Research On Feature Extraction And Classification Algorithm Of Hyperspectral Image

Hyperspectral images are high-dimensional data with rich spatial and spectral information.They have been widely used in certain fields such as precision agriculture,geological exploration,food safety,and environmental monitoring.However,due to the fact that the hyperspectral image data itself has too much redundant information,and the similarity between the spectrum and the spectrum is very high,and the feature dimension of the data is too high,so this will make the hyperspectral image have a great Difficulties and challenges.This article is mainly related to the feature extraction and classification methods of hyperspectral images.The main research contents and results of this article are as follows:(1)Firstly,the domestic and international status of hyperspectral image feature extraction and classification is introduced;secondly,the theoretical basis of feature extraction and classification methods of hyperspectral images that are commonly used in recent years is introduced,and the main components are mainly introduced in the feature extraction method The principles of analysis and linear discriminant analysis mainly introduce the principles of support vector machine,neural network,K-nearest neighbor algorithm and fuzzy clustering in the classification method;finally,the common evaluation indexes of hyperspectral image classification are introduced.(2)In the current research on feature extraction of hyperspectral images,most existing feature extraction methods assume that the label information in the data set is correct,but this assumption may not always be correct in practical applications,which may Will affect the performance of feature extraction methods and hyperspectral image classification.In order to solve this problem in hyperspectral image classification,we propose a feature extraction method based on Regularized Fuzzy Discriminant Analysis(RFDA),which can effectively use the space of hyperspectral images with noise labels And spectral information.As a result,the proposed method not only effectively corrects the wrong label samples,but also retains the neighbor relationship between pixels in the spatial domain and the basic structure between the samples in the spectral domain,which is conducive to the classification of hyperspectral images.Simulation experiments were performed on three public hyperspectral data sets,and then the classification results were analyzed using overall accuracy and Kappa coefficients.The experimental results show that the proposed RFDA method is superior to several other feature extraction methods in terms of classification accuracy.(3)In the current research on the classification of hyperspectral images,based on traditional fuzzy clustering,in the classification of hyperspectral images,the samples of class boundaries are relatively close due to the membership values of the two categories.The degree of membership determines the type of membership.But there will also be deviations that lead to the final effect not reaching the desired effect.However,intuitionistic fuzzy sets(IFSs)also show advantages in describing fuzzy and uncertain data when considering both membership and non-membership.However,when intuitionistic fuzzy C-means clustering involves multimodal and unbalanced features,it usually does not consider the importance of individual attributes and the structure of the data.Multi-kernel intuitionistic fuzzy C-means(MKFCM)using IFSs to handle uncertainty can achieve higher clustering accuracy.Therefore,by introducing multi-core mapping,this paper proposes a hyperspectral image classification method based on intuitionistic multi-core fuzzy clustering(MKIFCM).Simulation experiments were conducted on two public hyperspectral data sets,and then the classification results were analyzed using overall accuracy.From the experimental results,it can be seen that compared with the traditional clustering algorithm and kernel function-based clustering algorithm,the proposed MKIFCM method has a significant effect on the classification of hyperspectral images.