Dimension Reduction And Accuracy Evaluation Of Hyperspectral Remote Sensing Image Oriented To Extraction Of Feature Information

The existing remote sensing observation technology provides hyperspectral remote sensing image data for various fields.However,how to obtain effective information quickly and effectively in hyperspectral remote sensing image data has become one of the crucial issues restricting the application of hyperspectral remote sensing image data.In recent years,the dimensionality reduction method for “hyperspectral remote sensing image” has received widespread attention.However,due to the multi-band,high redundancy and strong correlation between bands,the existing dimensionality reduction methods have the phenomenon that the computational efficiency and the accuracy of the classification result after dimension reduction cannot be considered at the same time.At the same time,the accuracy evaluation of surface information extraction for hyperspectral remote sensing images is also a key issue affecting its application.Therefore,the aims of this paper are to reduce the dimension of hyperspectral remote sensing images and evaluate the accuracy of the extraction results of ground object information after the reduction of dimension,and the research includesthe dimensionality reduction model for hyperspectral images analysis based on rough set theory and the accuracy inspection for remote sensing images based on multilevel non-uniform spatial sampling design.(1)In view of the multi-bands and high redundancy of hyperspectral remote sensing images,a dimension reduction method for hyperspectral remote sensing images based on rough set theory is proposed.In the dimension reduction study of hyperspectral remote sensing images,a dimension reduction method of rough set based on feature selection is proposed.The main idea of this method is: firstly,using the discernibility matrix of rough set theory to reduce the dimension of hyperspectral remote sensing image,and then retaining the important band to reduce the information redundancy of hyperspectral remote sensing image.Secondly,using the information entropy to sort the information of the remaining important bands,and the band combination is selected according to the precision requirements of the actual feature information extraction,so as to achieve the purpose of dimensionality reduction.(2)In view of the large area coverage characteristics of remote sensing images,an accuracy inspection method for remote sensing image object information based on multi-level non-uniform spatial sampling is proposed.In the sampling research of remote sensing images,an accuracy evaluation method based on multi-level non-uniform spatial sampling using fragmentation index is proposed.The main idea of this method is: firstly,by calculating the fragmentation index of the remote sensing image classifications,the remote sensing image is divided step by step from low to high,and a scheme of sample point layout from high to low is designed according to the classification level.It ensures both the high sampling rate for accuracy inspection in regions with complex,heterogeneous and high fragmentation,and the representativeness of sample points.Finally,the hyperspectral remote sensing image of Urban dataset is taken as experimental data to make an empirical analysis of the whole set of methods in this paper.And the results show that the proposed method has obvious advantages over the traditional dimensionality reduction method in time efficiency.At the same time,the accuracy inspection method based on multi-level non-uniform sampling guarantees both the balance of the sample points among different types of features,and the spatial representation of the sample points.