A semi-supervised multi-view feature selection method for remote sensing images with label learning

Abstract

The invention discloses a semi-supervised multi-view feature selection method for high-resolution remote sensing images with label learning, belonging to the technical field of semi-supervised feature selection for high-resolution remote sensing images. The present invention is to solve the problem that in the existing high-resolution remote sensing image feature selection methods, when only a small number of objects are marked, the supervised and unsupervised methods cannot achieve good enough performance and the views are not available in the high-resolution image features. question. It includes the following steps: 1. Collect the original image feature set and generate the feature group; 2. Calculate and obtain the weight coefficient of all feature vectors in all feature groups through the class probability matrix yu and the diagonal matrix F containing the exclusive group information The original feature weight coefficient vector β; 3. Iteratively calculate the feature weight coefficient vector β, select a preset number of weight coefficients, and use all the feature vectors corresponding to the selected weight coefficients as the selected feature set. The invention is used for feature selection of high-resolution remote sensing images.

Description

technical field [0001] The invention relates to a semi-supervised multi-view feature selection method for high-resolution remote sensing images with label learning, and belongs to the technical field of semi-supervised feature selection for high-resolution remote sensing images. Background technique [0002] High-resolution image VHRs can capture small or narrow objects, so they can be used to develop a corresponding infrastructure for continuous monitoring and map updating. This application often uses the object-based image analysis method OBIA to exploit the spatial relationship between pixels to cope with the salt and pepper effect when using a pixel-wise classifier. OBIA can extract a large number of features. Since not all features are beneficial for classification and classification performance may degrade as the number of features increases, feature selection becomes an important issue to address this issue. [0003] Spatial, texture and shape features can be extrac...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to solve the problem that in the existing high-resolution remote sensing image feature selection methods, when only a small number of objects are marked, supervised and unsupervised methods cannot achieve good enough performance and views are not available in the features of high-resolution images. problem, providing a semi-supervised multi-view feature selection method for high-scoring remote sensing images with label learning

Images