全色遥感影像地物信息自动分类方法研究

发布时间：2019-03-12 13:06

【摘要】：随着我国经济的快速发展,城市化发展政策正在紧锣密鼓的进行当中,因此对土地资源的需求量也在不断提高。利用遥感影像对城市典型地物进行分类提取已经成为现今的主流趋势,利用合适的算法实现对城市典型地物信息的高精度提取是现在最主要的研究课题之一。本文使用“吉林一号”光学A星全色遥感影像,使用基于像元和面向对象这两种分类方法对区域内的建筑、道路、林地及草地四种典型地物进行分类提取研究,主要工作如下:(1)对于基于像元的分类方法,主要研究监督分类法的最小距离法、最大似然法、BP神经网络法和支持向量积法以及非监督分类法的ISODATA算法和K-means算法;(2)对于面向对象分类方法,根据影像的光谱、形状和纹理特征,使用基于Sobel边缘算子的分割技术和基于Full Lambda-Schedule的分割-融合算法对影像进行分割,再分析每种地物的特点利用模糊分类方法建立信息提取规则。(3)研究了混淆矩阵的分类精度评价方法,针对每种算法的分类结果,使用总体分类精度、Kappa系数这两个指标对总体分类精度进行评价,使用错分误差、漏分误差及单类分类成功指数这三个指标对单类地物的分类精度进行评价。实验结果与数据表明:对于基于像元的分类方法来说,总体分类精度最高的分类方法是监督分类中的最大似然法,总体分类精度为83.8680%,Kappa系数为0.7561,对道路、建筑、草地这三种地物进行单类提取时,最大似然法的单类分类精度最高,对林地进行单类提取时,支持向量积法的分类精度最高;对于面向对象分类方法来说,总体分类精度为94.4721%,Kappa系数为0.903。总体上,面向对象方法的分类精度都要高于基于像元的分类精度。本文的研究成果对长春的城市化建设发展具有十分重要的指导意义,对“吉林一号”卫星后续获取的影像进行分类处理时也起到一定的借鉴指导作用。
[Abstract]:With the rapid development of China's economy, the development of urbanization policy is being carried out intensively, so the demand for land resources is also increasing. Using remote sensing images to classify and extract typical urban features has become a mainstream trend. It is one of the most important research topics to use appropriate algorithms to extract urban typical features with high accuracy. In this paper, "Jilin No. 1" optical A-star panchromatic remote sensing image is used to classify and extract four typical features of buildings, roads, woodlands and grasslands in the region using pixel-based and object-oriented classification methods. The main works are as follows: (1) for pixel-based classification methods, the minimum distance method, maximum likelihood method, BP neural network method, support vector product method, ISODATA algorithm and K-means algorithm for supervised classification are mainly studied. (2) for the object-oriented classification method, according to the spectral, shape and texture features of the image, the segmentation technology based on Sobel edge operator and the segmentation-fusion algorithm based on Full Lambda-Schedule are used to segment the image. Thirdly, the classification accuracy evaluation method of confusion matrix is studied. According to the classification results of each algorithm, the overall classification accuracy is used, and the fuzzy classification method is used to set up the information extraction rules. (3) the classification accuracy evaluation method of the confusion matrix is studied, and the overall classification accuracy is used for each algorithm. The Kappa coefficient is used to evaluate the overall classification accuracy. The classification accuracy of single-class ground objects is evaluated by using three indexes, namely, the error of classification, the error of missing classification and the index of success of one-class classification. The experimental results and data show that for pixel-based classification methods, the most accurate classification method is the maximum likelihood method in supervised classification, the overall classification accuracy is 83.868%, the Kappa coefficient is 0.7561, for roads, buildings, The single classification accuracy of the maximum likelihood method is the highest when the three kinds of ground objects are extracted, and the classification accuracy of the support vector product method is the highest when the forest land is extracted by the single class method. For the object-oriented classification method, the overall classification accuracy is 94.4721%, and the Kappa coefficient is 0.903. On the whole, the classification accuracy of object-oriented method is higher than that of pixel-based method. The research results of this paper have a very important guiding significance for the development of urbanization in Changchun, and the classification and processing of the images acquired by Jilin 1 satellite also play a guiding role in the classification of the images obtained by Jilin 1 satellite.
【学位授予单位】：长春理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】