Abstract: With the development of remote sensing technology, change detection for remote sensing image provides an effective method in environmental monitoring, disaster relief and many other fields. However, it is still a challenging problem to develop more effective change detection methods due to the complexity of ground-truth and the difficulty of labeling the samples and so on. This paper proposes a remote sensing image change detection method based on semi-supervised discrete potential theory. The suggested method first uses a new method to label the samples to get the training set, then constructs complex network by KNN approach. Finally, it improves the classical Wu-Huberman algorithm in complex network and divides the network. As a result, the obtained two community structures exactly correspond to the change part and the invariant part. Experimental results show that the change detection method based on semi-supervised discrete potential theory has perfect change detection performance.

Key words: remote sensing image, change detection, semi-supervised classification, discrete potential theory, Wu-Huberman algorithm