基于数据挖掘技术的航摄影像土地利用变化检测研究

发布时间：2018-07-01 12:27

本文选题：航摄影像 + 土地利用变化检测　；参考：《东南大学》2017年硕士论文

【摘要】：土地利用变更调查是我国每年需进行的重要调查之一,城市土地利用变化信息对于分析城市的经济发展状况、监测城市土地利用变化动态以及城市自然生态环境变化分析具有重大的意义。航摄影像数据具有空间分辨率高,数据获取实时性强、成本低等特点,是城市土地利用变化检测的良好数据源。目前,在采用航摄影像数据进行土地利用变化检测时主要存在的问题包括:数据预处理质量不稳定、影像光谱分辨率低、"同物异谱"现象明显以及变化检测方法自动化程度低等。针对以上问题,本文将数据挖掘C4.5算法应用于决策树分类规则的生成中,对基于数据挖掘技术的分类后土地利用变化检测方法进行了实验研究,并将实验结果与基于面向对象分类法的土地利用变化检测结果进行了对比,得到以下主要结论:(1)在航摄影像数据预处理过程中,采用基于灰度信息的自动配准结合人工筛选的方法进行影像配准,并确保同名匹配点选在同一高程基准面,可达到亚像元级配准精度;采用平均值-标准差的相对辐射归一化方法对航摄遥感影像进行辐射校正,校正结果与前时相影像的均方根误差波段平均值为16.346,前后时相影像辐射值达到了较高的相似性,且信息熵差值小于0.1,信息量损失较少。(2)高分辨率航摄影像分类过程中,由"同物异谱"现象造成的分类误差可通过控制样本采集与训练的质量来进行有效削弱。在采集样本时,对部分存在"同物异谱"现象的地物进行类别细化,采用J-M可分离度指数样本训练质量评价,样本间分离度均大于 1.8。(3)构建了包含影像光谱信息和纹理信息的特征空间,验证了纹理信息在土地利用变化检测中的有效性,弥补了航摄影像数据光谱分辨率的不足,使数据得到了更加充分的利用。(4)采用基于数据挖掘C4.5算法的决策树分类法进行土地利用变化检测,得到的分类精度和变化检测精度与基于面向对象的方法精度相当,总体分类精度大于85%,变化检测误检率均小于10%。结果表明,基于数据挖掘C4.5算法的决策树土地利用分类在人工依赖性方面优于面向对象方法,并且在提高工作效率的同时也能保障分类的精度,具有较高的实用价值。
[Abstract]:The investigation of land use change is one of the important surveys that need to be carried out every year in our country. The information of urban land use change is of great importance to the analysis of the economic development of cities. It is of great significance to monitor the dynamic of urban land use change and the analysis of urban natural ecological environment change. Aerial photography data is a good data source for urban land use change detection because of its high spatial resolution, real-time data acquisition and low cost. At present, the main problems in land use change detection using aerial photography data include: the quality of data preprocessing is unstable, The spectral resolution of image is low, the phenomenon of "isomorphism" is obvious, and the automation of change detection method is low. Aiming at the above problems, this paper applies the C4.5 algorithm of data mining to the generation of decision tree classification rules, and makes an experimental study on the detection method of land use change after classification based on data mining technology. The experimental results are compared with the results of land use change detection based on object-oriented classification. The main conclusions are as follows: (1) in the process of aerial photography data preprocessing, The method of automatic registration based on gray level information combined with manual selection is used to carry out image registration and ensure that the matching points of the same name are selected in the same elevation datum level, which can achieve sub-pixel registration accuracy. The radiometric correction of aerial remote sensing image is carried out by using the method of relative radiometric normalization of mean value and standard deviation. The correction result is 16.346 with the mean square error band of the pretemporal image, and the radiative value of the front and rear phase image reaches a high similarity. And the difference of information entropy is less than 0.1, and the loss of information is less. (2) in the process of classification of high-resolution aerial photography, the classification error caused by the phenomenon of "same object heterospectral" can be effectively weakened by controlling the quality of sample collection and training. When collecting samples, some of the objects with "isomorphism" phenomenon are classified and evaluated by J-M separability index. The separation degree between samples is greater than 1.8. (3) the feature space including image spectral information and texture information is constructed, which verifies the validity of texture information in land use change detection, and makes up for the lack of spectral resolution of aerial photography data. (4) the decision tree classification method based on data mining C4.5 algorithm is used to detect land use change. The accuracy of classification and change detection is similar to that of object-oriented method. The overall classification accuracy is more than 85%, and the error detection rate is less than 10%. The results show that the decision tree land use classification based on data mining C4.5 algorithm is superior to the object-oriented method in terms of artificial dependence, and it can improve the working efficiency and ensure the accuracy of the classification. It is of high practical value.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【相似文献】