The Study On Pixel-Level Data Fusion For Remote Sensing Images Of Medium And High Resolution

With the development of remote sensing technology, more and moremulti-spectral, multi-resolution and multi-temporal image data are generated bydifferent sensors. To get more correct, clear and complete understanding about thetarget, it is necessary to find means to synthesize different kinds of image data. Thenthe image fusion is developed.This paper focuses on image fusion for multi-spectral images and panchromaticimages of medium and high resolution. In order to enhance the fused image's spatialresolution and preserve spectral characteristics, researchs on algorithms of pixel-levelimage fusion are developed. This thesis mainly includes three parts.(1) Based on the comprehensive review and summarization of previous papersand research work, basic concepts, models, methods, techniques and applications ofremote sensing image data fusion, especially for pixel-level image fusion, arediscussed. Furthermore, assessment of image fusion performance is analysed. Severalevaluation criteria are presented and they are classified with condition and purpose.(2) Methods of pixel-level remote sensing image fusion are analyzed andstudied (including IHS transform, Brovey transform, PCA transform, SVR transform,Pansharp transform and Gram-Schmidt transform). And they are comparedqualitatively and quantitatively with sharpness, entropy, spatial resolution andpreserving the spectral characteristics of source multi-spectral images and so on.(3) A new image fusion method based on artificial neural BP network and SVRtransform is developed and tested. The results of experiment shows the new methodoutperforms SVR transform on enhancing spatial information and preserving thespectral characteristics of source multi-spectral images.