基于遥感技术的湖泊历史变迁检测及展示

发布时间：2018-09-05 15:52

【摘要】：本文通过对不同时期武汉市遥感影像资料进行预处理提取沙湖影像资料,并通过采用多尺度分割和决策树分类的方法对影像资料进行分类处理,利用智能手机等移动终端来展示湖泊历史的变迁情况,为湖泊执法人员检测违规占湖、填湖的现象提供一种快速、准确、便捷的办法,进一步提高执法人员的执法效率,同时也为违规填湖取证提供参考。本文用到的变化检测方法为图像分类比较法,研究主要分为两个方面:遥感影像的预处理和面向对象的影像分类;在移动端检测分类后影像的变化区域。主要研究内容和创新之处包括以下几方面:1、研究遥感影像的预处理技术。本文通过对湖北省地理信息中心获取的航空影像资料进行几何纠正、影像镶嵌、影像裁剪的处理,来获取武汉市沙湖的影像资料,并经过重采样使得历年的影像资料具有相同的分辨率(1M)。2、多尺度分割方法中最优分割尺度。在通过采用多尺度分割方法对影像进行面向对象遥感影像分类时,尺度参数的选取对分类结果的影响很大,不同的分割尺度对影像进行分割时得到的对象大小不同且数量不等,尺度越小生成的对象越多,会造成过度分割的现象,反之,尺度越大生成的对象越少,会导致分割不足。由于不同地物类别的光谱、纹理、结构等特征存在差异,所以,不同的地物类别有不一样的最优分割尺度。本文基于对象内部同质性高且外部异质性高的原则建立分割质量函数。通过对不同尺度参数下的分割质量函数值进行比较,来判断最优分割尺度。3、基于多尺度分割和决策树分类方法的研究与改进。本文对多尺度分割和CART决策树分类方法进行深入的研究和分析。在传统的多尺度分割方法上引入图像分块的概念,并提出两种分块的方法:基于影像“拼接线”分块的方法和基于大尺度的多尺度分割的分块方法。对子块进行多尺度分割和决策树分类。以81年的沙湖影像数据为实验对象,把这两种改进的分类方法与传统的多尺度分割和决策树分类的影像分类结果进行精度对比,结果显示:基于分块的多尺度分割和决策树分类方法的精度明显高于对整幅影像进行多尺度分割和决策树分类方法。其中基于大尺度的多尺度分块方法精度最高。4、基于智能移动终端的湖泊变迁区域检测。简要介绍武汉市水文化App系统,同时在系统的变迁功能模块中实现对分类后影像图的变迁区域的检测。
[Abstract]:In this paper, the sand lake image data are extracted by preprocessing the remote sensing image data of Wuhan in different periods, and the image data are classified and processed by using the method of multi-scale segmentation and decision tree classification. The use of mobile terminals such as smart phones to show the historical changes of lakes provides a fast, accurate and convenient way for lake law enforcement personnel to detect illegal occupation of lakes and fill lakes, and further improve the efficiency of law enforcement personnel. At the same time, it also provides a reference for evidence collection of illegal lake filling. The change detection method used in this paper is the image classification and comparison method. The research is mainly divided into two aspects: remote sensing image preprocessing and object oriented image classification; The main research contents and innovations include the following aspects: 1, the research of remote sensing image preprocessing technology. In this paper, the image data of Shahu Lake in Wuhan City is obtained through geometric correction, image mosaic and image tailoring of aerial image obtained by Hubei Geographic Information Center. After resampling, the image data have the same resolution (1m). 2, the optimal segmentation scale in the multi-scale segmentation method. When using multi-scale segmentation method to classify the object oriented remote sensing images, the selection of scale parameters has a great influence on the classification results, and the size and number of objects are different when different segmentation scales are used to segment the images. The smaller the scale is, the more objects will be generated, which will lead to over-segmentation. Conversely, the larger the scale is, the fewer objects will be generated, which will lead to insufficient segmentation. Because there are differences in spectral, texture and structure characteristics among different types of objects, there are different optimal segmentation scales for different types of objects. Based on the principle of high internal homogeneity and high external heterogeneity, the segmentation quality function is established in this paper. By comparing the value of segmentation quality function under different scale parameters, we can judge the optimal segmentation scale. 3. The research and improvement of the classification method based on multi-scale segmentation and decision tree are presented. In this paper, multi-scale segmentation and CART decision tree classification methods are deeply studied and analyzed. The concept of image segmentation is introduced in the traditional multi-scale segmentation method, and two methods are proposed: one is based on image "splicing lines" and the other is based on large-scale multi-scale segmentation. Multi-scale segmentation and decision tree classification for subblocks are carried out. Taking the 81 year sand lake image data as the experimental object, the accuracy of these two improved classification methods is compared with the results of traditional multi-scale segmentation and decision tree classification. The results show that the accuracy of multi-scale segmentation and decision tree classification based on block is obviously higher than that of multi-scale segmentation and decision tree classification of whole image. The method based on large scale multi-scale block has the highest precision of .4and the detection of lake transition area based on intelligent mobile terminal. This paper briefly introduces the Wuhan Water Culture App system, and realizes the detection of the transition region of the classified image map in the system transition function module.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】