Loading...

Development and Application of Miniaturized Nearinfrared Spectroscopy Coupled with Multivariate Classification Techniques for Food Authenticity

Ehsani, Samaneh | 2023

88 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 56653 (03)
  4. University: Sharif University of Technology
  5. Department: Chemistry
  6. Advisor(s): Parastar Shahri, Hadi; Yazdanpanah, Hassan
  7. Abstract:
  8. The need to measure the authenticity of food has always been an important concern of many countries. Since food fraud can cause serious problems, every year, much research is conducted in the field of food authenticity all over the world. Milk among dairy products and orange juice among fruit juices are two categories of widely consumed beverages all over the world, therefore they are exposed to various frauds from the manufacturers. Frauds such as adding water to milk and preservatives such as hydrogen peroxide, formaldehyde, and sodium hypochlorite are common frauds that are added to milk to increase its volume and storage time. Among the common cases of adulteration in orange juice, we can mention the addition of water, sucrose, citric acid, artificial flavors, and colors. Since the identification of these adulterants with devices such as chromatography is time-consuming and expensive, nowadays many researchers are interested in using spectroscopic systems to quickly identify food fraud. In the meantime, the use of portable spectrometers is suitable due to their speed and on-site analysis. In the first study, an effective method for rapid identification of water adulteration in milk samples using a portable near-infrared spectrometer and ensemble techniques is presented. The results of the classification method based on the random subspace ensemble (RSDE) algorithm was satisfied compared to two common methods in chemometrics, the partial least squares discriminant analysis (PLS-DA) , and the radial basis function-support vector machine (RBF- SVM). Moreover, the boosting regression tree method (BRT) was used to quantitatively measure the water adulteration level in milk and its performance was compared with the usual partial least squares regression (PLSR) method. The BRT model performed better in the form of calibration, validation, and prediction. In the second research, three miniaturized near-infrared spectrometers such as Linksquare (400-1000 nm), Tellspec in the spectral range of (900-1700 nm), and Neospectra (1350-2500 nm) along with ensemble learning techniques in machine learning were used to predict adulteration related to preservatives in cow's milk. Considering the low signal-to-noise of portable spectrometers, the use of ensemble classifiers in combination with Linksquare and Tellspec spectrometers is a suitable approach for classifying the original and the adulterated milk with sodium hypochlorite and hydrogen peroxide. The results of ensemble learning methods used in this research revealed that the use of these methods can provide better performance even compared to basic models such as soft independent modeling of class analogy (SIMCA) and partial least squares discriminant analysis (PLS-DA). In addition, in the case of the mixtures of two types of frauds, the ensemble approaches such as ensemble bagging tree (EBT), RSDE, and random subspace ensemble -k-nearest neighbor (RSE-kNN) method in combination with Tellspec spectrometer had the highest performance in terms of classification parameters.The results of these two studies can be used as a reference for other studies in the field of milk authenticity.In the third study, the measurement of the brix to citric acid ratio in pulp-wash as adulteration in fruit juice was done according to the measurement of brix (sucrose content) and citric acid content in three different samples of orange juice (North, South and blood oranges). On this matter, two hand-held spectrometers were used in combination with chemometrics tools as quick tools for screening. DD-SIMCA and soft-PLS-DA methods were used as modeling tools. In this regard, the DD-SIMCA method showed a good performance in the classification of genuine and adulterated samples. Furthermore, the gradient boosting tree (GBT) method, as one of the ensemble learning strategies, can detect adulteration in orange samples with LOD of 2% for brix to citric acid ratio and 5% of pulp-wash with high sensitivity, specificity, and accuracy. The results of this research can be used in the fruit juice industry as a quick and efficient approach to identify brix to citric acid adulteration in pulp-wash. In the fourth study, the ensemble learning approach (based on the random subspace strategy) was implemented to develop a new calibration method (RSE-MCR-ALS-CC) based on the MCR-ALS algorithm for the quantitative measurement of various components in the sample matrix. According to the results obtained from three near-infrared data sets, the RSE-MCR-ALS-CC model based on the CCD experimental design method, with considering the weighted average approach, as a new approach and with the high power has far better results than the results of the MCR-ALS-CC method itself (with R2 prediction value above 95%), which is even better than the PLSR method (as a conventional method in solving quantitative problems in analytical chemistry). Therefore, the use of ensemble learning approaches along with portable spectrometers can be used as fast, simple, and low-cost tools for food authenticity in industries related to food quality control.
  9. Keywords:
  10. Food Authenticity ; Adulteration in Foods ; Ensemble Learning ; Classification ; Calibration ; Chemometrics Method ; Handheld Spectrometers ; Adulteration In Orange Juice ; Adulteration In Milk

 Digital Object List

 Bookmark

No TOC