基于HJ-HSI高光谱影像的狼毒遥感提取研究

发布时间：2018-04-14 07:51

本文选题：狼毒 + HH-HSI高光谱影像　；参考：《西北大学》2015年硕士论文

【摘要】：瑞香狼毒是分布在青海省高寒草甸的主要毒害草之一,近年来其迅速蔓延对当地畜牧业危害严重并使草地生态系统日趋退化。应用遥感技术在大范围内及时获取狼毒分布、面积及变化信息,可为狼毒分布监测及危害程度评估提供重要基础,对于维护该地区畜牧业的可持续发展和草地生态系统平衡具有重要意义。本文采用2012年HJ-HSI高光谱数据,以青海省海北州为研究区,开展了狼毒分布遥感提取研究。高光谱影像能够对狼毒光谱特征进行精细探测,有效地克服传统遥感因波段有限导致的识别误差等问题。本研究从光谱波形匹配的角度出发,在对HJ-HSI数据降维去噪的基础上,分别用光谱角制图(SAM, Spectral Angle Mapper)和光谱信息散度(SID, Spectral Information Divergence)两种算法进行狼毒识别提取研究,主要取得了以下成果：1.狼毒处于盛花期时,其独特的物候特征使狼毒光谱反射率与同期牧草的光谱反射率差异极大,具体表现为盛花期狼毒顶花在整个波段的光谱反射率均高于同期牧草,并在900nnm处达到近70%。对狼毒群落而言,这种明显的光谱特征差异使盖度较高的狼毒群落光谱反射率明显区别于牧草地,为狼毒识别的可能性提供依据。在对狼毒群落光谱特征分析的基础上,检验经过大气校正后的HJ-HSI影像狼毒光谱有效性,结果表明,经过大气校正后的HJ-HSI影像能够提供可用的地物光谱信息,并且在近红外波段,影像狼毒群落与牧草群落具有可分性。2.在分析针对高光谱遥感影像各种降维去噪方法优劣性的基础上,最终采用PCA变换(Principal components analysis)和MNF变换(Minimum Noise Fraction Rotation)两种方法对HJ-HSI数据进行降维,结果表明,经过降维后的影像数据消除了相邻波段间的相关性,很好的分离了图像与噪声。然后利用PCA逆变换和MNF逆变换技术,使得到的图像不仅保留光谱波形信息,而且数据量明显压缩,数据噪声明显降低,峰值信噪比高,达到了降维去噪的目的。3.基于光谱波形匹配思想,通过构建狼毒群落与牧草群落的波谱库,以经过降维去噪后的影像数据为输入数据,分别运用SAM和SID两种光谱识别方法进行狼毒识别提取,结果表明,以经过MNF变换后的影像数据为数据源的SAM法狼毒信息提取总体精度最高,达到78.46%,Kappa系数为0.5260,统计实验中狼毒提取面积数据发现,该方法得到的狼毒危害面积与《2012年青海省毒杂草调查》统计的狼毒危害面积吻合,证明了基于MNF变换后的SAM提取技术能够应用于狼毒识别。本研究以HJ-HSI数据为数据源,充分研究了盛花期狼毒群落光谱与牧草光谱特征差异,运用两种方法识别提取研究区狼毒群落,最终证明HJ-HSI数据在狼毒识别提取领域的可用性及潜力,研究结果可为中国西部草原毒害草分布遥感监测和危害程度评估提供技术参考。
[Abstract]:Daphne chamaejasme is one of the main poisonous grass distributed in alpine meadow in Qinghai Province. In recent years, its rapid spread has caused serious harm to local animal husbandry and made grassland ecosystem degenerate day by day.The application of remote sensing technology to obtain the distribution, area and change information of werewolf venom on a large scale in time can provide an important basis for monitoring the distribution of werewolf venom and evaluating its harm degree.It is of great significance to maintain the sustainable development of animal husbandry and balance of grassland ecosystem in this area.In this paper, using HJ-HSI hyperspectral data in 2012 and Haibei state in Qinghai Province as the research area, the remote sensing extraction of the distribution of werewolf venom was carried out.Hyperspectral images can detect the spectral characteristics of werewolf venom and overcome the problem of recognition error caused by the limited band of traditional remote sensing.From the point of view of spectral waveform matching, on the basis of dimension reduction and denoising of HJ-HSI data, two algorithms of spectral angle mapping (SAM, Spectral Angle Mapper) and spectral information divergence (Sid, Spectral Information divergence) are used to extract and extract werewolf venom.The main achievements are as follows: 1: 1.The spectral reflectance of werewolf venom in the flowering stage was greatly different from that of the forage in the same period due to its unique phenological characteristics, which showed that the spectral reflectivity of the top flower in the flowering period was higher than that in the same period.And the 900nnm reached nearly 70.For the community of werewolf venom, the spectral reflectance of the community with high coverage is obviously different from that of the herbage land, which provides the basis for the possibility of the identification of werewolf venom.Based on the analysis of spectral characteristics of werewolf community, the spectral validity of HJ-HSI image after atmospheric correction is tested. The results show that the HJ-HSI image after atmospheric correction can provide the available spectral information of ground objects, and it can be used in near infrared band.The image showed that the community of werewolf venom and the community of forage had separability. 2.On the basis of analyzing the merits and demerits of various dimensionality reduction methods for hyperspectral remote sensing images, PCA transform Principal components Analysis (PCA) and MNF transform minimum Noise Fraction rotation (MNF) are used to reduce the dimension of HJ-HSI data. The results show that,After dimension reduction, the correlation between adjacent bands is eliminated, and the image and noise are separated well.Then, by using PCA inverse transform and MNF inverse transform technology, not only the spectral waveform information is preserved, but also the data volume is obviously compressed, the data noise is obviously reduced, and the peak signal-to-noise ratio is high, which achieves the purpose of dimension reduction and denoising.Based on the theory of spectral waveform matching, the spectral database of werewolf venom community and forage community was constructed, and the image data after dimension reduction and denoising were used as input data, and two spectral recognition methods, SAM and SID, were used to identify and extract werewolf venom.The total precision of MNF method is the highest, and the coefficient of 78.46 kappa is 0.5260. In the statistical experiment, the area data of the extracted area of werewolf venom is found.The results obtained by this method are in agreement with the results of the investigation of poisonous weeds in Qinghai Province in 2012. It is proved that the SAM extraction technique based on MNF transform can be applied to the identification of werewolf venom.In this study, HJ-HSI data was used as the data source to study the spectral characteristics difference between the community spectrum and the forage spectrum in the flowering stage, and two methods were used to identify and extract the community in the study area.Finally, the availability and potential of HJ-HSI data in the field of werewolf venom identification and extraction are proved. The results can provide a technical reference for remote sensing monitoring and hazard assessment of grassland toxic grass distribution in western China.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S812.6;S818.9

【相似文献】