The invention discloses an image change detection method which comprises the following steps: acquiring two time-phase images including a first image and a second image; obtaining a first difference image by using the two time-phase images, and training the first difference image to obtain a second difference image; classifying the second difference image by using the MRF to obtain an initial classification result, the initial classification result comprising a probability value and a first classification result; inputting the probability value into a unary potential function to obtain first energy, inputting the first classification result into a binary potential function to obtain second energy, and summing the first energy and the second energy to obtain total energy; segmenting the total energy by using a CRF method to obtain a second classification result; and circulating the second classification result by using an iteration method to obtain a final result. According to the method, the second difference image is obtained after the first difference image is trained, the second difference image serves as the initial detection image of the MRF random field, noise interference isrestrained, and the MRF random field is combined with the CRF random field, so that the accuracy of the detection result is improved.