摘要：
为了更加全面的建立中国土壤类型系统，了解中国土壤地域差异，从而提高土地资源的利用率，以及根据土壤类型指导农业科学生产。该研究利用激光诱导击穿光谱（laser-induced breakdown spectroscopy，LIBS）技术结合化学计量学方法对土壤类型进行判别分析研究。从6种标准土壤样品出发，分析所采集6种土壤的LIBS光谱谱线特征，结合其主要成分物质（SiO2，Al2O3，Fe2O3，FeO，MgO，CaO，Na2O，K2O）的含量，针对每种主要物质选取了Si I 390.55 nm、Al I 394.40 nm、Fe I 422.74 nm、Mg I 518.36 nm、Na I 588.96 nm、Ca II 393.37 nm、K I 766.49 nm为特征分析谱线。结合所选的7条特征谱线下的300个标准土壤样品的光谱（200个为训练集，100个为预测集），对训练集光谱进行主成分分析（principal component analysis，PCA），6种土壤有明显的聚类。然后根据训练集光谱值和预先赋予土壤类型的虚拟等级值分别建立最小二乘判别分析（partial least squares discriminant analysis，PLS-DA）和最小二乘支持向量机（least-squares support vector machine，LS-SVM）判别模型，分析预测结果二者总的判别准确率分别为98%和100%。用受试者工作特征曲线（receiver operating characteristic curve，ROC）评价这2个模型的性能，结果表明LS-SVM判别模型性能优于PLS-DA模型。基于以上结果，选取不同于标准土壤的另7种不同类型土壤进行试验验证所选特征谱线和判别模型，建立7种不同类型土壤的LS-SVM预测模型，其预测准确率达100%，ROC曲线对其评价的性能很好。研究证明，LIBS技术结合化学计量学方法能够实现对土壤类型的判别分析，这为土壤质量的正确评价，土壤的整治、规划和合理利用提供理论基础。%Laser-induced breakdown spectroscopy (LIBS), as a kind of atomic emission spectroscopy (AES), has been considered to be a future “Superstar” in the field of chemical analysis and green analytical techniques due to its unique features, like little or no sample preparation, stand-off or remote analysis, fast and multi-element analysis, wide application in various aspects. To establish the soil type system in China and more comprehensively understand the type of elements in the soil, soil types were studied to improve the utilization of land resources and offer a theoretical guide for agricultural scientific production. This research focused on investigating the soil types using LIBS coupled with chemometrics methods. A laboratorial LIBS device working in air was employed to obtain the 300 (every 50 LIBS spectra acquired from one type of soil) LIBS spectra of 6 soil samples. Based on the contents of main materials (SiO2, Al2O3, Fe2O3, FeO, MgO, CaO, Na2O, K2O) of 6 kinds of standard soil samples, their corresponding LIBS curve characteristics were analyzed. Then 7 characteristic spectral lines at SiI390.55 nm, AlI394.40 nm, FeI422.74 nm, MgI518.36 nm, NaI588.96 nm, Ca II 393.36 nm and KI766.49 nm (I represented atomic spectral line, II meant ionic spectral lines) were identified. Based on 300 spectra at 7 characteristic spectral lines from 6 standard reference soils, of which 200 were in training set and 100 in prediction set divided by sample set partitioning based on joint x-y distances (SPXY) method, the principal component analysis (PCA) was carried out on the training set and an obvious cluster was observed from the score plot of the first 2 principal components (PCs). Meanwhile, partial least squares discriminant analysis (PLS-DA) and least-squares support vector machine (LS-SVM) models were introduced to establish the discriminant models and the correct rates of discrimination were 98% and 100%, respectively. Then, the performances of PLS-DA and LS-SVM models were evaluated using receiver operating characteristic (ROC) curve. The results demonstrated that the LS-SVM discriminant model with the parameter area of 1 was superior to the model of PLS-DA with the area of 0.99569, which illustrated that the LS-SVM discriminant model was robust. Based on this, 7 types of soils from different places were used to conduct the same experiments to acquire 385 (every 55 LIBS spectra acquired from one type of soil) and then to verify the selected seven characteristic spectral lines and discriminant model. PCA on the training set of 255 LIBS spectra from 7 types of soil samples also displayed apparent cluster. Then, the LS-SVM model based on the training set from 7 types of soils was built to predict the prediction set of 130 LIBS spectra and the prediction accuracy was 100%. The performance of the model was also evaluated using ROC and it exhibited an excellent result. The research reveals that LIBS technology coupled with chemometrics methods can achieve the discriminant analysis of different types of soils, which provides a theoretical guidance for soil quality assessment, management, planning and reasonable use.
