【作者】 孙波；

【导师】 梁勇；

【作者基本信息】 山东农业大学 ， 农业电气化与自动化， 2020， 博士

【摘要】 土壤湿度是影响全球水循环的重要变量,是衡量地面与大气之间能量交换的主要因素。能够及时、可靠地获取农田土壤湿度值对于在农业生产中进行精准灌溉、减少水资源浪费、降低生产成本和提高农作物产量都是非常必要的。本论文依托国家863课题《多尺度农田信息获取与融合技术》、国家自然科学基金面上项目《数据融合的GNSS-IR农田土壤湿度反演方法研究》和北京航空航天大学横向课题《北斗土壤遥感系统试验基地建设》等课题,从电磁波反射分析、信号处理、反演模型的构建、多星多频数据的融合、外场试验的设计及试验数据验证几个方面展开了地基GNSS-IR土壤湿度反演方法研究。构建了基于GA-SVM的单天线土壤湿度反演模型;构建了单天线模式下北斗系统基于单星双频段熵融合的土壤湿度反演模型以及双频载波相位融合的土壤湿度反演模型;提出了基于GPS系统多星多频段观测量的自适应加权融合算法,构建了基于多星多频自适应加权融合的土壤湿度反演模型。最后对各模型进行了性能的横向对比并评估了其适用场景。本文研究的主要内容和结论如下:(1)分析了GNSS地基平台下土壤湿度的遥感方法,探讨了数据处理流程。首先从GNSS信号源的角度对四大导航系统的系统组成、星座结构、信号特点进行了分析对比。然后详细的分析了电磁波信号的极化方式、反射信号的几何关系,给出了反射信号的数学表达式以及镜面反射点、第一菲涅尔反射区的概念等跟反射区域有关的定义和计算方法。最后总结了近年来应用全球导航卫星反射信号进行土壤湿度遥感的方法——采用直、反射信号相关功率比的双天线模式和利用直、反射信号干涉现象的单天线模式,重点阐述了单天线模式反演的基本原理,给出了两种模式的数据处理流程,并对比了两种工作模式的工作特点和区别。(2)研究了基于单星单频观测量的GNSS单天线土壤湿度反演方法,构建了裸土条件下GNSS单天线的土壤湿度一元线性回归反演模型和基于GA-SVM的单天线土壤湿度反演模型。开展了裸土条件下GNSS单天线土壤湿度反演模型研究,从理论层面详细分析了“信噪比SNR数据—多径分量—土壤湿度”这一建模过程中直射分量的拟合、反射分量的计算、干涉特征观测量的获取等问题,给出了适用于本文条件下的GNSS单天线的裸土土壤湿度一元线性回归反演模型,并开展了相关试验对所提方法进行验证。从一元回归模型反演结果上可以看出:振荡频率、振荡幅度和初始相位三个干涉特征参量与土壤湿度均具有较好的相关性,土壤湿度反演值与土壤湿度实测值的回归相关系数R达到了0.7861-0.9353,均方根误差RMSE为0.593 cm~3/cm~3-0.841cm~3/cm~3。为了抑制植被和土壤粗糙度所引起的噪声、提高拟合精度,构建了基于GA-SVM的单天线土壤湿度反演模型,选择普适性好的径向基核函数。针对SVM模型参数人为调整的不确定性,选用遗传算法对SVM模型参数进行优化,并通过试验对模型进行了验证。结果表明:单星的GPS PRN 12 L1频段GA-SVM的反演结果与土壤湿度实测值的回归相关系数R达到了0.9782,均方根误差为0.182 cm~3/cm~3。对比由三种干涉特征参量分别构建的一元线性回归模型,相关系数R提高了4.59%-24.44%,均方根误差RMSE减小了69.3%-78.36%。为了进一步证明本文所提模型的有效性,在相同数据集配置下,分别与PSO-SVM和BP神经网络的机器学习方法相比,GA-SVM模型均方根误差RMSE分别降低了68.57%及84.09%,证明了GA-SVM模型可以有效的提高反演精度。(3)研究了基于单星双频数据融合的GNSS单天线土壤湿度反演方法,从信息熵的角度为了融合GNSS反射信号中不同频段信号所含的不同土壤信息,构建了单天线模式下北斗系统基于单星双频段熵融合的土壤湿度反演模型,给出了该模型的数据处理流程,并在北京通州开展了地基试验对反演方法进行验证。试验结果表明:北斗PRN 9、PRN 10、PRN 13三颗卫星熵融合后的振荡频率观测量与土壤湿度的回归相关系数分别达到了0.8118、0.8924、0.8609;均方根误差分别为2.073 cm~3/cm~3、1.689cm~3/cm~3、1.814 cm~3/cm~3。PRN 9的双频熵融合的反演结果相比于B1、B2频段振荡频率的传统回归方法,相关系数R提高了24.41%-37.11%,均方根误差RMSE降低了3.4%-8.48%;PRN 10反演结果相比于传统回归方法,相关系数R提高了35.64%-48.07%,均方根误差RMSE降低了10.82%-40.4%;PRN 13反演结果相比于传统回归方法,相关系数R提高了37.59%-44.62%,均方根误差RMSE降低了11.56%-16.67%。其次,考虑到部分GPS、北斗接收机不能提供SNR数据的问题,同时进一步丰富利用反射信号测量土壤湿度的手段,从信号的层面提出了单天线模式下基于北斗系统单星双频载波相位融合的土壤湿度反演模型,通过双频载波相位组合的方式消除载波相位中的几何信息及对流层延迟误差,对数据处理的流程进行了分析,最后利用通州试验数据进行了验证。试验数据表明:北斗PRN 9、PRN 10、PRN 13三颗卫星载波相位组合后的振荡主频观测量与土壤湿度的回归相关系数分别达到了0.5641、0.6122、0.5796;均方根误差分别为2.269 cm~3/cm~3、1.722 cm~3/cm~3、2.897 cm~3/cm~3,证明载波相位融合方法应用于土壤湿度反演是有效的,但反演精度还需要进一步提高。(4)研究了基于多星多频数据融合的GNSS单天线土壤湿度反演方法,从信息的层面为了充分融合不同轨道、不同频段的卫星观测信息,提出了基于GPS系统多星多频段观测量的自适应加权融合算法,构建了基于多星多频自适应加权融合的土壤湿度反演模型,分析了数据的处理流程,并通过试验数据进行了模型验证。试验数据表明:GPS PRN 1、PRN 6、PRN 8三颗卫星L1、L2、L5频段的振荡幅度观测量加权融合后的振幅观测量与土壤湿度的相关系数R达到了0.8059,均方根误差RMSE为2.075cm~3/cm~3。三颗星三频融合方法的反演结果相比于单星单频的传统回归方法,相关系数R提高了24.69%-79.21%,均方根误差RMSE降低了22.28%-33.58%;与所有频段振幅观测量取均值的均值融合法相比,相关系数R提高了26.77%,均方根误差RMSE降低了23.26%,证明了基于多星多频数据融合模型的有效性。最后,从反演精度、算法复杂度等评价指标对本文所提的反演模型进行了综合评价,并分析了各种模型的适用条件和场景。其中,与传统一元线性回归相比,GA-SVM模型反演精度提高最大,但是其算法复杂度和对硬件的要求较高;双频熵融合模型则在算法复杂度和反演精度两方面取得良好折衷;双频载波相位融合模型在低复杂度的优点下,可以充分利用无法存储SNR的早期GPS站点,扩大了全球观测的网络范围;多星多频融合模型在明显提高反演精度的同时,增大了观测范围,反演值更能代表探测区域土壤湿度的平均水平。在今后的具体应用中,还应根据需求和资源进行模型的选择。更多还原

【Abstract】 Soil moisture is an important variable of global water cycle and a key parameter to quantify the energy exchange between land and atmosphere.Timely and accurate acquisition of soil moisture data is very important for rational irrigation in agricultural production to reduce waste of water resources for achieving low cost and high yield in agricultural.Research activities of this dissertation are funding by the national 863 project"Multi-scale cropland information acquisition and fusion technology",general program of national natural science foundation of China"Cropland soil moisture inversion method research based on GNSS-IR data fusion",and the Beihang university crosswise task"Experimental base construction of Beidou soil remote sensing system",etc.This dissertation carries out the research on GNSS-IR soil moisture inversion method from the aspects of electromagnetic wave reflection analysis,signal processing,inversion model building,multi-satellite and multi-frequency data fusion algorithm,field experiment design and experimental data verification.Moreover,this dissertation propose a GA-SVM model of bare soil moisture based on GNSS single antenna,and constructs the soil moisture inversion model of single-satellite and dual-frequency band of Beidou system based on the entropy fusion and the carrier phase fusion respectively.Furthermore,an adaptive weighting fusion algorithm based on GPS multi-satellite and multi-frequency observations is proposed,and a soil moisture inversion model based on this algorithm is built.Finally,the performance of each model is compared and the applicable scenarios are evaluated.The main contents and conclusions of this research are as follows:(1)The principle of in-situ GNSS-IR soil moisture remote sensing is analyzed,and the data processing flows are discussed.Firstly,the system composition,constellation structure and signal characteristics of the four global navigation satellites systems are analyzed and compared from the signal source viewpoint.Then,the polarization mode and geometry of GNSS signal are analyzed in detail,and the mathematical expression of reflection signal is given,including the definition and calculation method of specular reflection point,first Fresnel reflection region and other related reflection region.Finally,this dissertation summed up the current two approaches of GNSS-R soil moisture inversion,as well as the dual antenna mode which utilizes the direct and reflected signal correlation power ratio,and the single antenna mode which utilizes the interference of direct and reflected signals.The basic principle of soil moisture inversion based on single antenna pattern is emphatically analyzed and the data processing flow is given.(2)The soil moisture inversion method of GNSS single antenna based on single-satellite and single-frequency observations is studied,and the unitary linear regression inversion model and GA-SVM model of bare soil moisture based on GNSS single antenna is presented respectively.Study of GNSS single antenna soil moisture inversion model research under the condition of bare soil is carried out.The method of deprive direct from SNR data to get the multipath component which can be used to build soil moisture model is studied theoretically.and a unitary linear regression inversion model of bare soil moisture based on GNSS single antenna technology is presented.The in-situ experimental campaign is performed for verification.From the inversion results of the unary regression model,it can be seen that the three interference characteristic parameters of oscillation frequency,oscillation amplitude and initial phase all have a good correlation with soil moisture.The regression correlation coefficient R of soil moisture inversion value and measured value reaches 0.7861-0.9353,and the root-mean-square error RMSE is 0.593 cm~3/cm~3-0.841 cm~3/cm~3.For the purpose of suppressing the noise caused by soil roughness and vegetation and improving the inversion accuracy,a single antenna soil moisture inversion model based on GA-SVM is proposed.In this model,radial basis kernel function is chosen for its good universality,and genetic algorithm introduced for SVM parameter optimization to avoid the uncertainty caused by artificial adjustment.The results show that the regression correlation coefficient R between the GA-SVM inversion results of single satellite GPS PRN 12 L1 band and the measured values of soil moisture reaches 0.9782,and the root-mean-square error is0.182 cm~3/cm~3.Comparing with the unary linear regression model constructed by three interference characteristic parameters respectively,the correlation coefficient R increases by4.59%-24.44%,and the root-mean-square error RMSE decreases by 69.3%-78.36%.RMSE of the GA-SVM model is reduced by 68.57%and 84.09%respectively compared with other machine learning methods such as PSO-SVM and BP neural network under the same data set configuration,which proves that the GA-SVM model can effectively improve the inversion accuracy.(3)The soil moisture inversion method of GNSS single antenna based on single-satellite and dual-frequency data fusion is studied,in order to fuse different soil information contained in different frequency band signals of GNSS reflection signals from the perspective of information entropy,the soil moisture inversion model under the single antenna mode of Beidou system based on single-satellite and dual-frequency entropy fusion is constructed,the data processing procedure of the model is given,and an in-situ experimental campaign has been carried out in Beijing Tongzhou for verification.The regression correlation coefficients between the soil moisture and the entropy fusion oscillation frequency of Beidou satellites PRN 9,PRN 10 and PRN 13 reaches 0.8118,0.8924 and 0.8609 respectively,and the root-mean-square error is 2.073 cm~3/cm~3,1.689 cm~3/cm~3 and 1.814 cm~3/cm~3 respectively.Comparing with the traditional regression method of oscillation frequency in B1 and B2frequency bands,the correlation coefficient R of PRN 9 is increased by 24.41%-37.11%,and the root-mean-square error RMSE is reduced by 3.4%-8.48%.PRN 10 inversion results show that the correlation coefficient R increases by 35.64%-48.07%and the RMSE decreases by10.82%-40.4%.PRN 13 inversion results show that the correlation coefficient R increases by37.59%-44.62%and the root-mean square error RMSE decreases by 11.56%-16.67%.Secondly,considering some of the GPS and Beidou receivers cannot provide SNR data,also to further enrich the methods of measuring soil moisture using the reflection signal,from the signal level,the soil moisture inversion model under the single antenna mode of Beidou system based on single-satellite and dual-frequency carrier phase fusion model is proposed,which eliminates the geometric information and tropospheric delay errors in carrier phase by dual frequency carrier phase combination method,.Then the data processing flow of the model is given.The results on Tongzhou experimental data show that the regression correlation coefficients of the soil moisture and the main oscillation frequency after the carrier phase fusion of Beidou PRN 9,PRN 10 and PRN 13 is 0.5641,0.6122 and 0.5796respectively,the root-mean-square error is 2.269 cm~3/cm~3,1.722 cm~3/cm~3 and 2.897 cm~3/cm~3respectively,which proves that the carrier phase fusion method is effective in soil moisture inversion,but the inversion accuracy still needs to be further improved.(4)The soil moisture inversion method of GNSS single antenna based on multi-satellite and multi-frequency data fusion is studied,from the information level viewpoint,in order to fully fusion the satellite’s observation information of different orbits and different frequency bands,an adaptive weighted fusion algorithm based on multi-satellite and multi-frequency observations of GPS system is proposed,and the soil moisture inversion model based on multi-satellite and multi-frequency adaptive weighted fusion is constructed,data processing flow is introduced,and the model is verified by Tongzhou experimental data.The results show that the correlation coefficient R between the soil moisture and the amplitude observations of L1,L2 and L5 frequencies of GPS satellites PRN 1,PRN 6 and PRN 8 after weighted fusion is 0.8059,and the root-mean-square error RMSE is 2.075 cm~3/cm~3.Comparing with the traditional single satellite and single frequency regression method,the inversion results of the triple-satellite and triple-frequency fusion method show that the correlation coefficient R is increased by 24.69%-79.21%and the root-mean-square error RMSE is reduced by 22.28%-33.58%.Comparing with the mean fusion method which averaging amplitude observations in all frequency bands,the correlation coefficient R increases by 26.77%and the root-mean-square error RMSE decreases by 23.26%,the validity of the multi-satellite and multi-frequency data fusion model is proved.Finally,comprehensive evaluation and comparison of the four proposed inversion model is carried out based on the evaluation indexes such as inversion precision and algorithm complexity,and the applicable condition and scenario of each model is analyzed.Comparing with the traditional unary linear regression,the inversion precision of GA-SVM model improves the most,but its algorithm complexity and hardware requirements are higher.Furthermore,the dual frequency entropy fusion model has a good compromise between algorithm complexity and inversion precision.Moreover,the dual frequency carrier phase fusion model can make full use of the early GPS stations that cannot store SNR to expand the scope of global observation network with low complexity.Finally,the multi-satellite and multi-frequency fusion model improves both the inversion precision and observation range obviously,and the inversion value represents the average level of soil moisture in the detection area better.In the future,the model should be selected according to the demands and resources.更多还原