光谱学与光谱分析, 2019, 39 (2): 584, 网络出版: 2019-03-06  

共线双脉冲LIBS结合变量筛选定量检测腐霉利含量

Double Pulse LIBS Combined with Variable Screening to Detect Procymidone Content
甘兰萍 1,2,*孙通 1,2刘津 1,2刘木华 1,2
作者单位
1 江西农业大学工学院, 江西省高校生物光电技术及应用重点实验室
2 江西省果蔬 采后处理关键技术及质量安全协同创新中心, 江西 南昌 330045
摘要
腐霉利(Procymidone) 作为一种新型的农产品杀菌剂, 具有防止农产品受病虫害的作用, 但其在施药过程中容易使用不当危害环境和人的健康。 为加强对腐霉利农药的检测, 本研究应用共轴双脉冲激光诱导击穿光谱技术(LIBS) 对溶液中的腐霉利含量进行定量检测研究。 为配置不同浓度的腐霉利样品, 将有效成分含量为98%腐霉利粉末与二甲苯按照不同比例混合并完全溶解。 由于液体样品在激光击打的过程中容易将液体溅出, 具有一定的危险性。 因此, 实验将液体样品转化为固体样品, 利用石墨吸附腐霉利溶液, 然后采用八通道高精度光谱仪采集样品的LIBS光谱, 并利用不同预处理方法对光谱数据进行预处理。 为提高腐霉利的检测精度, 选择氯元素信号最强的两通道(744.555~935.843, 893.107~1 057.058 nm) 光谱数据, 分别采用归一化函数(normalization) 、 基线校正(baseline correction) 、 标准正态变量变换(SNV) 、 多元散射校正(MSC) 方法进行光谱预处理, 并应用PLS方法建模。 通过比较各预处理方法数据后, 综合考虑, 选择Baseline方法为最佳预处理方法。 在baseline预处理方法的基础上使用无信息变量消除算法(UVE) 联合竞争性自适应重加权采样(CARS)算法剔除无信息的波长变量, 筛选与腐霉利相关的重要波长变量, 最后应用偏最小二乘回归建立溶液中腐霉利含量的定量预测模型。 建模结果表明: 经光谱预处理和UVE-CARS方法优选后, 可将原4096个波长变量个数减少至13个, 变量压缩率为99.68%; 经UVE-CARS变量优选后建立的PLS模型的校正集的决定系数和均方根误差分别为0.990 5和0.66, 预测集的决定系数和均方根误差分别为0.990 3和0.67, 其模型性能优于原始光谱建立的PLS模型。 结果表明, 利用共轴双脉冲LIBS技术定量检测溶液中的腐霉利含量具有一定的可行性, 经UVE和CARS方法筛选后可以有效提取腐霉利的特征变量及相关影响变量, 剔除冗余及噪声影响变量, 简化定量分析模型且提高了定量分析模型的稳定性。
Abstract
Procymidone, as a new type of agricultural fungicide, has the effect of preventing agricultural products from being affected by pests and diseases, but it is easy to be used improperly to harm the environment and human health during the application process. In order to strengthen the detection of procymidone pesticides, this study uses laser induced breakdown spectroscopy (LIBS) to quantitatively detect the content of procymidone in solution. In order to prepare different density of procymidone samples, this study mixed the ingredient content of 98% procymidone powder with xylene in different proportions and completely dissolved. Since liquid samples are easy to spill and cause dangers during laser striking, so this experiment converted the liquid samples into solid samples, used the graphite to adsorb the procymidone, and then used the eight-channel high-precision spectrometer to collect the LIBS spectrum of the sample, and applied different pretreatment methods to pretreat the spectral data. So as to improve the detection accuracy of procymidone, this research chose the strongest chlorine signalthe in two channels (744.555~935.843, 893.107~1 057.058 nm) and spectral data were preprocessed with normalization, baseline correction, standard normal variable transformation and multiplicative scatter correction methods respectively, and PLS method was used to model. After comparing the data of each pretreatment method, considering the comprehensive consideration, the Baseline method was selected as the optimal pretreatment method. Based on the baseline preprocessing method, uninformed variable elimination (UVE) combined with competitive adaptive reweighted sampling (CARS) algorithm was used to eliminate the wavelength variable without information, and screen out the important wavelength variables related to procymidone, and finally the partial least squares regression was used to establish a quantitative prediction model of procymidone content in solution. The modeling results showed that after the spectral preprocessing and optimized by VUE-CARS method, the number of original 4 096 wavelength variables reduced to 13, and the variable compression rate was 99.68%. The PLS model was established after the UVE-CARS variable was optimized. The correction set and prediction set determination coefficient and root mean square error were 0.990 5, 0.66, and 0.990 3, 0.67, respectively. The model performance was better than the PLS model established by the original spectrum. The results showed that it is feasible to detect the procymidone content quantitatively in the solution by using the coaxial double pulse LIBS technique. After screened by UVE and CARS methods, the characteristic variables and related influence variables of procymidone can be effectively extracted. The redundancy and noise influences variables can be eliminated effectively. The quantitative analysis model can be simplified and the stability of the quantitative analysis model can be improved.

甘兰萍, 孙通, 刘津, 刘木华. 共线双脉冲LIBS结合变量筛选定量检测腐霉利含量[J]. 光谱学与光谱分析, 2019, 39(2): 584. GAN Lan-ping, SUN Tong, LIU Jin, LIU Mu-hua. Double Pulse LIBS Combined with Variable Screening to Detect Procymidone Content[J]. Spectroscopy and Spectral Analysis, 2019, 39(2): 584.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!