Different Feature Extraction Methods Are Used To Compare The Classification Results Of Hyperspectral Remote Sensing Pixels

With the rapid development of hyperspectral remote sensing technology im-ages generated by hyperspectral remote sensing technology have been widely used in many fields in recent years.Hyperspectral data provides us with many conveniences in our research.As hyperspectral remote sensing data has many bands,it is necessary to preprocess the spectral information before using the data.The processed spectral information can be analyzed to obtain the information we need.Hyperspectral data are also used in a wide range of fields,such as atmospheric pollution monitoring,Marine water quality monitoring and treatment,chemical composition analysis of a-gricultural crops,identification of the authenticity of military targets,and prevention of natural disasters.However,due to the characteristics of multiple spectral bands of hyperspectral,data analysis will also bring disaster problems of data dimension,resulting in very high computing cost.So the band selection is particularly neces-sary.In this paper,Indian-pines and Pavia_U of hyperspectral remote sensing data were used to extract features by comparing all features and principal components of PCA(Principle Component Analysis),and features after dimension reduction by L-DA(Linear Discriminant Analysis).ACO(Ant Colony Optimization)selects features and uses GA(Genetic Algorithm)to optimize kernel function parameters and adjust parameters of SVM(Support Vector Machine)to compare the classification accura-cy of SVM classifiers trained by different feature extraction/selection methods.The work of this article includes:(1)The characteristics and preprocessing of hyperspectral remote sensing data,the main application direction of hyperspectral remote sensing.(2)GA principle and parameter optimization of classifier in this paper by GA.(3)The principle of different feature selection method and its application in this paper and the classification effect of classifier trained by this feature extraction method.