Abstract:Hyperspectral imagery contains both spatial structure information and abundant spectral features of imaged objects. However, due to sensor limitations, abundant spectral information always comes at the sacrifice of low spatial resolution, which brings about difficulties with object analysis and identification. The super-resolution (SR) of HSIs, restored by the traditional interpolation algorithms or the network models trained with the mean-square-error-based loss function, tends to produce over-smoothed images. In this paper, we propose a novel Hyperspectral imagery Spatial Super-Resolution algorithm based on a Generative Adversarial Network (HSSRGAN). The generator network in HSSRGAN consists of two interacting part, i.e., a spatial feature enhanced network (SFEN) and a spectral refined network (SRN), while the discriminator network is employed to predict the probability that the authentic HR image is comparatively more similar than the forged generated image. Concretely, SFEN with the special dense residual blocks is designed to fully extract and enhance more deep hierarchical spatial features of hyperspectral imagery, while SRN is constructed to capture spectral interrelationships and refine the spatial context information so as to increase spatial resolution and alleviate spectral distortion. Moreover, SFEN and SRN are trained by the least-absolute-deviation based loss function to investigate spatial context and the spectral-angle-mapper based loss function to refine spectral information. We validate two versions of our proposed algorithm, 3D-HSSRGAN and 2D-HSSRGAN, on Pavia Centre dataset and Cuprite dataset. Experimental results show that the presented approach is superior to several existing state-of-the-art works.

Deep Amended Gradient Descent for Efficient Spectral Reconstruction from Single RGB Images

U-Net-Based Multispectral Image Generation from an RGB Image

Hyperspectral Image Reconstruction Based on Improved Residual Dense Network

Hyperspectral Image Reconstruction from RGB Input Through Highlighting Intrinsic Properties

Hyperspectral Unmixing Via Projected Mini-Batch Gradient Descent

Hyperspectral Reconstruction From RGB Images via Physically Guided Graph Deep Prior Learning

Hyperspectral Reconstruction Using Auxiliary Rgb Learning From A Snapshot Image

Deep Parameterized Neural Networks for Hyperspectral Image Denoising

Unmixing of Large-Scale Hyperspectral Data Based on Projected Mini-Batch Gradient Descent.

Learning Hyperspectral Images from RGB Images Via a Coarse-to-fine CNN.

Reconstructing hyperspectral images from rgb inputs based on intrinsic image decomposition

Semantic-embedded Unsupervised Spectral Reconstruction from Single RGB Images in the Wild

Hyperspectral Imagery Spatial Super-Resolution Using Generative Adversarial Network.

Deep-learning-based Hyperspectral Recovery from a Single RGB Image.

Cluster-based deep convolutional networks for spectral reconstruction from RGB images

HDNet: High-resolution Dual-domain Learning for Spectral Compressive Imaging

HSCNN+: Advanced CNN-Based Hyperspectral Recovery from RGB Images

RAW2HSI: Learning-Based Hyperspectral Image Reconstruction from Low-Resolution Noisy Raw-RGB

Enhanced Deep Unrolling Networks for Snapshot Compressive Hyperspectral Imaging

Double Ghost Convolution Attention Mechanism Network: A Framework for Hyperspectral Reconstruction of a Single RGB Image

Cross Spectral Image Reconstruction Using a Deep Guided Neural Network