Luo Xia, Hong Tian-sheng, Luo Kuo, Dai Fen, Wu Wei-bin, Mei Hui-lan, Lin Lin
Guang Pu Xue Yu Guang Pu Fen Xi. 2016 May;36(5):1345-51.
The objective of present study was to find out an accurate, rapid and nondestructive method to detect total acid content (TA) of pitaya with visible/near-infrared spectrometry, wavelet transform (WT) and successive projections algorithm (SPA), which will provide scientific basis for non- destructive measurement of pitaya. Maya2000 fiber-optic spectrumeter was used to collect spectral data of pitaya on the wavelength in the range of 380~1 099 nm; and then with the methods of WT denosing pretreatment, SPA and partial least squares regression (PLSR) quantitative forecasting model of TA of pitaya was established. The result showed that the precision of WT-SPA-PLSR model, which combine the WT with SPA, was better than that of PLSR model based on the whole wave variables. The relation coefficient of the PLSR model (Rp) that predicted TA based on the original spectrum of all samples as the input variables was 0.851 394 and RMSEP was 0.086 848. The original spectrum variable of the all samples were processed by using wavelet function dbN(N=2, 3, …, 10) for wavelet decomposition and de-noising. The optimal results of noise reduction were decomposed in level 2 using wavelet function db4 (db4-2). The Rp of WT-PLSR model was 0.915 635 and RMSEP was 0.066 752. The prediction of model using wavelet transform de-noising was improved significantly. After the original spectrum processed by db10-3 and SPA, 12 preferred variables were selected from 570 spectrum variables, such as 530, 545, 604, 626, 648, 676, 685, 695, 730, 897, 972, 1 016 nm spectrum variables. The WT-SPA-PLSR model based on these 12 variables as input variables was established. Rp of the WT-SPA-PLSR prediction model was 0.882 83 and RMSEP was 0.077 39. SPA algorithm was suitable for the selection of spectrum variables which could effectively obtain the spectrum variables which were strong correlation with TA and increase the accuracy and stability of the prediction model. The results indicated that the nondestructive detection for TA of pitaya based on the diffuse reflectance visible/near-infrared spectrometry, WT and SPA was feasible.
本研究的目的是利用可见/近红外光谱法、小波变换（WT）和连续投影算法（SPA）找出一种准确、快速且无损的方法来检测火龙果的总酸含量（TA），这将为火龙果的无损检测提供科学依据。使用Maya2000光纤光谱仪在380~1 099 nm波长范围内采集火龙果的光谱数据；然后采用WT去噪预处理方法，建立了火龙果TA的SPA和偏最小二乘回归（PLSR）定量预测模型。结果表明，将WT与SPA相结合的WT-SPA-PLSR模型的精度优于基于全波变量的PLSR模型。以所有样品的原始光谱为输入变量预测TA的PLSR模型的相关系数（Rp）为0.851 394，预测均方根误差（RMSEP）为0.086 848。对所有样品的原始光谱变量使用小波函数dbN（N=2, 3, …, 10）进行小波分解和去噪处理。使用小波函数db4（db4-2）在第2层分解得到的降噪效果最佳。WT-PLSR模型的Rp为0.915 635，RMSEP为0.066 752。使用小波变换去噪后的模型预测效果有显著提高。对原始光谱经db10-3和SPA处理后，从570个光谱变量中筛选出12个优选变量，如530、545、604、626、648、676、685、695、730、897、972、1 016 nm光谱变量。以这12个变量为输入变量建立了WT-SPA-PLSR模型。WT-SPA-PLSR预测模型的Rp为0.882 83，RMSEP为0.077 39。SPA算法适用于光谱变量的选择，能有效获取与TA相关性较强的光谱变量，提高预测模型的准确性和稳定性。结果表明，基于漫反射可见/近红外光谱法、WT和SPA对火龙果TA进行无损检测是可行的。
