The Research Of Remote Sensing Image Segmentation And Its Quantitative Relationship To Object-Oriented Classification Accuracy

Object-based Image Analysis(OBIA)is a kind of image information extraction method.Its basic idea is to divide the image into several objects with relatively internal homogeneity which is the smallest unit of image processing.This method can effectively take the shape,texture,inter-object relationship and other features of high-resolution image into consideration,and can enhance the anti-noise,solve some problems in image interpretation effectively,such as "same thing heterogeneous" and "foreign matter with spectrum".Image segmentation is the basis for further analysis,understanding and recognition.The segmentation accuracy of image directly determines the accuracy of image recognition and analysis.Hence image segmentation is of great significance in OBIA.The factors that affect the segmentation mainly include the segmentation algorithm and the selection of the segmentation scale parameters,and the segmentation quality affects the classification accuracy to a great extent.The research on the quantitative relation between them can improve the efficiency and precision of object-oriented classification.In particular,the scientific and systematic evaluation of image segmentation method is very important to accurately evaluate the quality of segmentation.According to the data access and practical experiments,this paper make a systematic and comprehensive introduction and summary for the current high-resolution remote sensing image segmentation method and segmentation evaluation methods and measures.The work accomplished in this paper isas follows:Based on the current status of the segmentation evaluation method,this paper presents a classification system ofgeo-application oriented evaluations of remote sensing image segmentation,and uses three kinds of mainstream multi-scale segmentation methods to process two high-resolution remote sensing images with different characteristics,and then takes the quantitative evaluation to test the rationality of the proposed strategy.In addition,based on the research of the influencing factors of image segmentation,segmentation algorithm and scale parameter selection,the influence of remote sensing image segmentation on object-oriented classification is analyzed.Taking the scale parameter,segmentation quality and object-oriented classification accuracy into consideration,three high-resolution remote sensing images with different characteristics is processed by different segmentation methods and different scales.The results show that the measures of multi-scale segmentation evaluation,homogeneity and heterogeneity,have a negative correlation with scales changing in most cases.Specifically,the homogeneity of segmentation results increases while the heterogeneity declines with the scales enhance,and the comprehensive evaluation index has a high positive correlation with the classification accuracy.This paper reveals the quantitative relationship of them by experiments of multi-scale segmentation and object-oriented classification,which can provide quantitative basis for the further research of OBIA,in particular,the choice of segmentation algorithms and the determination of segmentation scale parameters.