SSCU-Net: Spatial–Spectral Collaborative Unmixing Network for Hyperspectral Images

Linear spectral unmixing is an essential technique in hyperspectral image processing and interpretation. In recent years, deep learning-based approaches have shown great promise unmixing, particular, unsupervised methods based on autoencoder networks are a trend. The model, which automatically learns low-dimensional representations (abundances) reconstructs data with their corresponding bases (endmembers), has achieved superior performance unmixing. this article, we explore the effective utilization of spatial information autoencoder-based networks. Important findings use framework discussed. Inspired by these findings, propose spatial-spectral collaborative network, called SSCU-Net, network end-to-end manner to more effectively improve performance. SSCU-Net two-stream shares alternating architecture, where two efficiently trained way for estimation endmembers abundances. Meanwhile, new introducing superpixel segmentation method abundance information, greatly facilitates employment improves accuracy network. Moreover, extensive ablation studies carried out investigate gain SSCU-Net. Experimental results both synthetic real sets illustrate effectiveness competitiveness proposed compared several state-of-the-art methods.