光谱预处理对LIBS检测脐橙中Cu的偏最小二乘定量模型影响

应用激光诱导击穿光谱(LIBS)对脐橙中Cu元素进行快速检测，并结合偏最小二乘法(PLS)进行定量分析，探索光谱数据预处理方法对模型检测精度的影响。针对实验室污染处理后的52个赣南脐橙样品的光谱数据，进行不同数据平滑、均值中心化和标准正态变量变换三种预处理方法。然后选择包含Cu特征谱线的319~338 nm波段进行PLS建模，对比分析了模型的主要评价指标回归系数(r)、交互验证均方根误差(RMSECV)和预测均方根误差(RMSEP)。采用13点平滑、均值中心化的PLS模型3个指标分别达到了0.992 8，3.43和3.4，模型的平均预测相对误差仅为5.55%，即采用该前处理方法模型的校准质量和预测效果都最好。选择合适的数据前处理方法能有效提高LIBS检测果蔬产品PLS定量模型的预测精度，为果蔬产品LIBS快速精准检测提供了新方法。

Influence of Spectral Pre-Processing on PLS Quantitative Model of Detecting Cu in Navel Orange by LIBS

Cu in navel orange was detected rapidly by laser-induced breakdown spectroscopy (LIBS) combined with partial least squares (PLS) for quantitative analysis, then the effect on the detection accuracy of the model with different spectral data pretreatment methods was explored. Spectral data for the 52 Gannan navel orange samples were pretreated by different data smoothing, mean centralized and standard normal variable transform. Then 319~338 nm wavelength section containing characteristic spectral lines of Cu was selected to build PLS models, the main evaluation indexes of models such as regression coefficient (r), root mean square error of cross validation (RMSECV) and the root mean square error of prediction (RMSEP) were compared and analyzed. Three indicators of PLS model after 13 points smoothing and processing of the mean center were found reaching 0.992 8, 3.43 and 3.4 respectively, the average relative error of prediction model is only 5.55%, and in one word, the quality of calibration and prediction of this model are the best results. The results show that selecting the appropriate data pre-processing method, the prediction accuracy of PLS quantitative model of fruits and vegetables detected by LIBS can be improved effectively, providing a new method for fast and accurate detection of fruits and vegetables by LIBS.