基于嗅觉和光谱技术融合的面粉脂肪酸值定量检测

提出了一种基于比色传感器数据和近红外光谱特征融合的储藏期面粉脂肪酸值的定量检测方法。开发比色传感器阵列、搭建便携式近红外光谱测量系统,分别采集不同储藏期面粉样本的比色传感器数据和近红外光谱。利用主成分分析分别对预处理后的比色传感器数据和近红外光谱数据进行特征降维,采用五折交互验证法在反向传播神经网络(BPNN)模型校正过程中进行优化,确定基于单技术分析模型的最佳主成分(PCs)个数。将优化后的基于单技术模型的最佳PCs在特征层进行融合,建立基于融合特征的BPNN分析模型,以实现对面粉储藏过程中脂肪酸值的快速检测。实验结果显示,基于比色传感器特征和基于近红外光谱特征建立的最佳BPNN模型的最佳PCs数量分别为3和4,基于融合特征建立的BPNN模型在预测集中的相关系数和预测均方根误差的均值分别为0.9276和1.9345 mg/(100 g)。研究表明,与单技术数据分析模型相比,基于比色传感器数据和近红外光谱特征融合模型的检测精度和泛化性能都有所提高。本研究可为粮食储藏品质的高精度原位监测提供一种技术方法。

Quantitative Determination of Fatty Acid Value of Flour Based on Fusion of Olfactory and Spectral Technique

A novel quantitative detection method for the fatty acid value of flour during storage was innovatively proposed based on the feature fusion of colorimetric sensor data and near-infrared spectra.A colorimetric sensor array was developed to collect colorimetric sensor data of flour samples in different storage periods.A portable near-infrared spectroscopy system was built to collect near-infrared spectra of flour samples in different storage periods.Principal component analysis was used to perform feature reduction on the preprocessed colorimetric sensor data and near-infrared spectra.In the back propagation neural network(BPNN)model calibration,the five-fold cross-validation method was used to optimize and determine the best principal components(PCs)of the single technique analysis model.The optimized PCs based on the single technique model were fused in the feature layer,and a BPNN analysis model based on the fusion feature was established to realize the rapid detection of the fatty acid value during the flour storage.Experimental results showed that the number of PCs in the best BPNN model based on the characteristics of the colorimetric sensor and the near-infrared spectra were three and four,respectively.The average values of correlation coefficient and root mean square error of prediction of the BPNN model based on the fusion features in the prediction set were 0.9276 and 1.9345 mg/(100 g),respectively.The overall results showed that compared with the single technique data analysis model,the detection accuracy and generalization performance of the colorimetric sensor data and the near-infrared spectral feature fusion model were improved.The results can provide a new technical method for high-precision in-situ monitoring of grain storage quality.