基于Hyperion影像的高光谱数据线性解混与目标检测—土地覆被识别实证

发布时间：2018-01-09 13:28

本文关键词：基于Hyperion影像的高光谱数据线性解混与目标检测—土地覆被识别实证　出处：《中国地质大学(北京)》2017年博士论文　论文类型：学位论文

【摘要】：高光谱遥感数据丰富的光谱信息有助于更准确的描述土地覆被的类型,但由于传感器的空间分辨率较低,当以城市、郊区为研究对象时高光谱影像中混合像元的存在不可避免。本文模拟高光谱数据混合光谱的解混,对端元不同所带来的解混误差进行了详细的分析;以北京市为研究区,利用Hyperion高光谱数据,结合V-I-S模型和线性光谱混合分析对北京市区、郊区的土地覆被进行了研究;通过信息检测算法,对特定地物目标的识别进行了实验验证,并给出了改良方法。本文对中等空间分辨率的高光谱影像在城市/郊区遥感方面的利用进行了比较全面的分析。论文研究的主要内容和结论如下:(1)利用光谱库的光谱数据模拟由植被、土壤、不透水层构成的混合光谱,使用不同的端元组合对混合光谱线性解混,提出适用于高光谱数据的复合端元解混方法。在同类地物光谱曲线幅度差异较大的情况下,使用复合端元做全约束的线性解混可以有效减少误差,提高相关系数。(2)基于V-I-S模型,利用单端元组合、复合端元组合对Hyperion高光谱影像进行了线性解混。结果显示,复合端元的组合有更小的误差。从真实的高光谱影像上证明了在对影像覆盖面积较大的高光谱数据做全约束的线性解混时,更宜采用复合端元的组合。(3)在遥感影像的处理中,利用植被指数在城市的不同区域发现了与植被丰度相关性更高的波段组合的特征。从整体上看,更靠近植被光谱“红边”区域的波段计算的NDVI与植被丰度的相关性更显著,植被指数与植被丰度的二次多项式拟合相关性更显著。(4)通过对2004年和2010年Hyperion影像的解混,揭示了北京南、北部区域土地覆被的变化特征。土地覆被的变化显示出2004年~2010年间北京南部和北部区域均有一定程度的植被、土壤向不透水层的转变,越靠近市区变化越明显,在离市区较远的区域不透水层主要沿干道向两侧扩张。(5)利用高光谱数据通过信息检测的方法,对小的目标进行了识别。利用多种信息检测方法对Hyperion数据进行了实验,优化了信息检测输出的方法,并改良了一种多目标识别方法的应用,使其可以更有效的检测Hyperion影像中的目标信号。
[Abstract]:The rich spectral information of hyperspectral remote sensing data is helpful to describe the land cover type more accurately, but because of the low spatial resolution of the sensor, the city should be used. The existence of mixed pixels in hyperspectral images is inevitable when the suburb is the object of study. This paper simulates the demultiplexing of the mixed spectra of hyperspectral data and analyzes in detail the unmixing errors caused by different end-components. Using Hyperion hyperspectral data, V-I-S model and linear spectral mixing analysis were used to study the land cover in the urban and suburban areas of Beijing. Through the information detection algorithm, the recognition of specific objects is verified by experiments. The application of hyperspectral images with medium spatial resolution in urban / suburban remote sensing is analyzed comprehensively. The main contents and conclusions of this paper are as follows: 1). The spectral data of the spectral database are simulated by vegetation. The mixing spectrum of soil, impermeable layer, is linearly unmixed using different end-element combinations. In this paper, a new method of compound end-element demultiplexing for hyperspectral data is proposed. When the amplitude of spectral curves of similar ground objects varies greatly, the linear demultiplexing with composite endpoints as full constraints can effectively reduce the error. Based on the V-I-S model, the Hyperion hyperspectral images are linearly unmixed using single-terminal combination and compound end-element combination. From the real hyperspectral images, it is proved that in the case of fully constrained linear unmixing of hyperspectral data with larger image coverage. In the processing of remote sensing image, vegetation index is used to find the characteristics of band combination with higher correlation with vegetation abundance in different areas of the city. The NDVI calculated in the band closer to the "red edge" region of the vegetation spectrum has a more significant correlation with the vegetation abundance. The correlation between the vegetation index and the quadratic polynomial fitting of vegetation abundance is more significant. Through the unmixing of Hyperion images on 2004 and 2010, this paper reveals the south of Beijing. The change of land cover shows that there was a certain degree of vegetation in the southern and northern regions of Beijing during 2004 and 2010, and the soil changed to impermeable layer. The closer to the urban area, the more obvious the change is, in the area far away from the urban area, the impermeable layer mainly expands along the main road to both sides.) using hyperspectral data through the information detection method. The small target is recognized, and the Hyperion data is tested by many information detection methods, and the method of information detection and output is optimized, and the application of a multi-target recognition method is improved. It can detect the target signal in Hyperion image more effectively.
【学位授予单位】：中国地质大学(北京)
【学位级别】：博士
【学位授予年份】：2017
【分类号】：P237

【相似文献】