Spectral angle and Euclidean distance based remote-sensing image classification method

Abstract

The invention is applicable to the field of remote-sensing image classification and provides a spectral angle and Euclidean distance based remote-sensing image classification method. The spectral angle and Euclidean distance based remote-sensing image classification method comprises the steps of preprocessing remote-sensing images to filter out noise; screening effective information for classification; segmenting the remote-sensing images into multiple homogenous image map spots serving as minimum research units; calculating mean values and variances of training samples at all wave bands; calculating mean values and variances of testing samples at all wave bands; further calculating Euclidean distances and spectral angles; determining the comprehensive similarity as the sum of weights of the spectral angles and the Euclidean distances and determining weights; calculating the comprehensive similarity of classification objects and surface features to enable the type of the surface features with minimum comprehensive similarity to serve as the final type of the classification objects. The spectral angle and Euclidean distance based remote-sensing image classification method integrates the advantages of two classifiers, achieves complementation of different classification methods, determines optimal weight through verification at minimum intervals, effectively improves classification accuracy, ensures classification efficiency, achieves algorithm automation and is high in classification efficiency.

Description

technical field [0001] The invention belongs to the technical field of computer remote sensing image automatic classification, in particular to a remote sensing image classification method based on spectral angle and Euclidean distance. Background technique [0002] Remote sensing image classification is the process of dividing pixels in remote sensing images into different object categories. The classification basis mainly includes the spectral characteristics of ground objects, the shape characteristics of land objects, and the characteristics of spatial relationship. At present, most studies are still based on the spectral characteristics of ground objects, combined with the results of visual interpretation. Using computer automatic classification algorithm is the main method of remote sensing image classification at present. Computer automatic classification algorithms mainly include unsupervised classification and supervised classification. Unsupervised classification...

AI Technical Summary

Problems solved by technology

[0003] Each classification method has its own defects. In order to improve the stability and reliability of the classification results, the existing technology proposes the idea of ​​multi-classifier fusion, which improves the classification accuracy to a certain extent, but there are still some problems, such as simple Blindly fusing different classifiers, ignoring their complementarity, resulting in extremely limited improvement in classification accuracy and affecting classification efficiency, so the selection of classifiers and fusion methods is particularly important

Images