AI and Laser-Induced Spectroscopy for Food Industry

Zhou Y., Jiao L., Wu J., Zhang Y., Zhu Q., Dong D.

There is strong interest in non-destructive and rapid determination of food freshness in food research. In this study, mid-infrared (MIR) fiber-optic evanescent wave (FOEW) spectroscopy was applied to monitor shrimp freshness through the evaluation of protein, chitin, and calcite contents in conjunction with a Partial Least Squares Discriminant Analysis (PLS-DA) model. Shrimp shells were wiped with a micro fiber-optic probe to obtain a FOEW spectrum which quickly and nondestructively allowed evaluation of the shrimp freshness. Peaks for proteins, chitin, and calcite, which are closely related to shrimp freshness, were detected and quantified. Compared with the standard indicator for evaluating shrimp freshness (total volatile basic nitrogen), the PLS-DA model gave recognition rates for shrimp freshness using calibration and validation sets of the FOEW data of 87.27%, 90.28%, respectively. Our results show that FOEW spectroscopy is a feasible method for non-destructive and in-site detection of shrimp freshness.

Shin S., Wu X., Patsekin V., Doh I., Bae E., Robinson J.P., Rajwa B.

Laser-induced breakdown spectroscopy (LIBS) is a widely acknowledged elemental analysis approach used in various study domains due to its rapid measurement capability and minimal sample-preparation requirements. Recently, there has been an increase in interest in the applications of LIBS in the realm of food safety and quality. Given that the majority of commonly consumed foods exhibit only modest trace-element variations, discovering predictive spectral patterns through multivariate analysis is crucial for the data-analysis pipeline. The efficacy of multivariate analysis and machine-learning algorithms to identify the most predictive spectral features, conduct class recognition and classification was evaluated in this paper, utilizing both a custom-developed benchtop LIBS system and a commercially available portable one. Specifically, this study's objective was to evaluate the performance of spectral variable selection using elastic-net multinomial logistic regression. The data processing pipeline and the LIBS hardware were evaluated in the context of food authentication and identification, a rising field of research addressing the issue of food fraud. Our findings indicated that classifying food samples with carefully selected fewer variables reduces model overfitting and improves the accuracy of LIBS pattern classification. In a broader sense, the results support the continued development of field-deployable, portable LIBS equipment designed for food authentication and fingerprinting activities.

Shin S., Doh I., Rajwa B., Bae E., Robinson J.P.

Wei K., Teng G., Wang Q., Xu X., Zhao Z., Liu H., Bao M., Zheng Y., Luo T., Lu B.

Fritillaria has a long history in China, and it can be consumed as medicine and food. Owing to the high cost of Fritillaria cirrhosa, traders sometimes mix it with the cheaper Fritillaria thunbergii powder to make profit. Herein, we proposed a laser-induced breakdown spectroscopy (LIBS) technique to test the adulteration present in the sample of Fritillaria cirrhosa powder. Experimental samples with different adulteration levels were prepared, and their LIBS spectra were obtained. Partial least squares regression (PLSR) was adopted as the quantitative analysis model to compare the effects of four data standardization methods, namely, mean centring, normalization by total area, standard normal variable, and normalization by the maximum, on the performance of the PLSR model. Principal component analysis and least absolute shrinkage and selection operator (LASSO) were utilized for feature extraction and feature selection, and the performance of the PLSR model was determined based on its quantitative analysis. Subsequently, the optimal number of features was determined. The residuals were corrected using support vector regression (SVR). The mean absolute error and root mean square error of prediction obtained from the quantitative analysis results of the combined LASSO-PLSR-SVR model for the test set data were 5.0396% and 7.2491%, respectively, and the coefficient of determination R2 was 0.9983. The results showed that the LIBS technique can be adopted to test adulteration in the sample of Fritillaria cirrhosa powder and has potential applications in drug quality control.

Van den Eynde S., Díaz-Romero D.J., Zaplana I., Peeters J.

One of the most promising innovation strategies for sorting and recycling post-consumer aluminium scrap is using quantitative Laser-Induced Breakdown Spectroscopy (LIBS) analysis. However, existing methods to estimate alloying element concentrations based on LIBS spectra, such as linear univariate regression and Machine Learning models, are still too limited in their performance to achieve the accuracy demanded by the industry. Therefore, this study presents novel Deep Learning approaches and compares their performance to those of traditional univariate regression and Machine Learning methods in terms of RMSE, MAE, and R2 value. For this evaluation, two sample sets of aluminium pieces are used: one containing 27 certified aluminium reference samples and the second containing 733 post-consumer scrap pieces for which the ground truth concentrations are determined by X-Ray Fluorescence (XRF). Adopting multiple loss functions, one for each element, has proven its significant value for the regression performance. It improves the results for all performance metrics in the Scrap Sample set, and the same is true for the Reference Sample set, except for the coefficient of determination of Fe, Mn and Mg. In addition, the proposed methodology considers the learning prioritisation problem to prevent that learning the concentration of the base element is prioritised over the alloying elements. Although the effect of excluding the base alloy aluminium from the learning is small and not always positive for the performance, demonstrating this effect is also considered valuable. Since the average RMSE on the prediction is just 0.02 wt% for Al and Si, and not more than 0.01 wt% for Fe, Cu, Mn, Mg, and Zn, the best-performing Deep Learning model shows promise for the future of LIBS in metal sorting applications.

Brar D.S., Pant K., Krishnan R., Kaur S., Rasane P., Nanda V., Saxena S., Gautam S.

Globally, honey is consumed as natural functional food, featuring high economic value, related to its authenticity and purity. Due to the high demand for honey, it has been wildly targeted for food adulteration with substandard honey or low-price syrups, and emerging absorbent resin technology has also affected the honey market. Researchers worldwide have been working persistently to invent and innovate advanced technologies for detecting honey adulteration and assuring its authenticity. The compositional intricacy of pure and unauthentic honey could be easily detangled by a combined approach of chemometrics and instrument. In this palimpsest, various detection methods like ISCIRA, NMR, AT-FTIR, Sensors, PCR based assay united with an appropriate Multivariate approach that provides accurate and acceptable results for the determination of honey authentication and adulteration were described. The botanical origin authentication of honey was determined with the application of δ13C-EA-IRMS and δ13C-LC-IRMS coupled SVM, which discriminate samples based on specific markers. LIBS, NMR, HPTLC, UHPLC, GC, and real-time PCR, generated data and processed with LDA, OPLS, PCA, ANN, CNN etc. The generated data discriminate adulterated honey from pure, as NMR clubbed with PCA can detect 1% of adulterants in honey. Hence, various methodologies in chemometrics have manifested their proficiency during ground application. Perhaps, there is a long way to go in this field to develop a universal technology for honey authentication and adulteration detection.

Brunnbauer L., Gajarska Z., Lohninger H., Limbeck A.

LIBS-based classification has experienced an ever-increasing interest in the last few years. LIBS is a well-suited technique for classification tasks based on elemental fingerprinting, providing fast simultaneous multi-element analysis with stand-off, online, and portable capabilities. The topic of classification gained even more momentum due to the current hype on machine learning, big data, and chemometrics. Nevertheless, with many LIBS users not being data scientists by training, classification algorithms are often used and considered “black boxes,” hindering the adequate application of these tools. This review provides a comprehensive introduction and overview of the steps necessary (e.g., normalization, background correction, feature selection) to go from recorded data to a well-performing classifier. Additionally, the basic principles, advantages, and limitations of the most used machine learning algorithms reported in LIBS-classification literature are discussed. Finally, the review offers an overview of the literature published in the field, highlighting the great diversity of applications.

Peng J., Liu Y., Ye L., Jiang J., Zhou F., Liu F., Huang J.

Minerals in rice leaves is a crucial indicator of plant health, and their concentrations can be used to guide plant management. It is important to predict mineral content in contaminated rice rapidly. In this study, laser-induced breakdown spectroscopy (LIBS) was applied to quantify minerals (Ca, Cu, Fe, K, Mg, Mn, and Na) in rice leaves under chromium (Cr) stress. Two feature extraction methods, including principal component analysis (PCA) and extreme gradient boosting (XGBoost), were compared to identify important variables that related to mineral concentrations. Results showed that partial least square regression (PLSR) achieved good performance in Ca, Fe Mg, K, Mn, and Na, with correlation coefficient of 0.9782, 0.8712, 0.8933, 0.9206, 0.9856, and 0.9865, root mean square error of 219.25, 14.78, 1192.47, 385.12, 9.56, and 124.32 mg/kg, respectively. In addition, the correlation between different spectral lines were further analyzed. Cr exhibited a positive correlation with Ca, Mg, and Na, and a negative correlation with Mn, Cu, and K. The proposed method provides a high-accuracy and fast approach for minerals prediction in rice leaves under Cr stress, which is important for environmental protection and food safety.

Wu X., Shin S., Gondhalekar C., Patsekin V., Bae E., Robinson J.P., Rajwa B.

Laser-induced breakdown spectroscopy (LIBS) is an atomic-emission spectroscopy technique that employs a focused laser beam to produce microplasma. Although LIBS was designed for applications in the field of materials science, it has lately been proposed as a method for the compositional analysis of agricultural goods. We deployed commercial handheld LIBS equipment to illustrate the performance of this promising optical technology in the context of food authentication, as the growing incidence of food fraud necessitates the development of novel portable methods for detection. We focused on regional agricultural commodities such as European Alpine-style cheeses, coffee, spices, balsamic vinegar, and vanilla extracts. Liquid examples, including seven balsamic vinegar products and six representatives of vanilla extract, were measured on a nitrocellulose membrane. No sample preparation was required for solid foods, which consisted of seven brands of coffee beans, sixteen varieties of Alpine-style cheeses, and eight different spices. The pre-processed and standardized LIBS spectra were used to train and test the elastic net-regularized multinomial classifier. The performance of the portable and benchtop LIBS systems was compared and described. The results indicate that field-deployable, portable LIBS devices provide a robust, accurate, and simple-to-use platform for agricultural product verification that requires minimal sample preparation, if any.

Birkmaier A., Imeri A., Riester M., Reiner G.

Grundy H.H., Brown L.C., Romero M.R., Donarski J.A.

Offal tissues carry a lower market value compared to skeletal meats in some global markets. The inclusion of offal in any meat product in the EU and UK must be declared on the label. While many technologies have been applied to the challenge of determining adulteration with offal in meat products, no single method has been recognised and validated as a reliable test to support legislative requirements. This literature review investigated appropriate methods to determine the adulteration of meat with offal. The aim was to identify technologies suitable for future validation to underpin a high throughput, low-cost method suitable for application by enforcement laboratories. Considering all of the methods, technologies which determine elemental composition and peptide markers were particularly highlighted as demonstrating potential for future development to determine a wide range of offal tissues to support the safety and integrity of the food chain.

Jeong S., Seol D., Kim H., Lee Y., Nam S., An J., Chung H.

Laser-induced breakdown spectroscopy (LIBS) and near-infrared (NIR) spectroscopy were combined to enhance discrimination of soybean paste samples according to geographical origin. Since element and organic component compositions of soybean pastes depend on soybean cultivation areas and fermentation conditions, utilization of two complementary spectroscopic signatures would be synergetic for the discrimination. When the areas of C (AC) and Ca (ACa) peaks in the LIBS spectra were used as the inputs for linear discriminant analysis, the accuracy was 95.4%. The accuracy became 92.1%, when the principal component (PC) scores obtained by principal component analysis of the NIR spectra were employed. To enhance NIR discrimination, two-trace two-dimensional (2T2D) correlation analysis was adopted to recognize minute spectral differences. With using the 1st/2nd PC scores of 2T2D slice spectra, accuracy increased to 95.0%. When the ratios of ACa/AC and the 2nd PC scores of the samples were combined together, the accuracy improved to 99.6%.

Ren L., Tian Y., Yang X., Wang Q., Wang L., Geng X., Wang K., Du Z., Li Y., Lin H.

There has been an increasing demand for the rapid verification of fish authenticity and the detection of adulteration. In this work, we combined LIBS and Raman spectroscopy for the fish species identification for the first time. Two machine learning methods of SVM and CNN are used to establish the classification models based on the LIBS and Raman data obtained from 13 types of fish species. Data fusion strategies including low-level, mid-level and high-level fusions are used for the combination of LIBS and Raman data. It shows that all these data fusion strategies offer a significant improvement in fish classification compared with the individual LIBS or Raman data, and the CNN model works more powerfully than the SVM model. The low-level fusion CNN model provides a best classification accuracy of 98.2%, while the mid-level fusion involved with feature selection improves the computing efficiency and gains the interpretability of CNN.

Liu Y., Zhao S., Gao X., Fu S., Chao Song, Dou Y., Shaozhong Song, Qi C., Lin J.

Combined laser-induced breakdown spectroscopy (LIBS) and hyperspectral imaging (HSI) with machine learning algorithms can be used to identify rice quality and the place of origin of rice production rapidly and accurately.

Qu C., Li Y., Du S., Geng Y., Su M., Liu H.

Meat nutrition and safety is highly related to people's health and quality of life. There is a huge demand to rapidly analyze meat quality during product processing and storage, but few rapid detection tools. Traditional strategies have certain disadvantages, including time-consuming, expensive, damage to samples, and the need for professional operators. Nowadays, Raman spectroscopy is drawing more and more attention due to its potential in fingerprint, specificity, speed, non-destructive and portable. This comprehensive review first briefly introduces the principles of meat analysis by common Raman techniques, e.g. Raman spectroscopy, surface-enhanced Raman spectroscopy (SERS), Raman chemical imaging (RCI), and spatially offset Raman spectroscopy (SORS), and then focus on their analytical applications on structure analysis, quality evaluation, and security control. This review also prospects the future development trend and challenges in detecting and analyzing meat and meat products.