Abstract: This paper describes how to detect change of land with multi-temporal remotely sensed image. The key is how to match images, detect change and confirm the change threshold. According to the idea of data processing in compute vision research, this paper puts forward a kind of method for change recognition by (2n+1)×(2n+1) judge templet. The result is considerably improved by judge templet than by only one pixel. After many tests we can conclude that the method mentioned in this paper is feasible. The nicety is up 85 percent using this land change detection technique. So that it can replace artificial recognition.