Generation And Classified Evaluation Of False Colored Image Composition Feature With High Dimensional Remote Sensing Data

Visualization of remote sensing data is the basic work of remote sensing applications such as visual interpretation.Color images can contain up to 3 features.And,the current dimension of remote sensing data(that is the number of bands)is often greater than three characteristics;The higher spectral resolution,data access to the more similar between adjacent band,the correlation of between bands,the greater the information redundancy is higher,the computing and the processing are more complex.Therefore,the reduction of the high-dimensional remote sensing data dimension is particularly important.This paper aim to high-dimensional remote sensing data visualization issues;through the exploration of different supervised feature generation methods to proposed a supervised visualization method for remote sensing data.This paper discusses the importance of the combination of visual and visual interpretation of high-dimensional remote sensing data.In this paper,a supervised principal component transform(Fisher)is proposed to reconstruct the parameters of regression parameters,on the basis of the principal component transform method or Fisher discriminant method;the least squares method is used to estimate the Fisher's coefficient.The first three components of the training samples are selected as feature information,and three new bands are given red,green and blue respectively,which are combined into a false color remote sensing image to realize the supervised visualization of high dimensional remote sensing data.At the same time,the first 3 features are selected for classification,and the method was evaluated by classification accuracy.The visualization results are compared with the main component transformations of remote sensing images and the false color images after Fisher discriminant transformation,and the new method of image has good visual interpreting results.In the case of supervised classification,the data used in this paper are constitute by training samples and test samples.Selected through the principal component transformation,the training samples were reconstructed and the classification model was established by Fisher discriminant method.The samples were classified and compared with the simple Fisher discriminant classification.The results show:reconstruction of regression parameters using principal component transform to perform supervised classification,its total accuracy is 76.55%,kappa coefficient is 0.7163.When using Fisher's regression parameter reconstruction method to carry out supervised classification,its total classification accuracy is76.01%,kappa coefficient is 0.7100,the classification result of principal component transform method is the same as that of the Fisher discriminant method,and the classification accuracy is 75.74% and kappa coefficient is 0.7068.The result of parameter reconstruction was improved.Using the first three features to verify this visualization,principal component transformation,principal component transformation regression parameter to refactor,Fisher discriminant and Fisher discriminant regression parameters of the classification accuracy were 64.69%,71.43%,69.00% and 64.69% respectively,the classification accuracyof high and low can affect visual result.