Method for picking up and comparing spectral features in remote images

Abstract

A method for extracting and comparing spectral features of remote sensing images. After the de-correlation between multispectral image spectra is completed by principal component transformation, each cluster center vector of the image is determined, and each vector element of the image is divided into each cluster according to the "proximity principle". Go to the class represented by the class center, and then perform merging and splitting operations on the clustering results, so that the clustering results tend to be balanced, and the final result of feature extraction is obtained. Finally, after the feature extraction of the two images is completed, compare them according to the four features The standard scalarizes the comparison between vector arrays, so that the similarity and difference between images can be qualitatively and quantitatively reflected. The present invention adopts the improved iterative self-organizing data analysis technology ISODATA, which can greatly reduce the amount of calculation and the required parameters while ensuring the effect of the original algorithm, especially when there are many remote sensing image bands, and make the control of the parameters easier. And the feature extraction results can be used for image database retrieval.

Description

Technical field: [0001] The present invention relates to a method for extracting and comparing spectral features of remote sensing images, in particular to a method for extracting and comparing spectral features of remote sensing images based on an improved ISODATA (Iterative Self-Organizing Data Analysis Technique, Iterative Self-Organizing Data Analysis Technique A) algorithm , can be widely used in content-based image database retrieval, image classification and pattern recognition. Background technique: [0002] Content-based image retrieval is the technique of image retrieval by analyzing the image content (spectrum, texture, shape, etc.). Its characteristics are: 1. The retrieval process is interactive, and users can participate in the retrieval process; 2. It introduces the concept of feature database and knowledge assistance; 3. It pays more attention to the rapid query of information. [0003] The spectral feature extraction and feature comparison of remote sensing...

AI Technical Summary

Problems solved by technology

[0003] The spectral feature extraction and feature comparison of remote sensing images is an important part of remote sensing image database retrieval. The multispectral remote sensing image based on principal component transformation (K-L transformation, Karhunen Loeve transformation) and ISODATA clustering algorithm proposed by Chen Hua et al. Feature extraction method (Chen Hua, An Bin, Chen Shuhai, Liu Yongchang. Application of KL transform in multispectral image clustering. Infrared and Laser Engineering, 2002, 30(2): 79-82.) has achieved good results, However, due to the large amount of multispectral image data, the calculation of the algorithm is very large. On the other hand, the algorithm has too many parameters in the ISODATA clustering, and it is difficult to control. At the same time, the features extracted by the algorithm have no For comparison, it cannot be used for image database retrieval

Images