基于全极化SAR图像的植被生物量信息提取技术研究
发布时间：2018-12-18 09:58
【摘要】：作为地球生态系统的重要组成部分,植被与人类活动息息相关。植被信息提取对于监测环境变化、农业发展以及地质灾害预测等方面具有重要意义。特别在山区,植被信息可以作为预测塌方、泥石流等灾害的重要指标。传统获取植被信息的方法非常有限,主要通过人工地表实测以及光学遥感两种方法。人工地表实测通常难以实施并且无法获取大尺度数据;光学遥感易受天气影响,在灾害多发的雨季无法获得遥感图像。SAR技术具有不受云、雾、雨影响的特点,能够全天候监测地表信息。因为微波具有穿透性,对植被与其他地物具有良好的区分性,并且更容易反演植被生物量信息。本文内容是基于全极化SAR图像的植被生物量信息提取技术研究,主要包括目标极化分解方法研究、植被覆盖信息提取技术研究以及植被生物量信息提取三部分内容,具体如下:(1)Cloude分解对于后散散射系数较低的平面散射机制区域会错误地分解为高熵体散射机制,对于这个现象,本文给出了解释:这是由于后散散射系数较低区域信噪比较低,造成各极化通道后散散射系数差异性降低,从而表现为高熵体散射机制。(2)本文针对Yamaguchi分解受地形影响问题,使用极化方位角补偿方法,降低了地形对分解方法的影响。(3)本文提出结合最大类间方差法的H/α-Wishart分类方法,有效改善水体、道路以及阴影错分为植被的问题,同时分类结果比传统H/α-Wishart分类方法保留更多细节。(4)本文提出基于区域的Yamaguchi-SVM分类方法,改善了传统Yamaguchi-SVM分类方法对于复杂地形区域分类结果零散点较多情况,并把分类准确率从62.4%提高到71.3%。(5)本文对邛崃和昭觉两个区域实现了完整的基于全极化SAR图像植被生物量信息提取流程。邛崃研究区测量值与反演均值的均方误差为0.6622kg/m,相关系数为0.893;昭觉研究区实验点一有75.0%的像元的反演值在实验范围内的实验点为83.3%,两个实验点的最大值偏移于1.62kg/m以内。
[Abstract]:As an important part of the earth's ecosystem, vegetation is closely related to human activities. Vegetation information extraction plays an important role in monitoring environmental change, agricultural development and geological hazard prediction. Especially in mountainous areas, vegetation information can be used as an important index to predict landslides, debris flows and other disasters. The traditional methods of obtaining vegetation information are very limited, mainly through artificial surface measurement and optical remote sensing. Artificial surface measurement is usually difficult to carry out and can not obtain large-scale data. Optical remote sensing is easy to be affected by weather, and remote sensing images can not be obtained in rainy season. SAR technology is not affected by cloud, fog and rain, and can monitor surface information all the time. Because microwave is penetrating, it is better to distinguish vegetation from other ground objects, and it is easier to retrieve vegetation biomass information. The content of this paper is the research of vegetation biomass information extraction technology based on fully polarized SAR image, which includes three parts: target polarization decomposition method, vegetation cover information extraction technology and vegetation biomass information extraction. The main results are as follows: (1) the Cloude decomposition can misrepresent the plane scattering mechanism region with low backscattering coefficient into a high entropy volume scattering mechanism. The explanation given in this paper is that the difference of backscattering coefficient is reduced due to the low signal-to-noise ratio in the lower region of the backscattering coefficient. Therefore, the scattering mechanism of high entropy volume is presented. (2) in this paper, the polarization azimuth compensation method is used to solve the problem that Yamaguchi decomposition is affected by topography. The influence of terrain on decomposition method is reduced. (3) in this paper, the H / 伪-Wishart classification method combined with the maximum inter-class variance method is proposed to effectively improve the classification of water bodies, roads and shadows into vegetation. At the same time, the classification results retain more details than the traditional H- 伪-Wishart classification method. (4) this paper proposes a region-based Yamaguchi-SVM classification method, which improves the situation that the traditional Yamaguchi-SVM classification method has more scattered points for the classification results of complex terrain regions. The classification accuracy is improved from 62.4% to 71.3%. (5) the extraction process of vegetation biomass information based on fully polarized SAR images is realized for Qionglai and Zhaojue regions. The mean square error between the measured value and the inversion mean in Qionglai study area is 0.6622 kg / m, and the correlation coefficient is 0.893; In Zhaojue research area, 75.0% of the pixel inversion value is 83.3% in the experimental range, and the maximum value of the two experimental points is less than 1.62kg/m.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52
本文编号：2385684
[Abstract]:As an important part of the earth's ecosystem, vegetation is closely related to human activities. Vegetation information extraction plays an important role in monitoring environmental change, agricultural development and geological hazard prediction. Especially in mountainous areas, vegetation information can be used as an important index to predict landslides, debris flows and other disasters. The traditional methods of obtaining vegetation information are very limited, mainly through artificial surface measurement and optical remote sensing. Artificial surface measurement is usually difficult to carry out and can not obtain large-scale data. Optical remote sensing is easy to be affected by weather, and remote sensing images can not be obtained in rainy season. SAR technology is not affected by cloud, fog and rain, and can monitor surface information all the time. Because microwave is penetrating, it is better to distinguish vegetation from other ground objects, and it is easier to retrieve vegetation biomass information. The content of this paper is the research of vegetation biomass information extraction technology based on fully polarized SAR image, which includes three parts: target polarization decomposition method, vegetation cover information extraction technology and vegetation biomass information extraction. The main results are as follows: (1) the Cloude decomposition can misrepresent the plane scattering mechanism region with low backscattering coefficient into a high entropy volume scattering mechanism. The explanation given in this paper is that the difference of backscattering coefficient is reduced due to the low signal-to-noise ratio in the lower region of the backscattering coefficient. Therefore, the scattering mechanism of high entropy volume is presented. (2) in this paper, the polarization azimuth compensation method is used to solve the problem that Yamaguchi decomposition is affected by topography. The influence of terrain on decomposition method is reduced. (3) in this paper, the H / 伪-Wishart classification method combined with the maximum inter-class variance method is proposed to effectively improve the classification of water bodies, roads and shadows into vegetation. At the same time, the classification results retain more details than the traditional H- 伪-Wishart classification method. (4) this paper proposes a region-based Yamaguchi-SVM classification method, which improves the situation that the traditional Yamaguchi-SVM classification method has more scattered points for the classification results of complex terrain regions. The classification accuracy is improved from 62.4% to 71.3%. (5) the extraction process of vegetation biomass information based on fully polarized SAR images is realized for Qionglai and Zhaojue regions. The mean square error between the measured value and the inversion mean in Qionglai study area is 0.6622 kg / m, and the correlation coefficient is 0.893; In Zhaojue research area, 75.0% of the pixel inversion value is 83.3% in the experimental range, and the maximum value of the two experimental points is less than 1.62kg/m.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52
【参考文献】
相关期刊论文 前4条
1 付海强;汪长城;朱建军;解清华;赵蓉;;Neumann分解理论在极化SAR植被分类中的应用[J];武汉大学学报(信息科学版);2015年05期
2 李坤;邵芸;张风丽;;基于RadarSat-2全极化数据的水稻识别[J];遥感技术与应用;2012年01期
3 范立生;高明星;杨健;彭应宁;;极化SAR遥感中森林特征的提取[J];电波科学学报;2005年05期
4 韩春明,郭华东,王长林;SAR图像斑点噪声抑制的本质[J];遥感学报;2002年06期
相关博士学位论文 前3条
1 贾明权;水稻微波散射特性研究及参数反演[D];电子科技大学;2013年
2 徐星欧;基于目标分解的全极化雷达数据估算生物量相关参数方法研究[D];武汉大学;2010年
3 曹芳;基于Cloude-Pottier分解的全极化SAR数据非监督分类的算法和实验研究[D];中国科学院研究生院（电子学研究所）;2007年
相关硕士学位论文 前2条
1 胡昊;极化SAR图像精细地物分类方法研究与实现[D];上海交通大学;2012年
2 郑波浪;机载高分辨率合成孔径雷达运动补偿研究[D];中国科学院研究生院（电子学研究所）;2006年
，本文编号：2385684
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2385684.html
