【作者】 贾明权；

【导师】 童玲；

【作者基本信息】 电子科技大学 ， 检测技术与自动化装置， 2013， 博士

【摘要】 雷达遥感具有全天候、全天时和地物穿透性的特点，适用于多云、多雾、多雨地区地物目标的快速、宏观、定量探测；但是，由于不同地物尤其是植被微波散射机理的复杂性和相关理论研究的滞后性，严重阻碍了雷达遥感巨大的应用潜力。近年来，随着星载SAR观测平台的迅速发展，为了更高效、精确地挖掘雷达遥感数据的应用潜力，迫切需要研究地物的散射机理和参数定量反演算法。本论文以星载SAR和陆基散射计同步观测为数据基础，以正演、反演算法研究为理论依据，详细分析了水稻微波散射特性和参数敏感性，建立了水稻散射经验、半经验和理论散射模型，研究了以散射模型为基础的神经网络等反演算法，实现了SAR图像水稻覆盖区域制图和生物量反演。本论文的主要工作概括如下：（1）建立了具有不同频率、不同角度和不同极化测量能力的微波散射测量系统，研究了天线非平面波的近场散射效应和电磁波多路径叠加效应抑制技术，消除了收发天线间因通道不平衡和天线串扰带来的失真矩阵等误差。实现了充分的独立取样和定标，确保了陆基散射计测量的精度。（2）完成了2010和2012年两个水稻季的8个不同生长期的散射测量实验。包括：C波段、全极化（HH、HV、VH和VV）和不同入射角（0°-90°）的后向散射系数测量；水稻生物量、高度、LAI、密度、叶片和茎杆参数、下垫面淹水或土壤参数等稻田参数的获取。以水稻生长参数实验数据为基础，建立了水稻生长模型，并对模型的有效性和合理性进行了验证；分析了水稻的散射特性，包括：不同生长期水稻入射角散射特征，水稻时域散射特征，以及水稻散射值与生长参数的相关性特征。（3）建立了水稻经验、半经验和理论微波后向散射模型。根据不同输入参数，提出了多参数非线性建模的思路，建立了水稻的多生长参数经验模型。根据散射项的不同，分别建立了水稻的水云模型、改进的水云模型和简化MIMICS模型，并利用实测数据确立了模型的经验参数，对模型分析对比。根据蒙特卡洛方法建立水稻理论微波散射模型，针对水稻结构特征对模型进行了修正，并利用水稻生长模型提供的输出数据进行了仿真，对比模拟和实测数据，验证了模型的准确性。利用建立的理论模型分析了后向散射系数对水稻主要参数的灵敏性。（4）根据建立的水稻散射模型，发展了水稻参数经验、半经验和基于Monte-Carlo理论模型的神经网络反演算法，建立了分别针对单极化、双极化和全极化的反演模型。利用水稻散射实验测量数据，分别确立了水稻经验反演模型和不同极化的半经验模型参数，半经验模型包括水云反演算法、改进的水云反演算法和简化的MIMICS反演算法，并比较了算法的优劣性。研究了BP神经网络的设置、训练数据的生成、网络训练精度验证，散射实验测量数据验证了基于Monte-Carlo理论模型的神经网络反演算法的精度。（5）将建立的基于简化的MIMICS半经验模型的反演算法和基于Monte-Carlo模拟的神经网络反演算法，用于双极化的ASAR图像和全极化的RADARSAT-2图像上，实现了水稻生物量的反演和验证。其中，双极化ASAR图像水稻生物量反演部分，结合光学TM图像首先实现了水稻区域制图和SAR图像后向散射系数的提取，利用半经验反演算法得到了不同时期的水稻生物量分布，并结合地面实验的实测数据对反演结果进行了验证。全极化RADARSAT-2图像水稻生物量反演部分，给出了生物量反演流程，将水稻散射测量实验、生长模型和Monte-Carlo散射模型的建立、神经网络的训练、多时相RADARSAT-2图像的处理和生物量反演相互关联起来，利用多时相数据分类实现了RADARSAT-2图像上水稻区域的制图，进而利用训练好的神经网络实现了生物量的反演和验证，将反演算法推广到了更大观测区域，实现了大面积水稻的参数反演和长势监测。水稻生长参数与后向散射系数之间存在复杂的非线性关系，从有限的雷达观测数据中提取这些参数是一个典型的病态反演问题，难以获得精确量化结果。本文提出以水稻为对象的微波散射特性和参数反演算法研究，不但丰富了植被散射机理的理论和实验研究，也将推动SAR图像植被参数定量反演算法的研究，拓宽雷达遥感技术的应用研究领域，对实现基于雷达遥感的农作物长势监测及估产，乃至植被生态环境的监测具有重要的学科意义和巨大的潜在经济价值。更多还原

【Abstract】 Radar remote sensing, which has characteristics as penetration and is available inall weather and all time, can be applied to detecting cloudy, foggy, rainy areas in aquick, macro and quantitative method. However, since microwave scatteringmechanism of the different surface features especially vegetation is complicated, whichmade the theoretical research lack of support; it is serious impediment to the radarremote sensing application potentiality. In recent years, several space-borne SARobservation platforms were constructed quickly, so it is possible to exploit moreefficient and accurate potential applications by studying the scattering mechanism of thesurface features and the quantitative inversion algorithm of the parameters. In thedissertation, with the theoretical basis of modeling and inversion algorithms, microwavescattering properties of rice, parameters and sensitivity was analyzed in detail, and thenempirical, semi-empirical and theoretical scattering model of rice were establishedbased on synchronization observations of space-borne SAR and ground-basedscatterometer data. Moreover, focus on neural network inversion algorithm based on thescattering model; the rice coverage area mapping and biomass inversion were achievedfrom SAR images. The main work of the dissertation is summarized as follows:(1) A microwave scattering measurement system was constructed with differentfrequencies, different angles and different polarization measurement capability. Thedistortion matrix of the transceiver antennas caused by the unbalanced channel andantenna crosstalk and other errors were eliminated after research on the non-planarwave antenna near-field electromagnetic wave scattering and suppression technology ofmulti-path superimposed effect. Enough independent sampling and precise calibrationguaranteed accurate measurement of the land-based scatterometer.(2) Eight different scattering measurements experiments were executed during tworice growing seasons (2010and2012). The measured C-band backscatteringcoefficients included that of full polarization (HH, HV, VH and VV) and differentincident angles (0°-90°). Sampling parameters of rice paddies included rice biomass,height, LAI, density, leaf and stem parameters, underlying surface parameters,waterlogged or soil parameters. The rice growth model was established and verified with the rice growth parameters, and the results were effective and reasonable. Thescattering properties of rice were analyzed, which included scattering characteristics ofangle at different rice growth periods, scattering characteristics of rice time domain, andthe correlation coefficient between scattering coefficients and the rice growthparameters.(3) The empirical, semi-empirical and theoretical microwave backscatteringmodels of rice were established. Based on different input parameters, the idea ofmulti-parameter nonlinear modeling was proposed, and then a multi rice growthparameters empirical model was established. According to the different scatteringmechanism, three semi-empirical model of rice, WC (Water Cloudy) model, theimproved WC model and the simplified MIMICS were established. These modelsparameters were acquired from the measured data and were compared and analyzedtheir accuracy separately. With Monte-Carlo method, the theory of microwavescattering model of rice was constructed and the model was modified with ricestructural characteristics. The rice growth model was used to provide input data for themodel simulation, and then the accuracy of the model was verified by comparing thesimulated data with that measured. The sensitivity between backscattering coefficientand the main rice parameters was analyzed by using the theoretical model.(4) Inversion algorithms of rice parameters were established according to the ricescattering model, including empirical, semi-empirical and neural network, which werebuilt based on Monte-Carlo theoretical models for single, dual and full polarizationsdata, respectively. The parameters of rice empirical inversion model and semi-empiricalinversion model at different polarization were obtained by using the measured data fromrice scattering experiments. The advantages and disadvantages of WC, improved WCand simplified MIMICS inversion models were compared and analyzed. The BP neuralnetwork (NN) was studied, including parameter settings of NN, the training datageneration, network training accuracy verification. The measured scattering dataverified the accuracy of NN inversion model based on Monte-Carlo model.(5) The semi-empirical model as simplified MIMICS and NN inversion algorithmbased on Monte-Carlo model were used to inverse rice biomass from dual polarizationASAR images and full polarization RADARSAT-2images, respectively. The ricemappings were achieved by combining the dual polarization ASAR images with opticalTM images, and then backscattering coefficients of rice area were extracted from SAR images. The rice biomass distribution of different periods was inversed by using thesemi-empirical inversion algorithm, and the inversion results were verified with theground measured data. Biomass inversion process of full polarization RADARSAT-2image was achieved, which involved the rice scattering measurement experiments,growth model and the Monte-Carlo scattering model, neural network training,multi-temporal RADARSAT-2image processing and biomass inversion. Rice areamappings were obtained by using the multi-temporal RADARSAT-2images, and ricebiomasses were inversed and validated by using the trained neural network andmeasured biomass data. The inversion algorithm was extended to the larger observationarea, and the rice biomass inversion of large area and growth monitoring was achieved.The complex nonlinear relationship between rice growth parameters andbackscattering coefficient is a typical ill-posed inversion problem, when the limitedradar data were used to extract these rice parameters, which is difficult to obtainaccurate quantitative results. The dissertation studied rice microwave scatteringcharacteristics and parameters inversion algorithm, which can enrich theoretical andexperimental research of the vegetation scattering mechanisms, and promote the SARimage research of quantification inversion algorithm of vegetation parameters. theresearch also expands the application field of radar remote sensing technology, and isproved to own enormous potential economic values for crop growth monitoring,yieldestimation, and even vegetation ecological environment monitoring.更多还原