Representation of genetic individuals for unmixing multispectral data

The traditional classification algorithms for multispectral images assign only one class to each pixel. However, such pixels are actually a mixture of the spectral reflectance values of several different types of ground, of which the various abundances characterize the final shape of the observed spectrum. Within the framework of supervised learning, a representative solution was defined to solve this kind of problem using a genetic algorithm. This paper introduces a representation of the selected and various associated genetic operators (fitness, crossover, mutation) used in remote sensing image classification, as well, it describes a comparison of various representation using two more algorithms on three data sets.