Multispectral Image Processing With Genetic Methods

Remote sensing technology is one of the most important methods for human beings to make an investigation on resource and environment of the earth and other celestial bodies. It takes more and more significant effect on some applications, such as managing soil resource, inspecting and surveying of environment and so on. Realizing the significance of multispectral data classification, scholars always strive for designing the quick and high-accuracy classification algorithms.Remote sensing images contain a lot of mixed pixels, and mixed pixel unmixing and classification is one of the key technologies of remote sensing information system. Making the investigation on it has important social and economic benefits. In this thesis, a new method based on Genetic Algorithm is proposed.By researching the principles of the mixed pixel unmixing algorithm, this paper proposes a new GA-based method for mixed pixel unmixing and classification problems, which uses the GA evolution models to fit (approximate) the hyperplane of the unmixing results. The validity and feasibility of the algorithm are demonstrated by remote multispectral images. Experimental results show that the approach is able to get the almost the same category proportion with the method of Full Constrained Least-Squares (correlation coefficients are above 0.99), and had more flexibility than the method of least squares.