High-resolution remote sensing image classification method guided by multi-level spatial context features

Abstract

The invention discloses a high-resolution remote sensing image classification method guided by multi-level spatial context features. According to the method, texture features of an object are extracted; geometric features and spatial context features are used as high-dimensional features of a deep learning framework; the classification result is put into a full-connection classifier to carry out classification training to obtain an object-oriented multi-feature fusion classifier, and finally, pixel-level spatial context guidance classification is performed on the classification result of the object level by using a conditional random field to further improve the classification precision. According to the high-resolution remote sensing image classification method based on multi-level spacecontext guidance, a plurality of features of objects are fused into object classification, a deep learning method exceeding general object classification can be realized, in addition, a conditional random field method guided by pixel-level space context is introduced, and high-precision high-resolution remote sensing image classification is realized.

Description

[0001] field of invention [0002] The invention belongs to the field of remote sensing image processing, relates to the field of computer deep learning, and in particular relates to an object-oriented high-resolution remote sensing image classification method based on deep learning. Specifically, it is a method of using deep learning network to extract multiple object features, construct a multi-feature fusion classifier, and improve the classification accuracy of high-resolution remote sensing images. Background technique [0003] The use of satellite remote sensing images for high-precision classification and interpretation has always been an important content with both application value and challenges in the field of remote sensing image processing. It has important scientific guiding significance for urban planning and disaster relief. With the rapid development of high-precision sensor technology and UAV and aerial photography technology in recent years, the spatial reso...

AI Technical Summary

Problems solved by technology

This method of feature extraction by taking a fixed-size patch in the object ignores the contour information of the object, and does not consider enough spatial context features.

Images