Research Of Land Cover Classification Based On Object-oriented Method

Land cover is an important part of terrestrial life-support systems. It is of greatsignificance to know the situation of land cover timely and accurately. As acomprehensive means of detection, remote sensing plays an important role in the fieldof land cover classification. With the rapid development of remote sensing technology,the higher spatial resolution has become a development trend. High resolution imagessuch as GeoEye, IKONOS, QuickBird, has been widely applied in the field ofsurveying and mapping, forestry, land and so on. High resolution remote sensingimage has abundant spatial information and texture information, but the spectralinformation is relatively less. The traditional image classification method based onpixels, is often difficult to obtain satisfactory results. Object-oriented classificationmethod arises at the historic moment. Object is the minimum unit of Object-orientedclassification method. This classification is based on not only spectral information,but also spatial information and texture information. The accuracy and efficiency ofhigh resolution images’land cover classification both are significantly increased.By reading lots of literatures at home and aboard, the author exposed thePrinciple of the object-oriented method, has analyzed the superiority of theobject-oriented classification method compared whit the pixel-based classificationmethod. Research of land cover classification is based on the high resolution imagesand the eCognition software. The research content and research results of this paperare as follows:(1) In the study area the experimental data is preprocessed. Dealing with thetopographic map data is generated to DEM. The remote sensing image data isradiation-corrected, shot-corrected, band-selected and fused. The band combination isbased on the correlation matrix. The best fusion method is chose by comparison. (2) The optimal partition scale is detailed discussion. Through the experiment,four segmentation scales are choose to build object hierarchy.(3) The object image information, geometry information and spectra textureinformation is made full use to construct fuzzy classification rules. Then use fuzzyclassification method to achieve the classification of land cover in the study area.(4) It takes classification accuracy and kappa coefficient as evaluation index toevaluate the classification results which based on object-oriented classification andmaximum likelihood classification method. By comparisons, the object-orientedclassification method of land cover is obviously more accurate than the classificationmethod based on pixel.