HyperReconNet: Joint Coded Aperture Optimization and Image Reconstruction for Compressive Hyperspectral Imaging

Lizhi Wang, Tao Zhang, Ying Fu*, Hua Huang

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

129 Citations (Scopus)

Abstract

Coded aperture snapshot spectral imaging (CASSI) system encodes the 3D hyperspectral image (HSI) within a single 2D compressive image and then reconstructs the underlying HSI by employing an inverse optimization algorithm, which equips with the distinct advantage of snapshot but usually results in low reconstruction accuracy. To improve the accuracy, existing methods attempt to design either alternative coded apertures or advanced reconstruction methods, but cannot connect these two aspects via a unified framework, which limits the accuracy improvement. In this paper, we propose a convolution neural network-based end-to-end method to boost the accuracy by jointly optimizing the coded aperture and the reconstruction method. On the one hand, based on the nature of CASSI forward model, we design a repeated pattern for the coded aperture, whose entities are learned by acting as the network weights. On the other hand, we conduct the reconstruction through simultaneously exploiting intrinsic properties within HSI - the extensive correlations across the spatial and spectral dimensions. By leveraging the power of deep learning, the coded aperture design and the image reconstruction are connected and optimized via a unified framework. Experimental results show that our method outperforms the state-of-the-art methods under both comprehensive quantitative metrics and perceptive quality.

Original languageEnglish
Article number8552450
Pages (from-to)2257-2270
Number of pages14
JournalIEEE Transactions on Image Processing
Volume28
Issue number5
DOIs
Publication statusPublished - 2019

Keywords

  • Compressive hyperspectral imaging
  • coded aperture optimization
  • convolution neural network
  • hyperspectral image greconstruction

Fingerprint

Dive into the research topics of 'HyperReconNet: Joint Coded Aperture Optimization and Image Reconstruction for Compressive Hyperspectral Imaging'. Together they form a unique fingerprint.

Cite this