Remote Sensing Image Fusion Based On Quick Bird Satellite & Its Influence On Image Classification

With the continuous development of satellite technology,the earth observing system also tends to be matured,and more and more multispectral and high resolution image data have provided valuable resources for scientific research.How to make full use of these data and integrate them into an image that has both high spatial resolution and spectral characteristics has become a hot topic over the past years.Remote sensing data fusion technology as a multi-level,multi-angle,multi-faceted process,combining different remote sensing data information will provide stronger information interpretation capabilities and more reliable analysis results.This article firstly introduces the background and significance of the remote-sensing image fusion as well as current researches of it.To be specific,this article contains the research status at home and abroad and hierarchical classification of remote sensing image fusion.And it also analyzes the related problems which the technology might face in the future and also,it predicts the possible direction of further research.Secondly,the QuickBird(Panchromatic,Pan)and Multispectral(MS)images of the QuickBird satellite are used as data sources to test and perform pre-processing tests,including the mean,standard deviation,information entropy,and correlation of the images.The coefficient and the spectral characteristic data of each band are analyzed,and the best index factor OIF is calculated to determine the best band combination for subsequent fusion experiments.Besides,the method of combining àtrous wavelet transform with non-subsampled contourlet transform is adopted.At the same time,the color bias caused by the IHS transform is reduced for the QuickBird image.The mean value method is used to calculate the chroma component and the saturation component.The differences in chroma and saturation of the components,the final fusion method of inverse transform and IHS transform,HSV transform,GS transform,Brovey transform,Ehlers transform,DWT transform,IHS and wavelet transform,àtrous wavelet transform,non-subsampling Contourlet transforms are used for fusion experiments,and subjective qualitative and objective quantitative analyses were performed on these ten fusion methods.Finally,on considering the accuracy of classification can better test the pros and cons of the fusion performance,this article adopts the maximum image likelihood classification method,the random forest classification method as well as the support vector machine classification method to classify the fused images,with a combination of the accuracy evaluation method.By longitudinally comparing the classification accuracy of different fusion methods under the same classification method,comprehensive qualitative evaluation and quantitative evaluation,it is concluded that the combination of àtrous wavelet transform and non-subsampled contourlet transformation method is the best;horizontal comparison under the same classification method.According to different fusion methods,the classification accuracy of the random forest classification is better than that of the support vector machine classification and maximum likelihood classification,which is most suiTab.le for the classification of the fusion image.in addition,this article deeply analyzes the impact of fusion on the classification and make a comparison between the results and classification of the fusion image and the unfused image,and find the features that are difficult to sample on the original multi-spectral image can be well sampled on the fusion image.The classification accuracy is improved;the features that cannot be well-distributed before fusion can be better classified after fusion;the image classification accuracy of the fusion image is higher than that of the unfused image,but the phenomenon of foreign matter with the same spectrum still exists;the applicability of fusion algorithm,playing a vital part in enhancing the original spatial resolution and maintaining its spectral properties,which also directly affects the accuracy of classification...