Parallel Spectral–Spatial Attention Network with Feature Redistribution Loss for Hyperspectral Change Detection

Change detection methods using hyperspectral remote sensing can precisely identify differences of the same area at different observing times. However, due to massive spectral bands, current change are vulnerable unrelatedspectral and spatial information in images with stagewise calculation attention maps. Besides, arrange hidden features a random distribution form, which cannot express class-oriented discrimination advance. Moreover, existent deep have not fully considered hierarchical features’ reuse fusion encoder–decoder framework. To better handle mentioned problems, parallel spectral–spatial network feature redistribution loss (TFR-PS2ANet) is proposed. The contributions this article summarized as follows: (1) module (PS2A) introduced enhance relevant suppress irrelevant maps extracted from original image patches; (2) function (FRL) construct distribution, organizes advance improves discriminative abilities; (3) two-branch framework developed optimize transfer fusion; Extensive experiments were carried out on several real datasets. results show that proposed PS2A significant effectively FRL distribution. method outperforms most methods.