摘要
为了提高农产品中重金属的光学快速检测的精度,运用偏最小二乘(PLS)法结合激光诱导击穿光谱(LIBS)对马铃薯中的Pb含量进行了定量分析,探讨了数据预处理方法对模型精度的影响。针对96个污染马铃薯样品的LIBS数据,分别进行3点到17点平滑处理,然后将平滑后的数据分别进行标准正态变量变换(SNV)、多元散射校正(MSC)、均值中心化(MC)、一阶导数(FD)和二阶导数(SD)求导去噪预处理。采用湿法消解结合原子吸收分光光度计(AAS)获取样品中Pb元素的真实浓度,选择包含Pb特征谱线的401-417nm波段进行PLS建模,对比分析模型的相关系数r、交互验证均方根误差(RMSECV)和预测均方根误差(RMSEP)。结果表明,采用13点平滑、均值中心化预处理的PLS模型的校准质量和预测效果最好,模型的r、RMSECV和RMSEP分别达到了0.9963、16.4和11.5,说明选择合适的数据预处理方法能有效提高LIBS检测果蔬产品定量模型的质量。
Lower detection accuracy restricts the development of laser induced breakdown spectroscopy (LIBS). A partial least square (PLS) method is proposed to analyze the Pb content in potatoes by LIBS. The PLS model is composed of two steps which include different smoothing points and standard normal variate transformation (or multiple scatter correction,or the mean centralized, or the first derivative of denoising,or second derivative of denoising) ,to pretreat the LIBS spectrum. Spectral data of the 96 potatoes samples are pretreated by different data smoothing from 3 points to 17 points. The samples of potatoes are pretreated by acid wet digestion, and the real content of Pb is obtained by atomic absorption spectrophotometer (AAS). The LIBS spectrum chooses the 401--417 nm wavelength range which con- tains characteristic spectral lines of Pb. The main evaluation indexes of PLS models, such as correlation coefficient, root mean squared error of cross-validation and root mean square error of prediction, are compared and analyzed. The obtained three indexes of PLS model after 13 points smoothing and processing of the mean center are reach 0. 996 3,16.4 and 11.5, respectively. In one word, the calibration and predic- tion of this quantity model are the best results. Selecting the appropriate data for preprocessing can effectively improve the quality of agricultural products detected by LIBS.
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2015年第1期141-148,共8页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(31460419
31271612)
江西省重大科技项目(20143ACB21013)
江西省教育厅科技计划(GJJ10681)
江西省学术带头人计划(09004004)
赣鄱"555"英才和大学生创新计划(201410410010)资助项目