According to the invention, the difference graph is extracted firstly by use of the sparse automatic code machine and then change detection is carried out on the remote sensing images by use of traction of the characteristics, such that the quantity of the undetected and wrongly detected pixels in the change detection result is reduced, and the precision of change detection is improved.