Adaptive non-dominated sorting genetic algorithms for wavelength selection of molecular hyperspectral images

Hyperspectral data cube usually includes hundreds of high-correlated single band images. It is necessary to reduce the dimensionality of hyperspectral images to facilitate the studies and analysis. This paper presents an adaptive non-dominated sorting genetic algorithm to process the wavelength combination of molecular hyperspectral images. In this algorithm, dynamic reproduction probabilities are employed to regulate the selection pressure. To evaluate the performance of this new algorithm on the combination optimization, the simulation results are compared with those of a non-dominated sorting genetic algorithm without adaptation and of a single-objective genetic algorithm having the same adaptive mechanism. The comparison revealed its better performance in the wavelength selection of molecular hyperspectral data of rat retinal sections.