Difference Curvature Multidimensional Network for Hyperspectral Image Super-Resolution

In recent years, convolutional-neural-network-based methods have been introduced to the field of hyperspectral image super-resolution following their great success in RGB super-resolution. However, images appear different from that they high dimensionality, implying a redundancy high-dimensional space. Existing approaches struggle learning spectral correlation and spatial priors, leading inferior performance. this paper, we present difference curvature multidimensional network for exploits help improve resolution. Specifically, introduce enhanced convolution (MEC) unit into learn through self-attention mechanism. Meanwhile, it reduces dimension via bottleneck projection condense useful features reduce computations. To remove unrelated information space extract delicate texture image, design an additional branch (DCB), which works as edge indicator fully preserve eliminate unwanted noise. Experiments on three publicly available datasets demonstrate proposed method can recover sharper with minimal distortion compared state-of-the-art methods. PSNR/SAM is 0.3–0.5 dB/0.2–0.4 better than second best