Find in Library

Recently, deep convolutional neural network (CNN) hyperspectral change detection methods have achieved significant improvement. However, most CNN hyperspectral change detection methods do not make full use of spectral–spatial feature information. In this article, we propose a novel multidirection and multiscale spectral–spatial residual network for hyperspectral multiclass change detection. Specifically, a multiscale structure and a multidirection mechanism are introduced to investigate feature variation of hyperspectral images and improve the accuracy of hyperspectral change detection. Experiments on multiple hyperspectral datasets show that the proposed method achieves improved performance in comparison with other advanced hyperspectral multiclass change detection methods.