Coupled Convolutional Neural Network With Adaptive Response Function Learning for Unsupervised Hyperspectral Super Resolution

Due to the limitations of hyperspectral imaging systems, imagery (HSI) often suffers from poor spatial resolution, thus hampering many applications imagery. Hyperspectral super-resolution refers fusing HSI and MSI generate an image with both high spectral resolutions. Recently, several new methods have been proposed solve this fusion problem, most these assume that prior information Point Spread Function (PSF) Spectral Response (SRF) are known. However, in practice, is limited or unavailable. In work, unsupervised deep learning-based method - HyCoNet can problems HSI-MSI without PSF SRF proposed. consists three coupled autoencoder nets which unmixed into endmembers abundances based on linear unmixing model. Two special convolutional layers designed act as a bridge coordinates nets, parameters learned adaptively two convolution during training process. Furthermore, driven by joint loss function, straightforward easily implemented end-to-end manner. The experiments performed study demonstrate performs well produces robust results for different datasets arbitrary PSFs SRFs.