腔体约束LIBS结合机器学习对土壤重金属元素的定量分析

土壤重金属元素含量检测及防治，对我国农业、生态环境修复具有重大意义。利用外加腔体约束结合激光诱导击穿光谱技术（LIBS）获得土壤光谱数据，采用机器学习对土壤中重金属元素Ni和Ba含量进行分析。实验设置延迟时间为0.5～5 μs，选择Ni Ⅱ 221.648 nm和Ba Ⅱ 495.709 nm作为目标研究特征谱线，计算两种LIBS条件下延迟时间对信噪比、光谱强度及增强因子的影响。结果表明，腔体约束LIBS（CC-LIBS）可以增大光谱强度及目标元素信噪比，同时随着采集延迟时间增长，等离子体数目变少，光谱强度及信噪比逐渐减小并趋于稳定；当延迟时间设置为1 μs时，CC-LIBS条件下Ni和Ba元素特征谱线信噪比达到最优，确定此时为LIBS最优实验条件。通过最优条件获取9种含Ni和Ba元素土壤样品的光谱数据，由于采集到的每组光谱信息有12 248个数据点，利用主成分分析（PCA）对CC-LIBS条件下的光谱数据降维，在保留95%以上的土壤原始信息后，选择9个主成分作为定量分析模型的输入变量，以提高模型的运算速度。采用机器学习中的Lasso，AdaBoost和Random Forest模型，对PCA降维后的光谱数据进行建模及预测，实现土壤重金属元素Ni和Ba的定量分析。结果表明，与Lasso和AdaBoost模型相比，Random Forest模型在训练集和测试集中表现出的预测性能最优。Random Forest模型下Ni元素在测试集中的R2为0.937，RMSEP为3.037；Ba元素在测试集中的相关系数R2为0.886，均方根误差RMSEP为90.515。基于腔体约束LIBS技术结合机器学习，为土壤重金属元素的高精度检测提供了技术指导。

Quantitative Analysis of Soil Heavy Metal Elements Based on Cavity Confinement LIBS Combined With Machine Learning

The detection and control of the content of heavy metal elements in the soil are of great significance to the restoration of agriculture and the ecological environment. This study used external cavity confinement combined with traditional laser-induced breakdown spectroscopy (LIBS) to obtain soil spectral data. Then machine learning was used to analyze the content of heavy metal elements Ni and Ba in the soil. During the experiment, the delay time was set to 0.5～5 μs, Ni Ⅱ 221.648 nm and Ba Ⅱ 495.709 nm were selected as the target characteristic spectrum to study, and calculated the influence of delay time on the signal-to-noise ratio (SNR), spectral intensity and enhancement factor under two LIBS conditions. Experimental results show that cavity confinement LIBS (CC-LIBS) can increase the target element’s spectral intensity and SNR. As the acquisition delay time increases, the number of plasmas decreases, and the spectral intensity and SNR gradually decrease, then become stable; when the delay time is set to 1 μs, the SNR of the characteristic spectrum of Ni and Ba elements reaches the best under CC-LIBS conditions, which is determined to be the optimal experimental condition for LIBS at this time. Obtain the spectral data of 9 soil samples containing Ni and Ba through optimal conditions. Since there were 12 248 data points for each set of collected spectral information, the principal component analysis algorithm (PCA) was used to reduce the dimensionality of the spectral data under CC-LIBS conditions. After retaining more than 95% of the original soil information, 9 principal components were selected as the quantitative analysis model’s input variables to improve the model’s calculation speed. The Lasso, AdaBoost and Random Forest models in machine learning were used to model and predict the spectral data after PCA dimensionality reduction to realize the quantitative analysis of soil heavy metal elements Ni and Ba. The experimental results show that the Random Forest model has the best prediction performance in the training and test sets compared with Lasso and AdaBoost models. Under the Random Forest model, the correlation coefficient R2 of the Ni element in the test set is 0.937, and the root mean square error (RMSEP) is 3.037; the R2 of the Ba element in the test set is 0.886, the RMSEP is 90.515. This paper is based on the research of cavity-confinement LIBS technology combined with machine learning to provide theoretical support and technical guidance for the high-precision detection of heavy metal elements.