CSSNET: A Learning Algorithm for the Segmentation of Compressed Hyperspectral Images
Abstract
The paper presents a semantic segmentation method which is directly applicable to compressed hyperspectral images acquired with a dual-disperser CASSI instrument. It introduses an algorithm based on a shallow neural network that exploits the spectral filtering performed by the optical system and the compressed hyperspectral images measured by the detector. Encouraging results that exploit 50 to 100 less data than the whole hyperspectral datacube on PaviaU and IndianPines datasets are presented.
Origin : Files produced by the author(s)