Research On The Application Of MRCD Algorithm In The Classification Of High-dimensional Data

With the continuous development of smart cities,the information age background of big data has provided strong support for the planning of modern cities.This is precisely because the high-resolution remote sensing data provides a large amount of high-dimensional data information about the spatial,spectral,and texture features of urban land.In the collection of remote sensing data,a large number of outliers appear due to reasons,so it brings certain challenges to the classification of high-dimensional data with obvious outliers.The traditional classification method is very sensitive to outliers,which will affect the results of the model estimation,and may even lead to wrong conclusions,causing major economic losses to the city.This paper will propose a class of robust quadratic discriminant analysis algorithms,which can effectively process high-dimensional categorical data,in view of the limitations of classic multivariate statistical analysis methods in high-dimensional data analysis.It is supported by numerical simulation and used to solve the problem of urban land classification in high-resolution remote sensing images.Regarding the effectiveness of the Q-MRCD method,The main work of this paper is as follows:The simulation results show the good properties of the robust estimation of MRCD.The Q-MRCD classification method combined with Quadratic Discriminant Analysis achieves a high accuracy classification effect in low and high dimensional sample data with 10% outliers.In high-dimensional data,the classification accuracy rate is higher than 94%,and it has a wider scope than MCD robust estimation.In the empirical analysis,the Q-MRCD classification method is used to classify and analyze the high-dimensional small sample data of urban land with high-resolution remote sensing images.The accuracy rate obtained by 10-fold cross-validation is as high as 87%.Compared with Machine Learning,the Q-MRCD classification method has better classification effect as a traditional statistical method and has good model interpretation and high stability.