Research On The Object-oriented Classification Technology Of High Resolution Remote Sensing Imagery

Compared with the low resolution remote sensing image, high resolution remotesensing image has more Spatial information and texture information, and can showmore clearly feature details, Traditional classification methods are generally focused onlocal pixels，use of the image spectrum information only. Traditional classificationmethod is based on the pixels, and due to ignoring the image texture features, structuralhierarchy, etc,"salt and pepper" phenomenon in classification results is serious.Therefore，the traditional pixel-based classification method，is not suitable for highresolution remote sensing image classification. The object-oriented informationextraction method of remote sensing emerged as a reflection of the times.This paper studies the technology of object-oriented classification forhigh-resolution remote sensing image，including high resolution remote sensing imagesegmentation technology and object-oriented classification technology. On this basis，taking Quickbird image in parts of xuzhou city as an example，the results of this paperwas verified. The main results are as follows:（1） Segmentation technology of high resolution remote sensing image.Discussed the multi-scale segmentation theory，and traditional and new segmentationmethod， color structure code segmentation method based on local homogeneitygradient，which is suitable for high resolution remote sensing images.（2） Object-oriented classification method. Due to the different object hasdifferent spectral，texture and spatial structure，so choosing different classificationmethods according to different terrain features is necessary.. Based on the multi-scalesegmentation， custom threshold method is used to extract water， shadow andvegetation；the method of support vector machine is used to extract road and building.Taking the Quickbird images of part of Jiangsu Normal University as the datasource，the segmentation and classification are made using the method described above.As is object-oriented classification method，the traditional error matrix analysis is notsuitable for the accuracy evaluation，so use the best classification results and thestability classification based on of fuzzy mathematics to evaluate the classification results. The results show that，the classification results of the method used in this paperare complete and accurate，and are suitable for high resolution remote sensing imageinformation extraction.