| In recent years, high spatial resolution remote sensing technology for its highspatial resolution capabilities, can achieve ground information finely and become thegeneral trend of development of remote sensing applications. How to make full useof these advantages to doing object recognition, has become the focus and hotspot ofthe development of remote sensing. With the development of society, urbangeographic information updating is becoming increasingly important and remotesensing images provide a wealth of data sources. But the social function of the city’sdecides that the city has a wide variety of categories. How fast and efficient of toachieve urban land use information extraction has significance. Currently,object-oriented methods of analysis and computation has unparalleled advantages interms of information extraction, has some research significance.In this paper, object-oriented information extraction carryed out mainly byeCognition software. The main research contents are: high spatial resolution remotesensing images preprocessing, object-oriented overview of the theory,selection ofsegmentation parameters and the optimal segmentation scale, classifier selection,application of classification characterization,the research of classification system ofsequential decision-making, development of the rule set and classification accuracyevaluation so on.The basis for a large number of studies in the literature reference, according tothe characteristics of image data to be processed, in allusion to the object-orientedclassification method for obtaining an object in an image, introduce mean-variancemethod, the maximum area method and the automatic threshold method to gain theoptimal segmentation scale; in allusion to the level of difficulty of informationextraction, design strategy based on the decision tree classification order; in allusionto single category of information for urban land use, designed object hierarchynetwork method. According to their spatial distribution and geographicalcharacteristics, by the combinations of the spectral characteristics, geometry, texturefeatures and features class-related features do extraction information efficiently. Inaddition, due to the dense urban tall buildings, so there is a lot of shadow informationinevitably and cause significant distress to information extraction, this paper, contrapose this difficulty, designed shadow compensation method and introduceregion growing method and to bridge the gap method, makes the impact of theshadow information have been resolved.This paper selects high-resolution remote sensing images, Futian District,Shenzhen, to do information extraction. We established set of rules by eCognition forland use information extraction. By analyzing the results of the classification, can weknow that: the classification tree strategy effectively avoid the interference caused bythe spectral information and reduced the leakage classification and misclassificationphenomenon; object hierarchy network method makes information extractionaccuracy improved and makes the impact caused by over-segmentation and undersegmentation of varying scale has been resolved; for information discontinuitiescaused by shadows, the proposed method make some improvement. Application ofthese methods in object-oriented land use information extraction have improved theaccuracy of information extraction in effectively and describes the object-orientedmethod has unparalleled advantages in dealing with high spatial resolution remotesensing images. The above shows that,in the three aspects of classification principle,visual effects and classification accuracy, object-oriented method has unparalleledadvantages in dealing with high spatial resolution remote sensing images. |
