基于BP-ANN和PLS的近红外光谱无损检测李果实品质的研究

可溶性固形物（SSC）和可滴定总酸（TA）含量是影响李果实品质的重要指标，经典的破坏性检测方法不适用于果实按品质分级，近红外光谱(NIRS)检测方法具有速度快、操作简便、可无损检测果实品质。为实现NIRS无损快速检测安哥诺李果实可溶性固形物和可滴定总酸含量，利用NIRS采集李果实的漫反射光谱，同时采用糖度计测定安哥诺李果实的SSC，采用滴定法测定了李果实TA含量，使用杠杆值和F概率值剔除异常样品，采用软件优化结合人工筛选光谱波段，使用了消除常数偏移量、减去一条直线、矢量归一化（SNV）、最大-最小归一化、多元散射校正（MSC）、一阶和二阶导数结合平滑处理、一阶导数结合减去一条直线和平滑处理、以及一阶导数结合SNV或MSC校正等光谱预处理方法，分别采用偏最小二乘法（PLS）和主成分分析结合反向传播人工神经网络（BP-ANN）建立李果实SSC、TA的定量分析模型。结果表明，李果实SSC和TA的最佳PLS建模效果波段范围分别为4 000～8 852和4 605～6 523 cm-1。SSC的PLS模型的最佳光谱预处理方法为MSC校正，最佳模型校正相关系数（R_c）为0.914 4，预测相关系数（R_p）为0.878 5，校正均方根误差（RMSEC）为0.91，预测均方根误差（RMSEP）为1.00。经一阶微分结合SNV和9点平滑的方法预处理后，TA的PLS模型效果最佳，R_c，R_p，RMSEC，RMSEP分别为0.860 3，0.819 6，0.80和0.86。提取了李果实SSC和TA光谱数据的主成分，并基于前10个主成分得分建立了李果实SSC和TA最佳BP-ANN定量分析模型，其R_c，R_p，RMSEC和RMSEP分别为0.976 7，0.889 7，0.75和0.99；TA的BP-ANN模型的相应参数值依次为0.974 3，0.897 7，0.62和0.83，与采用PLS算法建立的定量模型相比较，BP-ANN模型具有较高的R_c，R_p和较低的RMSEC，RMSEP，因此BP-ANN模型对SSC和TA指标的定量分析结果更佳。

Quantitative Analysis of Soluble Solids and Titratable Acidity Content in Angeleno Plum by Near-Infrared Spectroscopy With BP-ANN and PLS

Soluble solid content (SSC) and titratable acidity (TA) are important indexes affecting the fruit quality and the fruit quality grading. Classical destructive detection methods are not suitable for fruit classification by quality. NIRS detection method is fast, easy to operate and can detect fruit quality without damage. In order to achieve non-destructive and rapid determination of SSC and TA in Angeleno plum fruits by near-infrared spectroscopy (NIR), diffuse reflectance spectra of plum fruits were collected by NIR spectrometer, SSC was measured by saccharometer, and TA content was determined by titration. Using leverage and F probability value to eliminate abnormal samples and software optimization combined with a manual screening of spectral bands, eliminating constant offset, subtracting a straight line, standard normal variate (SNV), max-minimum normalization, and multiplicative scatter correction （MSC), first and the second derivative combined smoothing, the first derivative combined minus a straight line and smoothing, and the first derivative combined with SNV or MSC correction. Partial least squares (PLS) and back propagation artificial neural network (BP-ANN) were used to establish the quantitative models of SSC and TA of plum fruit. Results indicated that the best Band ranges of plum fruit SSC and TA are 4 000～8 852 and 4 605～6 523 cm-1 respectively. The best spectral preprocessing method of the PLS model of SSC was MSC correction. The best model correction correlation coefficient (R_c) was 0.914 4, the prediction correlation coefficient (R_p) was 0.878 5, the correction root means square error (RMSEC) was 0.91, and the prediction root means square error (RMSEP) was 1.00. After the first order differential combined with SNV and 9-point smoothing, the PLS model of TA was the best, and the R_c, R_p, RMSEC and RMSEP were 0.860 3, 0.819 6, 0.80 and 0.86. The principal components of SSC and TA spectral data of plum fruits were extracted, and the optimal BP-ANN quantitative analysis model of SSC and TA were established based on the first 10 principal component scores. The SSC BP-ANN model’s R_c, R_p, RMSEC and RMSEP were 0.976 7, 0.889 7, 0.75 and 0.99. The corresponding parameter values of the BP-ANN model of TA were 0.974 3, 0.897 7, 0.62 and 0.83, respectively. Compared with the quantitative model established by the PLS algorithm, the BP-ANN model has higher R_c and R_p and lower RMSEC and RMSEP than that of the PLS algorithm, so the quantitative analysis results from the BP-ANN model were better than that of the PLS algorithm for SSC and TA indicators.