Automatic Extraction And Change Detection Method Based On A Typical Research Target Object-oriented

The rapid development of satellite remote sensing technology and the greatimprovement of spatial resolution and temporal resolution of remote sensing imageobtained make the information we obtain about the Earth’s surface richer and moreconvenient for us to track or monitor information about changes in the interestedregion. High spatial resolution image data while reducing the mixed pixel, increasesinternal spectral differences between the similar surface features. When the traditionalpixel-based spectral information extraction technology can not meet the currentrequirements of remote sensing image information extraction based on high spatialresolution, the object-oriented image analysis techniques provides new ideas andmethods for the high spatial resolution remote sensing image information extractionand change information detection.This paper through a comprehensive analysis and summarization of researchachievements about information extraction and change detection and by usingQuickBird-a high resolution image data, carrys out the following work areas directedat the automatically extraction of changes information of typical target in interestedregion:1. Comprehensive comparative analysis of the advantages and disadvantages ofthe traditional pixel-based classification method and object-oriented classificationmethod. Based on the inherent defects of the traditional classification methods, applythe object-oriented thinking into the extraction of target information, and elaboratessome important parts such as image segmentation, feature extraction and image objectclassification in the object-oriented classification method. Experimental analysis andinstructions are also conducted to solve scale effect problem in object-orientedinformation extraction.2. Summarizes main existing changes detection methods and analyzes the prosand cons of various methods of changes detection specific to characteristics of highspatial resolution image. Comparative experiment of the traditional image algebra change detection method and change detection method based on object-orientedclassification has been carried out on verify that change detection method based onobject-oriented classification can better eliminate the impact of changes in imageregistration errors on the test results.3. Experiment has been carried out to do comparative analysis about nowwidely-used fractal network evolution method, watershed approach and mean shiftmethod based on confidence level. And as for problems such as excessivesegmentation and region more fragmentation and other issues in segmentation method,proposes processing method of fusing both edge and spectral information forsegmentation processing and then combine small regions based on local homogeneity,which is evidenced through experiments by the result that the extraction effect ofimage objects is pretty well.4. Improve methods to deal with too much interference of subjective factors intocharacteristic space optimization and classification parameters selection inclassification process of images. Creatively put forward an approach to obtain thedistance distribution of characteristic parameters of each object through the wholesample statistical analysis of supervising object-based image classification, andautomatically select target classification threshold based on threshold selectionformulas. The advantage of this approach is to avoid the introduction of subjectiveerrors, thereby making the threshold more scientific and quantitative.