基于激光诱导击穿光谱技术的土壤理化信息检测方法研究

发布时间：2018-06-11 20:51

  本文选题：激光诱导击穿光谱技术 + 土壤　； 参考：《浙江大学》2016年博士论文

【摘要】：数字农业是实现农业精准化管理和科学化种植的一条重要途径,是现代农业最前沿的发展领域之一,也是当今农业高效、生态、安全和可持续发展的关键和核心。数字化精准农业的实施中最为基本和关键的因素是农作物-环境信息的准确感知、快速获取和智能控制。数字化和信息化技术可为农业绿色生产和高效管理提供快速、准确的信息获取、科学的辅助决策和高效的作业控制,已成为农业科技领域研究的热点。土壤作为人类赖以生存的重要自然资源之一,是农业生产的基础和根本所在,对土壤的信息获取和检测的技术和方法研究是农业环境信息领域的热点。对土壤类型的分析研究可以为建立土壤的肥力和质量评价系统,为土壤的整治、规划和合理利用提供科学依据；土壤的元素信息的检测能够为农业田间作物营养诊断,农田信息实时获取和科学的肥水管理奠定理论基础；对土壤的重金属检测可以有效防止农田的重金属污染,为农业的高品质安全生产提供理论指导作用。本研究在系统深入了解激光诱导击穿光谱(Laser-induced breakdown spectroscopy, LIBS)技术的工作过程和原理基础,展开阐述了激光诱导等离子体的形成机理和作用特性的基础上,以土壤为研究对象,研究分析LIBS系统单变量参数、土壤状态参数以及系统多变量参数对土壤的等离子体特性影响；建立了基于LIBS技术结合化学计量学方法对不同地域类型的土壤类型判别模型；对比探索了基于单波段和多变量回归的土壤中主要金属元素含量的定量分析模型和方法；探讨了基于LIBS技术结合单波段和多变量的定标方法对土壤重金属铅和镉元素含量快速检测的方法,为后期开发土壤理化信息(类型信息、元素的种类和含量信息等)检测仪器提供理论依据。具体的研究内容如下：(1)研究了LIBS系统参数与土壤状态参数对土壤等离子体特性影响规律。通过单因素分析,讨论了LIBS试验系统的主要参数(激光脉冲能量、重复频率、延时时间、采集方式以及聚焦透镜到样品的距离)以及土壤的状态参数(水分含量、颗粒大小和紧实度等)对土壤等离子体特性的影响,得到优化的试验参数：激光脉冲能量为100 mJ左右,重复频率为1 Hz,延时时间要依具体元素来定,采集方式为20次取平均,以及聚焦透镜到样品的距离为98 mm(透镜焦距为100 mm)；土壤含水率越低越好,土壤颗粒大小应小于0.15 mm(过100筛子),土壤压片的压力(即紧实度)在10 MPa较好。(2)设计了三因素(激光脉冲能量LE、延迟时间DT和聚焦透镜到样品的距离LTSD)二次回归旋转正交组合的实验,优化了LIBS系统对土壤检测的最佳实验条件。以土壤中主要元素的谱线综合信背比(signal-background-ratio, SBR) YSBR为目标函数,分析了三因素之间交互作用对YSBR的影响,结果表明：DT对YSBR的线性效果显著,而LE和LTSD对YSBR的线性效果均不显著；三者的交互影响对YSBR的交互效果都不显著；对于因素LE2、DT2和LTSD2对YSBR的曲面效应均显著。通过分析优化得到最佳的试验条件是：激光能量LE为103.09 mJ,延迟时间为2.92μs,透镜到样品的距离LTSD为97.69 mm时,得到最大综合信背比YsBR为198.602。(3)建立了基于LIBS技术结合化学计量学方法对不同类型的土壤的判别分析模型,并且验证了该模型方法的可靠性。通过对6种标准土壤样品的LIBS谱线特征进行主成分分析和元素含量对比,选取了7条特征谱线,并建立了基于7条特征谱线的偏最小二乘判别分析(Partial least squares discriminant analysis, PLS-DA)、簇类的独立软模式法(Soft independent modeling of class analogy, SIMCA)和最小二乘支持向量机(Least-squares support vector machine, LS-SVM)判别模型,判别的精度分别为98%、90%和100%,并用受试者工作特征曲线(Receiver operating characteristic curve, ROC)评价模型的性能,表明基于激光诱导击穿光谱的最小二乘支持向量机(LIBS-LS-SVM)判别模型的性能最好。针对选取的7条特征谱线,对选取另外8个不同类型的土壤样品进行分析验证,PCA得到8种土壤有明显聚类,建立的LS-SVM判别模型准确率为100%,ROC曲线也证明其预测性能的可靠性,这为研究土壤分类系统和农田土地的管理和合理利用提供理论依据；(4)应用LIBS技术结合定标曲线法以及化学计量方法,实现了对于土壤中多种元素(Al、Ca、 K、Mg、Na和Fe)同时定量检测。将LIBS数据经预处理(数据归一化,剔除异常光谱和平均处理)后,分别对比了基于谱线峰值强度、谱峰的积分信息(峰面积)和Si元素内标的定标方法。结果表明：基于峰值强度信息和谱峰的峰面积的定标曲线对多数元素都有较好的线性关系(Fe元素除外)；以Si元素内标的定标曲线的线性相关系数优于前两种定标方法；另外,利用自由定标法(Calibration free-LIBS, CF-LIBS)对土壤中主要元素Al、Ca、Si、Fe、Mg、Na和K的含量进行计算,结果有待于提高；建立了基于多变量偏最小二乘回归(partial least squares regression, PLSR)的土壤中Al、 Ca、K、Mg、Na和Fe预测模型,结果明显要优于定标曲线的分析精度,其各个的预测相关系数RP分别为：Ak,0.8455、Ca,0.9769、Fe,0.9744、K,0.8468、Mg,0.8260、 Na,0.9705,整体的预测精度要明显优于前几种定标方法,在应用LIBS技术对物质含量的定量分析中,多元的PLSR方法能够展现其较好的分析精度,也有更好的应用前景。(5)应用LIBS技术结合定标曲线法以及化学计量方法,实现了土壤中重金属铅和镉元素的快速定量检测。选取Pb Ⅰ 405.78 nm和Cd Ⅰ 361.05 nm为分析谱线,建立基于谱线峰强度,归一化后洛伦兹拟合强度以及谱峰面积与对应元素的浓度之间的关系模型。对于Pb元素,基于谱线峰强度、归一化后洛伦兹拟合强度以及谱峰面积与对应元素的浓度之间的线性关系分别为0.9839、0.9710、0.9932；而Cd元素,定标曲线法没有明显的线性关系,其分析精度有待提高;同时建立了基于PLSR方法的土壤Pb和Cd元素的定量分析模型,Pb元素的定标曲线法结果和PLSR模型的结果类似,其预测的相关系数RP为0.9485,预测均方根误差RMSEP为2.044 mg·g-1；而Cd元素的PLSR模型的结果提升较大,预测的相关系数RP为0.9949,预测均方根误差RMSEP为97.05 gg·g-1。
[Abstract]:Digital agriculture is an important way to realize the precision management and scientific planting of agriculture. It is one of the most advanced development fields of modern agriculture. It is also the key and core of agricultural efficiency, ecology, safety and sustainable development. The most fundamental and key factor in the implementation of digital precision agriculture is the quasi agricultural environment information. True perception, rapid acquisition and intelligent control. Digital and information technology can provide rapid, accurate information acquisition, scientific decision making and efficient operation control for agricultural green production and efficient management. It has become a hot spot in the field of agricultural science and technology. As one of the important natural resources for human survival, soil is an agricultural student. The research on soil information acquisition and testing is a hot spot in the field of agricultural environmental information. The analysis and study of soil types can be used to establish soil fertility and quality evaluation system for soil remediation, planning and rational use for scientific basis; the detection energy of soil element information can be found. It lays a theoretical foundation for agricultural field crop nutrition diagnosis, real-time acquisition of farmland information and scientific fertilizer management, which can effectively prevent heavy metal pollution in farmland and provide theoretical guidance for high quality and safe production of agriculture. In this study, laser induced breakdown spectroscopy (Laser-indu On the basis of the working process and principle of CED breakdown spectroscopy (LIBS) technology, the formation mechanism and function characteristics of laser induced plasma are expounded. Soil is taken as the research object, and the effects of the single variable parameters of the LIBS system, the soil state parameters and the multivariable parameters of the system on the plasma characteristics of the soil are studied and analyzed. A discriminant model of soil types based on LIBS technology combined with chemometrics method was established, and the quantitative analysis model and method for the content of major metal elements in soil based on single band and multivariable regression were compared and explored, and the soil weight based on LIBS technology combined with single band and multivariable calibration method to soil weight was discussed. The rapid detection method of metal lead and cadmium content provides a theoretical basis for the later development of soil physical and chemical information (type information, element type and content information etc.). The specific research contents are as follows: (1) the effects of LIBS system parameters and soil state parameters on soil plasma characteristics are studied. The main parameters of the LIBS test system (laser pulse energy, repetition rate, delay time, acquisition mode and the distance of focusing lens to the sample) and the influence of the state parameters of soil (moisture content, particle size and compaction) on soil plasma characteristics are discussed. The optimized experimental parameters are obtained: laser pulse energy is 100 Around mJ, the repetition rate is 1 Hz, the time delay time must depend on the specific elements, the acquisition mode is 20 times averaging, and the distance of the focusing lens to the sample is 98 mm (lens focal distance is 100 mm); the lower the soil water content the better, the soil particle size should be less than 0.15 mm (100 sieves), and the pressure of the soil press (i.e. compaction) is better in 10 MPa. (2) The experiment of three factors (laser pulse energy LE, delay time DT and the distance LTSD of focusing lens to sample) was designed for the experiment of two regression rotation orthogonal combination. The optimum experimental conditions for soil detection were optimized. The target function of the spectrum line integrated signal back ratio (signal-background-ratio, SBR) YSBR of the main elements in the soil was analyzed by three. The effect of interaction between factors on YSBR shows that DT has a significant linear effect on YSBR, while LE and LTSD have no significant linear effect on YSBR; the interaction effect of the three is not significant to the interaction effect of YSBR; for factors LE2, DT2 and LTSD2 are both obvious to the YSBR surface effect. The optimal test conditions are obtained by analysis and optimization. It is: the laser energy LE is 103.09 mJ, the delay time is 2.92 Mu s, and the distance LTSD of the sample is 97.69 mm, the maximum integrated signal back ratio YsBR is 198.602. (3), and the discriminant analysis model of different types of soil based on LIBS technology combined with chemometrics method is established, and the reliability of the model method is verified. 6 kinds of methods are verified. The LIBS line characteristics of the standard soil samples are analyzed by principal component analysis and element content comparison, and 7 characteristic spectral lines are selected, and the partial least squares discriminant analysis (Partial least squares discriminant analysis, PLS-DA) based on the spectral lines of the soil is established, and the independent soft mode method of Soft independent modeling of class (Soft independent modeling) is established. CA) and the least squares support vector machine (Least-squares support vector machine, LS-SVM) discriminant model, the accuracy of discrimination is 98%, 90% and 100% respectively, and the performance of the subjects' working characteristic curves (Receiver operating characteristic curve, ROC) is used to show the least square support vector machine based on the laser induced breakdown spectroscopy. LIBS-LS-SVM) the performance of the discriminant model is the best. According to the selected 7 characteristic spectral lines, 8 different types of soil samples are selected to be analyzed and verified by 8 different types of soil samples. 8 kinds of soil have obvious clustering, the accuracy rate of the established LS-SVM discriminant model is 100%, and the ROC curve also proves the reliability of its prediction performance. This is the study of soil classification system and agriculture. The management and rational utilization of field land provide theoretical basis; (4) the simultaneous quantitative detection of various elements (Al, Ca, K, Mg, Na and Fe) in the soil is realized by using LIBS technology and the method of calibration curve and chemical metrology. After the LIBS data is pretreated (data normalization, elimination of abnormal spectra and average processing), the data of LIBS are compared. The line peak strength, the integral information (peak area) of the peak and the calibration method of the Si element internal standard. The results show that the calibration curves of peak area based on peak intensity information and peaks have a good linear relationship with most elements (except Fe elements), and the linear correlation coefficient of the calibration curve with the internal standard of Si is superior to the first two calibration methods. In addition, the content of the main elements of soil Al, Ca, Si, Fe, Mg, Na and K in the soil are calculated by Calibration free-LIBS (CF-LIBS), and the results of the soil, which are based on the multivariable partial least squares regression (partial least), are established. Compared with the accuracy of the calibration curve, the prediction correlation coefficients RP are Ak, 0.8455, Ca, 0.9769, Fe, 0.9744, K, 0.8468, Mg, 0.8260, Na, 0.9705, and the overall prediction accuracy is obviously superior to the previous methods. In the quantitative analysis of the material content in the application of LIBS technology, the multivariate PLSR method can show its better analysis. Precision and better application prospect. (5) the rapid quantitative detection of heavy metal lead and cadmium in soil was realized with LIBS technique combined with calibration curve method and chemical metrology method. Pb I 405.78 nm and Cd I 361.05 nm were selected as the analytical spectral lines, and the intensity of spectral line peak, normalized intensity and peak surface after normalization were established. For the Pb element, based on the spectral peak intensity, the linear relationship between the Lorenz fitting strength and the peak area of the spectral peak and the concentration of the corresponding elements is 0.9839,0.9710,0.9932, while the Cd element, the calibration curve method has no obvious linear relationship, and the accuracy of the analysis needs to be improved. At the same time, the quantitative analysis model of soil Pb and Cd elements based on PLSR method is established. The calibration curve method of Pb element is similar to the result of PLSR model. The correlation coefficient of the prediction is 0.9485 and the mean square root error RMSEP is 2.044 mg. G-1, while the PLSR model of the Cd element is higher and the predicted correlation coefficient RP is 0.9949. The mean square root error of RMSEP is 97.05 GG. G-1.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S151.9;TN249
，

本文编号：2006654