Spatial-spectral hyperspectral image compressive sensing

Over the last decade, the number of missions including hyperspectral cameras of increasing resolution has grown considerably. Traditional compression techniques have been proposed as an efficient way to store and transmit the ever increasing amount of hyperspectral data and to cope with the limited transmission bandwidth between the onboard systems and the ground stations. On the other hand, compressive sensing techniques try to deal with the same issue from a different perspective: the signal is compressed while the acquisition takes place, and the decompression is carried out by solving an optimization problem. Due to scarce onboard resources, the compressing sensing paradigm in hyperspectral systems is attractive, namely because the main computational burden to recover the original data is carried out in the ground stations, where more computational resources are available. In this paper, we develop a new technique for compressive sensing of hyperspectral images, where the measurement process takes place both in the spatial and the spectral domains by efficiently exploiting the high correlation of hyperspectral images. The proposed technique was tested using synthetic and semi-real images yielding competitive results.