Remote sensing image change detection method and system based on style alignment and edge constraint

Abstract

The invention relates to a remote sensing image change detection method and system based on style alignment and edge constraint, and belongs to the technical field of deep learning and image processing. Sending the dual-temporal remote sensing image into an encoder, performing multi-layer feature extraction by using the same feature extraction network, performing style alignment on the extracted deep features by using adaptive instance standardization processing, and performing dual-temporal fusion on the extracted shallow features and the deep features after the style alignment to obtain time difference features; sending the data to a decoder adopting jump connection to carry out multi-scale feature fusion; performing unified up-sampling operation on the decoded features of each layer to the same size, then splicing to obtain a final feature map, and predicting the feature map; and calculating a mixed loss function by using the prediction map and label data to constrain and optimize the model. According to the method, the semantic consistency and the boundary definition of remote sensing image change detection in a complex scene can be remarkably improved, and meanwhile, the increase of the calculated amount is avoided.

Images