Compressive sensing approach to satellite hyperspectral image compression

Abstract

Hyperspectral image (HSI) processing plays a very important role in satellite imaging applications. Sophisticated sensors on-board the satellite generates huge hyperspectral datasets since they capture a scene across different wavelength regions in the electromagnetic spectrum. The memory available for storage and bandwidth available to transmit data to the ground station is limited in case of satellites. As a result, compression of hyperspectral satellite images is very much necessary. The research work proposes a new algorithm called SHSIR (sparsification of hyperspectral image and reconstruction) for the compression and reconstruction of HSI acquired using compressive sensing (CS) approach. The proposed algorithm is based on the linear mixing model assumption for hyperspectral images. Compressive sensing measurements are generated by using measurement matrices containing Gaussian i.i.d. entries. HSI is reconstructed using Bregman iterations, which advance the reconstruction accuracy as well as the noise robustness. The proposed algorithm is compared with state-of-the-art compressive sensing approaches for HSI compression and the proposed algorithm performs better than existing techniques both in terms of reconstruction accuracy as well as noise robustness.