Remote sensing image change detection method based on deep neural networks

A remote sensing image change detection method based on deep learning was proposed to obtain reliable training samples and improve the accuracy of remote sensing image change detection. Firstly, texture features (gray co-occurrence matrix method) are selected by structural similarity method (SSIM), and the multitemporal remote sensing image difference image (DI) and textural feature DI obtained by change vector analysis (CVA) are fused to construct the final DI, then the difference images are denoised by the constructed variational denoising model. Secondly, the frequency domain significance method is used to obtain the DI saliency map, and the coarse change detection map obtained by selecting a threshold for the DI saliency map is pre-classified (change, unchanged and undetermined) by the fuzzy c-means (FCM) clustering algorithm. Finally, the neighborhood features of the changed pixels and unchanged pixels extracted from remote sensing images are introduced into the deep neural network model for training, and the trained deep neural network model is used to detect the changes in multitemporal remote sensing image, then the final change detection map is obtained. Experiment on three real remote sensing image data sets shows that the change detection accuracy of the proposed method is higher than that of other comparison methods. © 2020, Zhejiang University Press. All right reserved.