Classification Of Hyperspectral Image Based On Imge Fusion

With the development of hyperspectral imaging, hyperspectral remote sensing remotesensing into the emerging field of direction, with the characteristics of a high spectralresolution, large amount of information, multi-band and a strong correlation between thebands. It propose a very high demand on the high-spectral image processing technology.Hyperspectral remote sensing image classification of hyperspectral remote sensingtechnology is a very important application, However, due to problem of the high spectraland spatial resolution images of the low-band information redundancy, classificationresults are not good.In this paper,against the characteristics for high-spectral images,combined with high spatial resolution panchromatic image, in the study of multi-spectralimage fusion algorithm, a method of High spectral classification is proposed based onthe proposed high-spectral image fusion. The specific content of this paper contains thefollowing three main parts:First, introduce the concept of remote sensing image classification, and the currentdomestic and foreign remote sensing image classification technology development status.And describes the two types of remote sensing image classification algorithms. Also byanalyzing the characteristics of hyperspectral image data, hyperspectral imageclassification methods are introduced.Second, introduce the concept of traditional remote sensing image fusion, includingremote sensing image fusion levels, methods, purpose, and integration of evaluationmethod. By introducing the integration of three levels: pixel level fusion, feature fusion,decision level fusion, detailing the IHS transform, principal component analysis (PCA),Brovey transform methods and their advantages and disadvantages. in addition, give themethod of Evaluation of the effectiveness of remote sensing image fusion.Finally, based on the analysis of traditional remote sensing image fusion, improve theclassification accuracy through the fusion of high-spectral image with high spatialresolution image In the preprocessing stage，use best exponential model to make the bestband selection. Fusion method using Gram Schmidt transformation method, with the effect ofcontrast to the IHS transform, principal component analysis (PCA), Brovey transform methods and so on, we can see, this method can better maintain the original hyperspectralimage spectral information, but also the larger enhance its spatial details.Finally,through acombination of texture features maximum likelihood classification to classify fusionimage and analysis.