DefinePK

Abstract

Classification consists of categorizing image pixels, specifically, in a spectral image (SI) it is used to determine environmental pollution agents, to stable land use, and to monitor crops. Due to the high dimensionality of the SI, classification is inefficient. However, compressive sampling (CS) has been established as a sampling protocol of SI allowing the reduction of data. Recently, CS classification is a promising research area, but it has only been tested on some specific systems. This paper proposes a general classification algorithm in compressive spectral imaging which uses singular value decomposition for estimating sparse dictionary of the data. From this dictionary, the algorithm performs a rough estimation which allows the classification of every spectral pixel in known classes by using discriminant analysis. The estimation is made by solving the inverse problem. Simulations with three state of the art compressive imagers show the outstanding performance of the proposed algorithm even in presence of noise.