Research On Object-oriented Information Extraction From High-resolution Remote Sensing Image

In recent years, it appears the high resolution remote sensing image which letspeople obtain more and more rich information from the surface of the earth. With theenhancement of spatial resolution of the remote sensing image, the more and moreabundant space information of landmark and the gradual increase of internal spectraldifference of the same kind of landmark, the traditional pixel-based spectralinformation extraction technology can not meet requirements of the high spatialresolution remote sensing image information extraction. Object-oriented imageanalysis technology provides new thought and method for high spatial resolutionremote sensing image information extraction.In this paper, it applies the object-oriented image analysis technology to highresolution remote sensing image information extraction. Firstly, the image is segmentedinto homogeneous image objects, and then analyzing the image object’s spectrum,shape, texture, levels and correlation characteristics with the quantitative descriptionmethod realizes information extraction by using fuzzy classification techniques ofmembership function. During the segmentation of image, it adopts the multi-scalesegmentation technique which can generate different scales hierarchy of image objectat fixed spatial resolution of remote sensing images, whereas different surface entityobjects are corresponding to different scales. This paper proposes the method to selectoptimal segmentation scale according to the ratio of heterogeneity and homogeneity,which can determine the different feature types of optimal spatial scale for the selectingproblem of scale based on the principle of "image object homogeneity of maximum and heterogeneity of minimum".Based on the above-mentioned theory, this paper, selecting GeoEye high spatialresolution image as experimental data and using eCognition as image processingsoftware platform, extracts feature information from the experimental area. During theexperiment, according to the different characteristics of the objects, it applies themethod of the ratio of homogeneity and heterogeneity that determines the differentfeatures of the optimal segmentation scale to experimental data segmentation. Then itsets up different features of the multi-level architecture, according to the establishmentof the fuzzy rule for information extraction based on the characteristics of image objectinformation. Finally, it makes accuracy evaluation on the extraction results with thefuzzy concept of membership degree angle. The results show that, informationextraction has high accuracy by using the object-oriented image analysis techniques forhigh resolution remote.