High-resolution Image Classificationi Based On Urban Application

Town of high-resolution remote sensing applications is the most important one of the remote sensing applications, how to get accurate image classification is a difficult of high-resolution remote sensing research. For high-resolution data classification only using the spectral information under the traditional classification which have a low classification accuracy and could not play to the advantages of high-resolution images and effectively distinguish between features.This article proposed in the use mathematics morphology the smooth filter method carries on the high resolution image in turn the filter, the filter is advantageous to the image division classification, carries on the image division and the image classification using the filter result.As high-resolution images is rich in shape, adding shape classified information is to improve the accuracy of a common approach. how to use the shape information is on the hot spots. Hu moments with the same rotation invariance, the scale invariance, as well as translation invariant features, so in this text calculating a different scale of the image segmentation processing, and calculate the segmentation of different scales's Hu moment, together with the spectrum of information Participate in the classification, can improve the classification accuracy.This article proposed the level classification method, the vegetation uses based on the spectrum classification, to the city spectrum extremely similar non-vegetation, through the different criterion image division, calculates the Hu not bending moment, joins the shape information, has carried on separately based on the element and the object-oriented two classified method, and has compared two classified methods through the experiment.