Multispectral Data Classification Based On Genetic Programming

Multispectral data classification is one of the typical applications of pattern recognition and has so many applications in various fields. Realizing the significance of multispectral data classification, scholars always strive for designing the quick and high-accuracy classification algorithms. In this thesis, a new method based on Genetic Programming is proposed.To begin with, the theory of remote sensing data classification is introduced, the feature of multispectral data is analyzed. Secondly, the traditional classification methods are studied and realized, and the results are compared. Thirdly, in order to avoid the deficiency of the traditional classification methods, the classifier based on Genetic Programming is proposed. Genetic Programming(GP) can discover relationships among observed data and express them mathematically. However, the existed algorithms based on GP can not resolve the multi-class classification issues accurately. In this thesis, the theory of Genetic Programming is introduced; the algorithm in realizing the GP classifier is shown; the application of GP in two-class classification problem is analyzed; and two novel methods, named as GP-ECC and GP-MECC, based on Genetic Programming combined with the binary error-correction codes are proposed and applied to the American AVIRIS and LANDSAT multispectral data classification problems. The results are satisfying compared to either the traditional methods such as the MLC and BP or the existed GP-based method, for the higher classification accuracy can be got by the new proposed methods. As a result, the work in this thesis can provide some new way for multispectal data classification.