High resolution remote sensing image oriented large scale semi-supervised feature selection method

Abstract

A high resolution remote sensing image oriented large scale semi-supervised feature selection method belongs to the technical field of semi-supervised feature selection. The present invention aims to solve the problems that in an existing high resolution remote sensing image supervision feature selection method, a large amount of training data needs to be marked and when the number of unmarked objects is much greater than marked data, rationality of selected features is influenced. The method comprises the steps of firstly, collecting remote sensing image data and obtaining normalized data x after performing processing on the remote sensing image data; then building a metric function of a probability distribution matrix {y<jk>} based on a loss function and unmarked samples; sequentially performing loop optimization on three parameters of the metric function and obtaining corresponding metric values of corresponding features; and according to the metric values, ranking the features, and obtaining a feature subset of the remote sensing image data, wherein the feature subset is used as data selected and acquired by the large scale semi-supervised feature selection method. The high resolution remote sensing image oriented large scale semi-supervised feature selection method, disclosed by the present invention, is used for feature selection of a remote sensing image.

Description

technical field [0001] The invention relates to a semi-supervised feature selection method, in particular to a large-scale semi-supervised feature selection method for high-resolution remote sensing images. Background technique [0002] By providing accurate and extensive land use and land cover information, high-resolution (VHR) remote sensing images play a great role in real life. The application of high-resolution remote sensing images in real life often relies on object-based image analysis (OBIA). OBIA requires various target features, including spectral, structural and shape features of the target, and too many underlying features will degrade the performance of OBIA. This contradiction can be alleviated by feature selection methods. By effectively selecting a small number of original features with higher discriminative power, feature selection methods have a direct and significant impact on the acceleration of data mining algorithms, the improvement of performance, ...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to solve the problem that in the existing high-resolution remote sensing image supervision feature selection method, a large amount of training data is required to be marked, and when the number of unmarked objects is far greater than the marked data, the problem of affecting the rationality of the selected features , providing a large-scale semi-supervised feature selection method for high-resolution remote sensing images

Images