极化干涉SAR图像森林高度估计算法研究

发布时间：2018-04-30 21:39

本文选题：极化干涉SAR + 目标分解　；参考：《哈尔滨工业大学》2014年博士论文

【摘要】：森林是人类赖以生存的重要因素。它们不仅支撑着像自然界气候变化和水循环这样的生态规律，也为人类提供必要的自然产品，例如木材、食物，家畜的饲料和药材等。因此，利用遥感技术对大面积的森林区域进行监测和管理成为了热点问题，而且人们对这一问题进行了大量研究。极化SAR干涉测量是现代遥感技术的一个重要分支。该技术结合了极化SAR和干涉SAR这两种独立的雷达技术。极化干涉SAR (PolInSAR)不仅对散射体的形状和方位不敏感，而且对散射体的位置和空间分布也不敏感。此外，利用PolInSAR图像对森林高度等森林参数进行提取是SAR图像解译和应用的热点，并且在理论上和实际应用上都具有重要意义。为了提高PolInSAR图像森林高度估计的精度，本文研究了基于PolInSAR极化信息与干涉信息的PolInSAR森林参数反演模型, PolInSAR目标分解(TD)和对森林参数的提取。首先，对目标散射特性和典型森林参数的反演模型已经有了很深的研究，例如随机体-地表散射和ESPRIT。在现有森林高度估计方法的理论和应用基础上，本文针对PolInSAR图像，提出了一种提高森里高度估计精度的方法，该方法结合了总最小二乘直线拟合和ESPRIT两种典型的处理方法。并且利用ALOS\PALSAR测量的马来西亚地区L波段全极化干涉数据和PolSARProSim软件仿真的数据对该方法进行了验证。实验结果表明，本文提出的方法可以显著提高森林高度参数的估计准确性. 其次，本文研究了基于散射模型的PolInSAR非相干目标的目标分解方法，提出了一种应用于PolInSAR图像森林高度估计的自适应分解模型(AMBD)。该模型把每种干涉的十字相关表示成奇次散射偶次散射和体散射的总和，不仅可以反演森林参数,而且可以得到每种散射过程的权重。除此之外，该模型还利用了协方差矩阵的所有信息，这是在之前的基于模型的目标分解方法中未能实现的。本文利用SIR-C/X-SAR PolInSAR图像进行了算法验证与性能检验。实验发现，与利用三层反演方法得到的森林高度估计结果比较，这种基于自适应模型目标分解的森林高度估计方法具有更高的精度。再次，本文还提出了基于通用三层散射模型(GTLSM)的PolInSAR图像森林高度估计方法。在GTLSM中，林冠顶层的相关参数可以利用AMBD估计出来，而树干与地表的参数提取则是一个基于非线性组优化的问题。这个模型根据极化特征和相干性的差异，把森林模型分离成三层：地表层,树干层和树冠层。GTLSM同样利用SIR-C/X-SAR的L波段PolInSAR图像进行验证。实验结果表明，本文提出的GTLSM可以更准确的反演森林参数。最后，本文提出了两种从PolInSAR图像中估计斜坡上森林高度的方法。第一种方法是基于一般模型的目标分解方法(GMBD)，在这个方法中，我们提出了一个用随机程度和方位角均值这两个参数描述的一般体散射模型，这两个未知的森林参数可以用非线性最小二乘优化得到。这种方法不仅可以反演森林参数，，也可以反演出每种散射过程的权重，同时还针对十字交叉极化和非对角的情况，通过分离一般奇次散射与偶次散射模型各自的方位角，改进了这两个模型。第二种方法是改进的三层散射模型方(MTLSM)。三层散射模型方法假设在斜坡地势上，可以将森林在垂直方向上分为三层：树冠顶层，树干层和地表层。这三层会同时影响到三种散射过程（体散射,表面散射和偶次散射）对斜坡上森林区域的作用。该方法也介绍了PolInSAR相干信息对森林高度,平均消减，特别是局部地势坡度等参数的影响。第二种方法不仅能够反演出斜坡上森林的参数，而且其对地表相位与树冠层相位的估计具有更强的鲁棒性和明确性。本文将这两种方法应用于ALOS/PALSAR测量的印度尼西亚地区L-波段图像，实验结果显示了这两种方法的有效性。
[Abstract]:Forests are important factors for human survival. They not only support the ecological laws such as natural climate change and water circulation, but also provide the necessary natural products for human beings, such as wood, food, livestock feed and medicinal materials. Therefore, the monitoring and management of large surface forest areas by remote sensing technology has become a hot spot. Problem, and people have done a lot of research on this problem. Polarization SAR interferometry is an important branch of modern remote sensing technology. This technique combines two independent radar technologies, polarizing SAR and interference SAR. Polarization interference SAR (PolInSAR) is not only insensitive to the shape and square of the scatterer, but also the position and space of the scatterer. In addition, the extraction of forest parameters such as forest height, such as the PolInSAR image, is a hot spot in the interpretation and application of SAR images, and it is of great significance both in theory and in practice. In order to improve the accuracy of the estimation of the height of the PolInSAR image forest, this paper studies the Pol based on the PolInSAR polarization information and the interference information. InSAR forest parameter inversion model, PolInSAR target decomposition (TD) and extraction of forest parameters.
First, a deep study of the scattering characteristics of the target and the model of the typical forest parameters has been deeply studied. For example, the theory and application of the random body surface scattering and the theory and application of the method of estimating the height of the forest height by ESPRIT., a method to improve the accuracy of the estimation of the height of the height of the PolInSAR is proposed in this paper. Two typical methods of small second linear fitting and ESPRIT are used. The method is verified by the data of L band full polarization interference data in the L band and the data simulated by PolSARProSim software measured by ALOSPALSAR. The experimental results show that the method proposed in this paper can significantly improve the accuracy of the estimation of the height parameters of the forest.
Secondly, the objective decomposition method of PolInSAR incoherent target based on scattering model is studied. An adaptive decomposition model (AMBD) applied to the estimation of forest height in PolInSAR images is proposed. This model represents the cross correlation of each interference into the summation of odd scattering even scattering and body scattering. It can not only invert the forest parameters. Moreover, the weight of each scattering process can be obtained. In addition, the model also uses all the information of the covariance matrix, which is not realized in the previous model based method of target decomposition. This paper uses the SIR-C/X-SAR PolInSAR image to carry out the algorithm verification and the ability test. Experimental discovery, and the use of three layer inversion method Compared with the results of forest height estimation, the forest height estimation method based on adaptive model decomposition has higher accuracy.
Thirdly, a forest height estimation method for PolInSAR images based on the general three layer scattering model (GTLSM) is proposed. In GTLSM, the correlation parameters of the canopy top layer can be estimated by AMBD, and the parameter extraction of the tree trunk and the surface is a nonlinear group optimization problem. The forest model is separated into three layers: the surface layer, the tree trunk layer and the tree crown.GTLSM also use the SIR-C/X-SAR L band PolInSAR image to verify. The experimental results show that the proposed GTLSM can more accurately retrieve the forest parameters.
Finally, this paper proposes two methods to estimate the height of the forest on the slope from the PolInSAR image. The first method is based on the general model's target decomposition method (GMBD). In this method, we propose a general body scattering model, which is described by the two parameters of the random degree and the azimuth mean, the two unknown forest parameters. This method can be obtained by nonlinear least squares optimization. This method can not only inverse the forest parameters, but also reverse the weight of each scattering process. At the same time, the two models are improved by separating the azimuth of the general odd scattering and the even scattering models for the cross and non diagonal conditions. Second methods are improved. It is an improved three layer scattering model square (MTLSM). The three layer scattering model assumes that in the slope terrain, the forest can be divided into three layers in the vertical direction: the top of the tree crown, the trunk layer and the surface layer. The three layers will simultaneously affect the effect of the three scattering processes (body scattering, surface scattering and even scattering) on the forest area on the slope. The method also introduces the influence of PolInSAR coherent information on forest height, average reduction, especially local terrain slope. The second methods can not only reverse the parameters of the forest on the slope, but also have stronger robustness and clarity to the estimation of the surface phase and the phase of the tree crown. In this paper, these two methods are applied in the paper. LSAR measured L- band images in Indonesia area. The experimental results show the effectiveness of the two methods.

【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN957.52

【参考文献】