Semi-supervised hyperspectral remote sensing image classification annotation method

Abstract

The invention discloses a semi-supervised hyperspectral remote sensing image classification annotation method for the problem in hyperspectral remote sensing image classification annotation. The method comprises the steps of obtaining a small amount of annotated training samples (with category annotation true values) and a large amount of non-annotated training samples and test samples through manual sample collection; constructing a correlation potential function of a conditional random field through a transductive support vector machine; constructing an interactive potential function of the conditional random field through an improved Potts model; training a conditional random field model through a genetic algorithm; and performing inference on the test samples through the trained conditional random field model to obtain a classification annotation result of the test samples. Compared with a classification annotation result obtained by separately using a conditional random field algorithm or the transductive support vector machine, a hyperspectral remote sensing image classification annotation result obtained by the method has the advantages that a large amount of isolated noise points are removed, so that relatively good regional continuity is achieved and the precision is relatively high.

Description

technical field [0001] The invention belongs to the technical field of image processing and pattern recognition, and in particular relates to a semi-supervised method for classifying and marking hyperspectral remote sensing images. Background technique [0002] With the rapid development of remote sensing imaging sensors, hyperspectral remote sensing imaging has gradually become an important means for human beings to explore the earth. The effective acquisition of a large number of hyperspectral remote sensing images demonstrates the ability of human beings to obtain remote sensing data, provides complete information resources for the development of scientific research, and also puts forward higher requirements for hyperspectral remote sensing image processing technology. [0003] In the field of remote sensing, hyperspectral image classification and labeling is one of the most basic problems in remote sensing image processing technology, and it is also the basis of image an...

AI Technical Summary

Problems solved by technology

[0005] (1) Purpose of the invention: In view of this, the present invention expects to provide a semi-supervised hyperspectral remote sensing image classification and labeling method to solve technical problems such as hyperspectral remote sensing image classification and labeling under the condition of a small number of labeled training samples

Images