The Study Of Scale Issues In Information Extraction From High Resolution Remote Sensing Image

With the development of remote sensing technology and the expansion of itsapplication, the problem of remote sensing scales gradually becomes an importantresearch direction. The problem of remote sensing scale focuses on three areas,including scale transformation, scale effect and the choice of optimal scale. This studyuses the data from IKONOS multi-spectral images that their spatial resolution is4m todo research on the problem of information extraction from high spatial resolutionremote sensing images. Firstly, it’s need to make scale transformation tomulti-spectral image data and evaluate its conversion effect, and then choose theoptimal scale of all types of surface features, and finally do the experiment ofextraction on basis of multi-scale remote sensing images. The specific contents andconclusions are as follows:(1) Using three methods to do scale expansion of the images of the experimentalarea, and analyze the effect of scale conversion using the criteria of remote sensingimage quality evaluation. The results show that: the nearest neighbor method canmaintain the spectral performance in each band of images better, and it’s also gooduse in the amount of information and clarity of the image effects. Especially, when theextended scale is less than40m, the effect of scale expansion is the best and the datafrom the result can be used in the follow-up study.(2) Using variation function method to calculate the optimal scale of all types ofsurface features, and find that as for different objects, the expression of the optimalscales are not the same; even the same feature types which have different layout in theform, the optimal scales are also different; to the same feature types in different bands,the optimal scale are not identical; the higher spatial resolution of images may notdescribe all the objects better. On the basis of the method of statistical separabilitymeasure to do the research of the separability between the all types of surface featuresand get the comprehensive conclusion: when do the information extraction of remotesensing images from surface features, multi-scale image data can be used to improvethe accuracy of information extraction. (3) Through the superimposing of optimal scale image data from all types ofsurface features and classification to multi-scale image data using supervisedclassification method, the results showed that, the method of the multi-scale imagedata based on the classification can improve the overall accuracy of classification ofhigh resolution remote sensing images.