Landslide Recognition Method Based On Object Oriented Research

Extraction of landslide disaster information using remote sensing technology hasbeen studied for several decades, though there are some challenges of inaccuracy ofextraction results in a specific application process.In recent years, the development ofremote sensing technology and information extraction technology has promoted thestudy about improving the accuracy of extraction results. Related researches focus onselection of remote-sensed data and methods of information extraction.In this paper, extracting the landslide disaster information respectively based onthe landslide hazard is carried out by two cases.The main research work and achievements include as following:1. The object-oriented image classification methods have been studied deeply,especially both image segmentation and properties selection.2. The study area was the Wenchuan county in Sichuan after the2008earthquakeThe data sources were the TM and ETM+images.The bands were analyzed to obtainthe optimum composition of spectral bands. Next the image fusion was performed toincrease the resolution with ETM+images,up to15meters.The fused images had beenconducted to maximum likelihood classification, and the classification samples andtesting samples were selected to test the precision of classified results.Correspondingly,The fused images had been conducted to the object-oriented image classification, andthen the better image segmentation was obtained. Further the comparison ofclassification methods had been studied and selected to test the precision of classifiedresults. At last, compared with the two methods, the classification method for betterprecision had been selected to extract the landslide disaster information.3. Weizhou town in Wenchuan country after the earthquake was selected as thestudy area and the data sources was spot5Panchromatic image, then the landslidedisaster information was extracted using the object-oriented image classificationmethod. The optimal segmentation scale about landslide disaster was gained byanalyzing the images. In the optimal segmentation scale, attribute characteristics were chose, and then the landslide disaster information was extracted by comparing theK-near classification method with the object-oriented image classification method.Meanwhile, the landslide disaster information was extracted by using the minimumdistance method as a regular supervised classification method. The object-orientedimage classification method was the best by comparing with the three methods, andthen the local-area landslide disaster information extraction process has been formed.