Synthetic Aperture Radar Images Changes Detection based on Random Label Propagation

Change detection using SAR images has drawn increasing attentions in remote sensing communities. It is important to take advantage of the label information in changed and unchanged pixels classification. However, most existing methods ignore the fact that the training process may get corrupt when noise labels exist. To overcome the problem, in this paper, we study the influence of the label noise in SAR image change detection, and introduce a random label propagation (RLP) algorithm to cleanse the label noise. The key idea of RLP is to use the probability transform matrix that considers the prior information simultaneously to propagate the label information. Experimental results on two real SAR datasets demonstrated that the proposed method can effectively reduce the noisy labels and therefore improve the change detection performance.