基于面向对象的柑橘类果林信息提取方法研究

发布时间：2018-07-06 11:07

本文选题：面向对象 + 模糊数学　；参考：《江西理工大学》2017年硕士论文

【摘要】：我国是柑橘类水果产品主要的生产国和消费国之一,有着4000多年的栽培历史。快速有效地提取柑橘类果林对指导果林病害防御、水果生产和产业规划有着重要意义。传统的基于像元的纯光谱柑橘类果林提取方法分类精度较低,很难避免“椒盐现象”,并且受噪声的影响导致同物异谱、异物同谱现象较严重。丰富的纹理信息、形状信息和上下文信息是高分辨率遥感图像突出的特点。本文以赣州市寻乌县柑橘类果林种植区为研究对象,针对传统基于像元分类方法精度较低的缺点,在高分一号数据源的基础上,做了以下研究:1.为了能够让高分一号遥感影像数据既保存影像数据多光谱特性,又提高空间分辨率,对数据进行了大气校正、正射校正、图像融合等图像预处理步骤。2.在多尺度分割的前提下,对象的各种特征信息存在一定的变化规律。利用e Cognition软件统计对象的面积均值、标准差、植被对象的NDVI均值在分割尺度上的变化情况。3.选用最大面积比法来衡量某种特定地物分割对象的优劣程度。4.深入研究了基于模糊数学方法的最近邻分类模型,在深入剖析模型数学表达的基础上,对整幅影像的光谱特征进行了全面的统计分析。5.构建了一种结合模糊数学理论的面向对象决策树分类模型,在多尺度分割建立对象层次组织结构的基础上,综合最近邻分类器,建立了融合模糊数学方法和决策树模型的面向对象决策树分层分类模型,选择样本对象进行统计分析光谱特征和纹理特征,分层使用单波段阈值、NDVI、波段组合、对象几何形态学特征和纹理特征等多种分析经验值,对研究区域进行信息提取,实现了多尺度分割和分层决策树分类,使用混淆矩阵对分类结果进行了精度验证,得到总体Kappa系数0.8643,特别是果林提取精度的明显提高。
[Abstract]:China is one of the main producing and consumer countries of citrus fruit products. It has more than 4000 years of cultivation history. It is of great significance to extract citrus fruit forest quickly and effectively to guide the defense of fruit forest diseases, fruit production and industrial planning. The phenomenon of "salt and pepper" is exempt from the phenomenon of "salt and pepper" and is influenced by the influence of noise. The phenomenon of ISO spectrum is serious. The rich texture information, shape information and context information are the prominent features of high resolution remote sensing images. This paper takes the citrus orchard planting area of Xunwu County, Ganzhou as the research object, and aims at the precision of the traditional pixel based classification method. On the basis of high score 1 data source, the following research is made: 1. in order to enable high score 1 remote sensing image data not only to preserve the multi spectral characteristics of the image data, but also to improve the spatial resolution, the image preprocessing step.2., such as atmospheric correction, orthophoto correction, image fusion, and so on, is on the premise of multiscale segmentation. The information of various features of the image has a certain change law. Using the e Cognition software, the mean area of the area, the standard deviation and the change of the NDVI mean of the vegetation object on the segmentation scale,.3. selects the maximum area ratio method to measure the good and bad degree of the object of certain specific objects,.4. based on the fuzzy mathematical method. The nearest neighbor classification model, based on the deep analysis of the mathematical expression of the model, makes a comprehensive statistical analysis of the spectral features of the entire image..5. constructs an object-oriented decision tree classification model combining the fuzzy mathematics theory. On the basis of the multi-scale segmentation and the establishment of the hierarchical organization structure of the object, the nearest neighbor classifier is built. The object oriented decision tree hierarchical classification model is combined with fuzzy mathematical method and decision tree model. The sample object is selected for statistical analysis of spectral features and texture features. A number of empirical values, such as single band threshold, NDVI, band combination, geometric morphological features and texture features of objects, are used to extract information from the research area. The classification of multi scale segmentation and stratified decision tree was realized. The accuracy of the classification results was verified by using the confusion matrix, and the total Kappa coefficient was 0.8643, especially the extraction precision of the fruit forest was obviously improved.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S666;S127

【参考文献】