Remote Sensing Image Pansharpen And Classification Research Based On Guided Image Filtering

The guided image filter is a kind of edge preserving filter,which makes the output result in preserving the original image feature,but also can get the edge information of the guidance image.With the increasing popularity of remote sensing technology in daily life,the data sources constantly enriched,high spatial resolution multispectral image means rich information,you can fully take full advantage of the information behind the data.Due to the physical constraint of sensor,remote sensing imagery cannot possess both high spatial and spectral properties simultaneously.Therefore,most earth observation satellites provide simultaneously low-resolution multispectral and highresolution panchromatic images,image pansharpen technique is an effective way to solve this contradiction.Remote sensing image classification is an effect way to extraction information,which is currently popular research in remote sensing technology.But there are some problems in traditional classification methods,especially for hyperspectral image.Hyperspectral images can provide abundant spectral information,playing more and more important role in earth observation system,has been widely used in military and civil fields.Traditional remote sensing image classification techniques is facing some problems for the hyperspectral data,such as: oversize of data,"Hughes" phenomenon,"salt and pepper" in classification results and so on.In this context,we analyze the theory of guide image filter,and then proposed a new remote sensing image pansharpen method and image classification optimization method based on guided image filter.The main contents and innovations are as follows:(1)The theory and process of guide image filter,and some application examples have given to verify effectiveness of this filter in the field of image processing.To understand the advantages of this filter in keeping guidance image's edge information.(2)An approach based on guided image filter is proposed for image pansharpening.The multi-spectral image is employed as a guidance image to the filtering process of lower resolution PAN image to extract the spatial information.Finally,inject the spatial information into the original MS image.The experiment conduct on Gaofen-2 and QuickBird imagery,compared to PCA,Gram Schmidt and UNB-Pansharpen methods.Meanwhile,brightness,spatial and spectral information three evaluation criterias are selected for quantitative assessment.The experimental results show that this algorithm can preserve the edge feature significantly to meet the actual needs.(3)Traditional classification results always exist some problems.In this context,a novel spectral–spatial classification framework based on edge-preserving filtering is proposed.The proposed framework utilizes morphological approach to extract feature of the image,then support vector machine classification method was used to get primary result,finally got the optimized classification result by guide image filter.University of Pavia and Indian Pines hyperspectral data sets are used in this paper.The experiments demonstrate that the guided image filter based method can achieve better results and better accuracies.