Research On Hyperspectral Image Feature Extraction And Ground Object Recognition Methods

Hyperspectral remote sensing image contains a lot of spectral feature information,which brings new opportunities to the problem of ground object recognition.However,there are difficulties in the feature extraction of hyperspectral images,such as data redundancy,insufficient spatial resolution,lack of labeled samples,etc.,which can easily lead to the problem of "dimensional disaster" if it is classified directly,which limits the application and development of hyperspectral remote sensing image classification.Aiming at the above problems,this paper focuses on the core issues such as global linear feature extraction,manifold learning feature extraction and ground object recognition,focusing on the features extraction of hyperspectral and high-dimensional data and hyperspectral ground object recognition based on extreme learning machine.The main research contents are as follows:(1)The global linear feature extraction method is used to reduce the dimension of hyperspectral images to solve the problem that the accuracy of ground object recognition is reduced due to the redundancy of data information.In order to solve the problem of hyperspectral image data redundancy caused by large amount of data and many bands,the global linear feature extraction method was adopted to reduce the dimensionality,and the principal component information of high-dimensional data was extracted by using the category information of hyperspectral image.Based on the hyperspectral image data set after dimensional reduction,the Extreme Learing Machine(ELM)ground object recognition model is established.Compared with the ground object recognition model established based on K Nearest Neighbor(KNN)and Support Vector Machine(SVM),the proposed model can improve the ground object recognition accuracy of hyperspectral images.(2)Manifold learning method is used to extract features of hyperspectral images to solve the problem of non-linear feature extraction of hyperspectral images.The local linear feature extraction method is difficult to extract the nonlinear structure features in the hyperspectral data,therefore the paper uses the Manifold Learning Locality Preserving Projection(LPP)method to extract the features of hyperspectral images.By constructing a projection matrix of high-dimensional spatial data in low-dimensional space,not only the local neighborhood structure of the sample in the space is preserved,but also the local manifold structure of the data can be represented.The experimental results show that the method of manifold learning can improve the accuracy of feature recognition model in hyperspectral images.(3)k-means clustering algorithm is used to solve the structural optimization problem of ELM ground object recognition model and improve the recognition accuracy of the model.The ELM ground object recognition model is affected by random parameters,which leads to the reduction of model recognition accuracy.The paper uses the k-means clustering algorithm to optimize the structure of the extreme learning machine,and the hidden layer activation function and other model parameters are determined from the clustering results.The k value of the k-means clustering algorithm is determined by the method of information entropy.The experimental results show that the Structure Optimized Extreme Learning Machine(SO-ELM)ground object recognition model can improve the ground object recognition accuracy.Taking hyperspectral images as the research object,this paper focuses on the research of global linear feature extraction,manifold learning feature extraction and structure optimization of extreme learning machine,etc.The established hyperspectral image ground object recognition model improves the accuracy of ground object recognition of hyperspectral images and verifies the effectiveness of the proposed algorithm.