遥感影像分割及其对面向对象分类精度影响的定量研究

发布时间：2019-03-15 13:19

【摘要】：面向对象影像分析(Object-Based Image Analysis,OBIA)是一种影像信息提取的方法,其基本思想是将影像分割成若干个内部性质相对均一的对象,以对象为最小单位对影像进行操作处理。这种方法可以有效的利用高分辨率影像丰富的形状、纹理、对象间关系等特征,且可增强抗噪性,有效解决“同物异谱”、“异物同谱”等问题。影像分割是对遥感影像进行进一步分析、理解和识别的基础,影像的分割精度直接决定着对影像的理解、识别和分析的准确程度,因此影像分割在OBIA中具有重要意义。影响分割结果的因素主要包括分割算法和分割尺度参数的选择两方面,而分割质量很大程度上影响分类精度,对于其间定量关系的研究有助于提高面向对象分类的效率和精度。其中,影像分割评价方法的科学性和系统性对准确评价分割质量至关重要。本文通过资料查阅及实际实验,对目前的高分辨率遥感影像分割、分割评价方法及评价测度做了较为全面、系统的介绍和总结。论文完成的主要工作如下:基于目前分割评价的研究现状,本文提出了面向地学应用的遥感影像分割评价体系框架,并利用三种主流的多尺度分割方法处理了两幅不同特性的高分辨率遥感影像,对分割结果进行定量评价,验证了本文提出的遥感影像分割评价及最优结果选择策略的合理性。通过对影响分割结果的两个重要因素,即分割算法和尺度参数选择的研究,分析遥感影像分割对面向对象分类的影响,并将多尺度分割的尺度参数、分割结果质量及面向对象分类精度综合考虑,对三幅不同特性的高分影像利用不同分割方法和分割尺度进行实验,实验结果表明斑块内均质度和斑块间异质度两个多尺度分割评价指标在多数情况下随尺度变化呈负相关关系,且斑块间异质度随尺度增加而增加,斑块内均质度随尺度增加而降低,综合评价指标与分类精度有较高的相关性。本文通过实验揭示了分割尺度、分割质量和分类精度的定量关系,为未来面向对象影像分析,尤其是分割算法的选择及分割尺度参数的确定提供了定量依据。
[Abstract]:Object-oriented image analysis (Object-Based Image Analysis,OBIA) is a method for extracting image information. Its basic idea is to segment the image into several objects with relatively uniform internal properties, and to operate and process the image with the object as the smallest unit. This method can effectively utilize the rich shape, texture and relationship between objects in high-resolution images, and can enhance the anti-noise ability, and effectively solve the problems of "homologues and different spectra" and "foreign objects and spectra". Image segmentation is the basis for further analysis, understanding and recognition of remote sensing images. The accuracy of image segmentation directly determines the accuracy of understanding, recognition and analysis of remote sensing images. Therefore, image segmentation is of great significance in OBIA. The factors that affect the segmentation result 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 relationship between them is helpful to improve the efficiency and accuracy of object-oriented classification. Among them, the scientific and systematic evaluation method of image segmentation is very important for the accurate evaluation of segmentation quality. In this paper, the current high-resolution remote sensing image segmentation, segmentation and evaluation methods and evaluation measures are comprehensively and systematically introduced and summarized through data retrieval and practical experiments. The main work of this paper is as follows: based on the current research status of segmentation and evaluation, this paper proposes a remote sensing image segmentation and evaluation system framework for geoscience application. Three mainstream multi-scale segmentation methods are used to deal with two high-resolution remote sensing images with different characteristics, and the segmentation results are quantitatively evaluated. The rationality of the segmentation evaluation and the optimal result selection strategy proposed in this paper is verified. Based on the study of two important factors affecting the segmentation results, namely, segmentation algorithm and selection of scale parameters, the influence of remote sensing image segmentation on object-oriented classification is analyzed, and the scale parameters of multi-scale segmentation are analyzed. Considering the quality of segmentation results and the accuracy of object-oriented classification, experiments are carried out on three high-resolution images with different characteristics using different segmentation methods and segmentation scales. The experimental results show that in most cases there is a negative correlation between intra-plaque homogeneity and inter-plaque heterogeneity, and the heterogeneity between patches increases with the increase of scale. The degree of homogeneity in plaques decreased with the increase of scale, and there was a high correlation between comprehensive evaluation index and classification accuracy. In this paper, the quantitative relationship among segmentation scale, segmentation quality and classification accuracy is revealed by experiments, which provides a quantitative basis for future object-oriented image analysis, especially for the selection of segmentation algorithms and the determination of segmentation scale parameters.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【相似文献】