基于可见-近红外光谱法无损检测赣南脐橙中总酸含量

Non-destructive detection of total acid content in Gannan navel orange based on visible-near infrared spectroscopy

中文关键词:  可见-近红外光谱法  无损检测  总酸含量  赣南脐橙

英文关键词:visible-near infrared spectroscopy  non-destructive examination  total acid content  Gannan navel orange

中文摘要:

      目的 基于可见-近红外光谱法建立一种无损测定赣南脐橙总酸含量的技术。方法 利用设计的可见-近红外光谱检测系统检测168个赣南脐橙总酸含量。以给定赣南脐橙的126个样品作为校正集, 42个未知样品作为预测集。本研究以去除首尾处噪声后的400~880 nm范围的光谱波段, 共481个波长点进行研究分析。结合SG (Savitzky-Golay)平滑法、多元散射校正法、变量标准化法、基线校正法4种预处理方法处理原始光谱数据, 通过PLSR数学模型确定最佳预处理模型; 再利用竞争性自适应重加权算法(competitive adaptive reweighted sampling, CARS)、随机蛙跳算法(random frog, RF)、遗传算法(genetic algorithm, GA)、连续投影算法(successive projections algorithm, SPA)和主成分分析法(principal component analysis, PCA) 5种算法对预处理后的数据提取特征变量, 降低维度, 随后分别建立基于特征变量的总酸偏最小二乘回归(partial least squares regression, PLSR)、主成分回归(principal component regression, PCR)、最小二乘支持向量机(least square support vector machine, LS-SVM)及多元线性回归(multiple linear regression, MLR)预测模型。结果 通过PLSR数学模型确定SG平滑预处理模型效果为最佳, 基于SG+GA+LS-SVM模型对总酸含量预测效果最佳, 预测集均方根误差(root mean square error of prediction, RMSEP)值为0.016, 预测集决定系数(prediction set coefficient of determination, )值为0.9834, 相对分析误差(residual predictive deviation, RPD)值为7.76。结论 基于可见-近红外光谱法实现赣南脐橙中总酸含量的无损检测是可行的, 结合SG+GA+LS-SVM预测模型可以实现赣南脐橙总酸含量的定量检测, 可用于评价赣南脐橙总酸含量。

英文摘要:

      Objective To establish a non-destructive method for the determination of total acid content in Gannan navel orange based on visible-near infrared spectroscopy. Methods The designed visible-near infrared spectroscopy detection system was used to detect the total acid content in 168 Gannan navel oranges. The 126 samples in the given Gannan navel orange were used as the modeling set, and the 42 unknown samples served as prediction sets. In this experiment, a total of 481 wavelength points were studied and analyzed in the 400-880 nm range of the spectrum band after removing the noise at the head and tail. The original spectral data were processed by 4 kinds of preprocessing methods including SG (Savitzky-Golay) smoothing method, multivariate scattering correction method, standard normal variable method, and baseline offset correction method, and the optimal pretreatment model was determined by PLSR mathematical model; then 5 kinds of algorithms including competitive adaptive reweighted sampling (CARS), random frog (RF), genetic algorithm (GA), successive projections algorithm (SPA) and principal component analysis (PCA) were used to extract characteristic variables and reduce the dimensions of the data information, the partial least squares regression (PLSR)、principal component regression (PCR)、least square support vector machine (LS-SVM) and multiple linear regression (MLR) prediction models of total acid content in Gannan navel orange based on the characteristic variables were established. Result The prediction effect of total acid content based on SG+GA+LS-SVM model was the best. The root mean square error of prediction (RMSEP) value was 0.016, the prediction set coefficient of determination ( value was 0.9834, and the residual predictive deviation (RPD) value was 7.76. Conclusion It is feasible to realize the non-destructive detection of the total acid content in the Gannan navel orange based on visible-near infrared spectroscopy, which can realize the quantitative detection of the total acid content of Gannan navel orange combining with the SG+GA+LS-SVM prediction model and can be used to evaluate the total acid content of Gannan navel orange.