Research On Worldview Short-wave Infrared Image Fusion Method Based On Guided Filtering And Application

With the development of sensor technology and remote sensing platforms,highresolution,hyperspectral,multi-platform and multi-temporal image data has grown rapidly,and thanks to the research of remote sensing image fusion,a large number of remote sensing image data and image fusion technologies have been applied to disaster monitoring,national defense and other fields.As one of the most advanced highresolution commercial remote sensing satellites,Worldview satellites can provide not only panchromatic band and eight multispectral bands,but also eight short-wave infrared bands.However,only short-wave infrared image with a resolution of 7.5m can be obtained in China,which is quite different from the 1.2m resolution of multispectral image.At the same time,Worldview multispectral image and short-wave infrared image belong to heterogenous images.There are few researches on fusion and classification applications,and the traditional image fusion methods have serious spectral distortion problems.The above factors limit the application value of Worldview images.In this regard,this paper takes the Worldview short-wave infrared image as the research object,proposes a new image fusion method and verifies the image classification results before and after the fusion.The following points is included in this paper.1)A Worldview short-wave infrared image fusion method(GFCS-P)based on guided filtering and principal component transformation is proposed.The method uses principal component transformation to separate spatial information and spectral information of an image,then the spatial detail information is extracted from multispectral image by guided filtering and fused into the short-wave infrared image,effectively overcoming spectral distortion and image blur.2)The GFCS-P fusion method is optimized,and the base layer obtained by the guided filter decomposition of the multispectral image is further decomposed into the edge layer and the low frequency layer.A guided filter fusion algorithm(TGF)based on the three-layer decomposition is proposed.That is,the multispectral image is decomposed into a detail layer,an edge layer,and a low-frequency layer,and the detail layer and the edge layer are combined to obtain richer spatial detail information.Compared with the GFCS-P method,the TGF method further enhances the information volume and clarity of the fused image.3)In the land use classification section,the short-wave infrared image is integrated with multispectral image into 16-band image.The classification results show that the 16-band image can significantly improve the overall accuracy of land classification compared to the 8-band multispectral image,especially Industrial plants and bare land.The overall accuracy can be improved by about 1.80%.In order to analyze the application effect of the TGF fusion method,the multispectral image combined with the short-wave infrared image before and after fusion was used for classification experiments.The results show that the TGF fusion method can obtain the highest overall classification accuracy compared to the traditional fusion method.This paper proposes the GFCS-P method and the TGF fusion method by exploring the fusion method of Worldview multispectral image and short-wave infrared image.The latter is optimized for the former,which not only overcomes the spectral distortion problem,but also improves the clarity of the fused image.And it has a good application effect in the land use classification.