English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The technology of hyperspectral imaging (HSI) records the visual informationupon long-range-distributed spectral wavelengths. A representativehyperspectral image acquisition procedure conducts a 3D-to-2D encoding by thecoded aperture snapshot spectral imager (CASSI) and requires a software decoderfor the 3D signal reconstruction. By observing this physical encodingprocedure, two major challenges stand in the way of a high-fidelityreconstruction. (i) To obtain 2D measurements, CASSI dislocates multiplechannels by disperser-titling and squeezes them onto the same spatial region,yielding an entangled data loss. (ii) The physical coded aperture leads to amasked data loss by selectively blocking the pixel-wise light exposure. Totackle these challenges, we propose a spatial-spectral (S^2-) Transformernetwork with a mask-aware learning strategy. First, we simultaneously leveragespatial and spectral attention modeling to disentangle the blended informationin the 2D measurement along both two dimensions. A series of Transformerstructures are systematically designed to fully investigate the spatial andspectral informative properties of the hyperspectral data. Second, the maskedpixels will induce higher prediction difficulty and should be treateddifferently from unmasked ones. Thereby, we adaptively prioritize the losspenalty attributing to the mask structure by inferring the pixel-wisereconstruction difficulty upon the mask-encoded prediction. We theoreticallydiscusses the distinct convergence tendencies between masked/unmasked regionsof the proposed learning strategy. Extensive experiments demonstrates that theproposed method achieves superior reconstruction performance. Additionally, weempirically elaborate the behaviour of spatial and spectral attentions underthe proposed architecture, and comprehensively examine the impact of themask-aware learning.

S^2-Transformer for Mask-Aware Hyperspectral Image Reconstruction