SAR image change detection based on multi domain convolution and self attention mechanism

DOI:	10.11809/bqzbgcxb2023.07.038
Keywords:	synthetic aperture radar (SAR) image; change detection; multi domain convolution; dilated convolution; self attention mechanism
Abstract:	As for the change detection methods of Synthetic Aperture Radar (SAR) images based on deep convolutional neural networks, it is easy to introduce noise to marginal areas in the process of using image blocks to analyze features, resulting in poor accuracy of SAR image change detection. To solve this problem, this paper presents a change detection method of SAR images based on multi domain convolution and self attention mechanism. Firstly, this method enhances the central region of the input SAR image blocks through a multi domain convolutional model to reduce the influence of edge noise. Then, the improved self attention mechanism model of the injected dilated convolution is used to fully excavate important spatial structure information of the SAR images to improve the performance of change detection. Finally, the experimental results on the three different types of the SAR datasets show that the proposed method can obtain a high detection accuracy and KC coefficient, which is better than various comparison methods.
Published:	2023-07-28
PDF HTML