Distributed Compressive Sensing of Hyperspectral Images Using Low Rank and Structure Similarity Property

摘要

An efficient method and system for distributed compressive sensing ofrnhyperspectral images is presented, which exploit the low rank and structure similarityrnproperty of hyperspectral imagery. In this paper, by integrating the respectiverncharacteristics of DSC and CS, a distributed compressive sensing framework isrnproposed to simultaneously capture and compress hyperspectral images. At thernencoder, every band image is measured independently, where almost all computationrnburdens can be shifted to the decoder, resulting in a very low-complexityrnencoder. It is simple to operate and easy to hardware implementation. At therndecoder, each band image is reconstructed by the method of total variation normrnminimize. During each band reconstruction, the low rand structure of band imagesrnand spectrum structure similarity are used to give birth to the new regularizers. Withrncombining the new regularizers and other regularizer, we can sufficiently exploit thernspatial correlation, spectral correlation and spectral structural redundancy inrnhyperspectral imagery. A numerical optimization algorithm is also proposed tornsolve the reconstruction model by augmented Lagrangian multiplier method.rnExperimental results show that this method can effectively improve the reconstructionrnquality of hyperspectral images.