黄河三角洲湿地典型植被高光谱遥感研究

发布时间：2018-05-14 23:22

本文选题：高光谱遥感 + 数据挖掘　；参考：《大连海事大学》2014年博士论文

【摘要】：高光谱遥感数据的特征提取和分类工作是高光谱遥感应用领域的研究重点和热点。滨海湿地区域面积广阔、地物分布复杂多样,且高光谱遥感数据本身维度高、数据量大,导致传统的特征提取方法对于蕴含于高光谱数据中的光谱特征信息利用有限,对于专家经验知识和统计信息以外的潜在特征,难以充分提取,进而难以形成适用于滨海湿地高光谱遥感数据的高精度分类算法。这不利于高光谱遥感技术在滨海湿地遥感研究领域的深入发展。本文的工作主要面向滨海湿地植被高光谱遥感精细化监测的需求,以黄河三角洲湿地为研究区域,利用高光谱遥感影像和现场采集的典型植被现场光谱数据,发展了基于数据挖掘技术的滨海湿地典型植被高光谱特征提取和分类方法,实现针对研究区域典型植被与地物类型的高精度提取和分类。具体内容如下： (1)开展了基于光谱可分性与季节光谱特征差异的现场光谱特征分析针对研究区域和高光谱遥感影像的特点,开展黄河三角洲滨海湿地典型植被现场光谱测量,得到了对研究区域植被类型光谱特征代表性良好的现场光谱数据；利用现场实测的典型植被光谱数据,开展基于反射率光谱的植被光谱分析和特征提取。为分析不同植被光谱间的可分性特征,发展了一种基于包络线去除光谱差值的特征波段提取方法,获得了典型植被的波段可分性查找表；为分析不同植被不同季节的光谱特征差异,基于导数变换方法开展典型植被春、秋两季光谱特征分析比较工作,获得了绿光反射峰、红光吸收峰、红边和近红外反射峰等4种光谱特征的位置和反射率差异信息。 (2)发展了基于数据挖掘的研究区域典型植被高光谱遥感特征提取技术针对PROBA CHRIS多视角高光谱遥感卫星影像,首先研究了该影像数据的预处理技术,并对不同视角影像的成像效果及分类能力进行分析研究,确定0°视角影像作为特征提取的数据源。在此基础上,为获取高光谱遥感影像中典型植被的遥感光谱特征和特征波段组合规律,指导研究区域典型植被遥感分类,发展了一种基于关联规则挖掘的滨海湿地典型植被高光谱遥感特征提取技术,利用关联规则挖掘中的广义规则归纳算法,配合关联规则定量指标分析,获得了研究区域8类典型植被与地物类型(包括芦苇、柽柳、碱蓬、大米草、清澈水体、浑浊水体、裸滩、裸地)的高光谱遥感特征集。该技术能够充分提取高光谱遥感数据中的潜在特征,并满足光谱特征在分类、波段和信息等多个维度的独立性要求。 (3)发展了基于决策树的研究区域典型植被高光谱遥感分层分类方法针对覆盖研究区域的PROBA CHRIS高光谱遥感数据,基于本文所建立的典型植被与地物类型高光谱遥感特征集,并结合现场光谱特征波段信息,确定研究区域8类典型植被与地物类型高光谱遥感分类规则,发展了一种基于决策树的黄河三角洲典型植被高光谱遥感分层分类方法。利用该方法对PROBA CHRIS高光谱遥感影像进行分类实验,将整景影像分为8类典型植被与地物类型,利用现场踏勘信息结合高空间分辨率遥感影像解译所获取的标准解译图像,对分类结果进行精度评价,并与基于相同训练样本的SVM分类结果进行对比,实验结果显示,本文所发展的分类方法其分类精度较SVM算法有明显提高,超过10%。
[Abstract]:The feature extraction and classification of hyperspectral remote sensing data is the focal point and hotspot in the field of hyperspectral remote sensing application .

The work of this paper mainly focuses on the demand of high spectral remote sensing precision monitoring of coastal wetland vegetation , uses the Yellow River Delta wetland as the research area , develops the typical vegetation hyperspectral feature extraction and classification method based on data mining technology using hyperspectral remote sensing image and typical vegetation field spectral data collected on site , and realizes high - precision extraction and classification of typical vegetation and figure types in the study area .

( 1 ) On - site spectral characteristic analysis based on spectral variability and seasonal spectral characteristics is carried out .

According to the characteristics of the research area and hyperspectral remote sensing image , the field spectral measurement of typical vegetation in the coastal wetland of the Yellow River Delta was carried out .
The vegetation spectral analysis and feature extraction based on reflectance spectra are carried out by using typical vegetation spectral data measured on site . In order to analyze the variability of vegetation spectra , a feature band extraction method based on envelope removal spectral difference is developed , and the band - separable look - up table of typical vegetation is obtained .
In order to analyze the spectral characteristic difference of different vegetation seasons , the spectral characteristic analysis of typical vegetation spring and autumn is carried out based on derivative transform method , and the position and reflectivity difference information of four spectral features , such as green reflection peak , red absorption peak , red edge and near infrared reflection peak are obtained .

( 2 ) The technology of hyperspectral remote sensing feature extraction based on data mining is developed .

In this paper , a high spectral remote sensing feature set of typical vegetation in coastal wetland , which is based on correlation rule mining , is developed , and the high spectral remote sensing feature set based on correlation rule mining is developed .

( 3 ) The hierarchical classification method of hyperspectral remote sensing based on decision tree is developed .

Based on the hyperspectral remote sensing feature set of typical vegetation and figure types established in this paper , the high spectral remote sensing classification rules of typical vegetation and ground objects in the study area are established based on the high spectral remote sensing feature set of typical vegetation and figure types established in this paper . The classification results are evaluated by using the method . The results show that the classification accuracy of the classification method is obviously improved compared with the SVM algorithm , which is more than 10 % .

【学位授予单位】：大连海事大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP79

【参考文献】