Segmentation Quality Evaluation Method Study Of High Spatial Resolution Remote Sensing Images

High spatial resolution remote sensing images have been widely applied in many fields such as land use,city planning and environmental evaluation,and it is of great importance to how to quickly and effectively extract image information from high spatial resolution remote sensing images.Traditional pixel-based image analysis may result in “salt and pepper” effect which directly affects the extract quality of high spatial resolution remote sensing images information.Now,object-based image analysis(OBIA)method appears.Image segmentation is the key process in OBIA which segmentation result quality directly affects subsequent image information extracting quality.Therefore,it becomes a worthy studying topic to evaluate image segmentation quality of high spatial resolution remote sensing images.Currently,supervised evaluation method and unsupervised evaluation method are main quantitative methods for image segmentation quality evaluation.Unsupervised evaluation method has strong subjective to design evaluation indicators,so it has some limits in application.Under guaranty of the reference objetcs accuracy,supervised evaluation is a common method which can be objective to evaluation image segmentation result quality.This dissertation applied supervised evaluation method to construct image segmentation quantity evaluation method and established individual image object evaluation index according to area discrepancy and position discrepancy between reference object and segment object.It establishes image segmentation quality evaluation index and the optimal segmentation evaluation criterion for whole image based on these indexes.The study main contents are as follows:(1)The establishment of high spatial resolution remote sensing images segmentation quality evaluation method.It construsts undersegmentation evaluation index and oversegmentation evaluation index according to the discrepancy of area and number between reference objects and all intersecting segmentation objects,and also analyzes the level of undersegmentation and oversegmentation of segmentation results,then reduce the segmentation results with high level undersegmentation or oversegmentation results and use the remainder segmentation results in the later quality evaluation of image segmentation.According to spatial topological relation and area discrepancy between reference object and segmention object,it determines the vaild segmentation objects that are used to evaluate the quality of segmentation result,and dissolve them to build a new polygon,then take is as the research object.By analyzing area discrepancy and position discrpancy between reference object and research object,the ESI(Error Segment Index)and CDI(Centroids Distance Index)are proposed to evaluation image single segmentation result quality,and new indexes that are AESIall and ACDIall to evaluate segmentation quality in entirety or certain areas are proposed with them.(2)The validation of high spatial resolution remote sensing images segmentation quality evaluation method.Using the quality evaluation method in this dissertation,the study evaluates segmentation result quality which are obained from different segmentation parameter combinations,and find out the optimal segmentation result of buildings in the image.The optimal segmentation result of buildings can high match the reference objects through visual analysis,which demonstrate the evalutation method is effective and reasonable.And the proposed method in this dissertation has certain superiority compared with the other evaluation method.(3)The application of the segmentation quality evaluation method of high spatial resolution remote sensing images.It applied the proposed evaluation method in this dissertation to evaluation the quality of any given image segmentaion results,and find out the optimal image segmentation result,which further verified the feasibility and universality of this method.