Font Size: a A A

Research On Hyperspectral Remote Sensing Image Classification Algorithm Based On FSVM

Posted on:2011-05-31Degree:MasterType:Thesis
Country:ChinaCandidate:L MaFull Text:PDF
GTID:2178360305477109Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Remote sensing technology is one of modern technology, which greatly expand the vision of people, rich in sources of earth observation information. In recent years, along with the continuous development of remote sensing technology, especially of hyper-spectral remote sensing, remote sensing, a big brought the revolutionary, greatly ex-panded the remote sensing applications. Therefore for hyperspectral remote sensing technology has been widely studied theoretical significance and application value, has been the research focus in the field of remote sensing.Hyperspectral remote sensing data of remote sensing data analysis and classifica-tion is an important method of information extraction. For support vector machine suit-able dimension characteristic, the superiority of small samples, is a kind of extremely potential of hyperspectral remote sensing classification method, but because it is sus-ceptible to noise cause is not high accuracy of classification, currently used in remote sensing data classification using only the spectral characteristics and classified using support vector machine (SVM) of different weights of characteristic vector with the same situation, this paper proposes a method of hyperspectral remote sensing image classification algorithm based on FSVM, so as to get a good classification accuracy.Classification algorithm based on fuzzy support vector machine chooses spectral characteristics and texture characteristics as sample characteristic vector. For those two different characteristics, according to the importance, it has to use different weight to normalize the different characteristics. The choice of the weight was measured by the anisotropic of different characteristics of the class, which is determined by the ratio be-tween the distance from the center of the projection of category to sample and distance of center of category to sample. In the sample training, first of all, it has to select the 9*9 matrix which uses pixel to be the center, calculating the graph of the texture cha-racteristics which will be the sample of the texture characteristics, and also using the method of calculating the images Band Index, and then calculating spectral characteris-tics of each sample clustering center. To each clustering center, find the spectrum and texture features of average weight between them and other sample spectra to be the ?sample's multi-characteristics weight. At first, when divide each test samples, which is disposing every testing sample after the normalized processing, then multiply the value of corresponding to be eigenvector, and finally to sort those forms. Classification algo-rithm adopts fuzzy sets and support vector machine (SVM) method, combining fuzzy membership as relaxation variables, punish factor and the coefficient of the product of fuzzy membership to measure the importance of different wrong points, including error of fuzzy membership of the dynamic test samples by adding to each category, through the calculation of sample before adding to measure the size of variance of change. For the multi-class classification problem, one-to-one Categories designed classifier was used.Through AVIRIS hyperspectral remote sensing images as experimental data ob-tained after the classification of confusion matrix and Kappa coefficient, through analy-sis that based on fuzzy support vector machine (SVM) characteristics of hyperspectral remote sensing image classification algorithm has good effect, the relative standard classification algorithm of support vector machine, the classification of precision, accu-racy and overall accuracy increased 3% ~ 10%.
Keywords/Search Tags:Remote Sensing, Hyperspectral, Band Index, Texture, Fuzzy Support Vector Machine
PDF Full Text Request
Related items