University of Patras
Development and application of an optical laser based diagnostic technique combined with machine learning for the authentication, quality control, and adulteration detection of olive oil and milk samples
Abstract
dc:descriptionThe growing public awareness for food quality and safety, and the increasing number of food fraud cases, have led to a greater demand for fast and reliable food authentication techniques. Food fraud, which can include composition alteration, adulteration, substitution, mislabeling, etc., is a worldwide issue affecting various food industry sectors. These deceptive practices pose significant health risks to consumers, particularly when harmful or inferior substances are used in food products. Furthermore, fraudulent activities result in substantial economic losses and reputational damage for producers and retailers. High-value foodstuffs, such as dairy products and premium olive oils, are especially susceptible to fraud due to their economic and nutritional significance. In addition to consumer-related concerns, food fraud presents challenges for regulatory authorities, who constantly try to enforce compliance and maintain food standards within an evolving global market. The traditional analytical methods applied for food authentication, while effective, can be time-consuming, costly, and require qualified personnel and extensive sample preparation. As a result, there is a growing need for rapid, accurate, and cost-effective analytical techniques that can ensure food authenticity, detect adulteration, and enhance traceability. In this context, Laser-Induced Breakdown Spectroscopy (LIBS) has emerged as a powerful, and very rapid non-destructive spectroscopic technique for food analysis. LIBS is a laser based spectroscopic technique that uses a focused laser beam to create a plasma onto the surface of a sample, which then emits light. The emitted light is collected, and analyzed, enabling the identification and quantification of the elemental composition of the sample. One of the key advantages of LIBS is its ability to perform rapid, real-time, and in-situ analysis, often requiring minimal or not at all sample preparation. Furthermore, LIBS can detect almost all of the elements of the periodic table, many of which serve as critical markers for determining food authentication, quality, and safety. The versatility of LIBS extends to its capability to analyze various food matrices, including solid, liquid, and powdered samples. Recent advancements in LIBS technology, particularly its combination with machine learning, have significantly enhanced the related data processing capabilities, allowing for the rapid interpretation of complex spectral data, and delivering highly accurate results. As a result, LIBS is considered a cutting-edge tool for food fraud prevention, offering an innovative approach to ensuring food transparency. The present thesis investigates the combined use of LIBS technique and machine learning on food authentication-related issues. In particular, the identification of origin, detection of adulteration, and quality control, with emphasis on milk and olive oil, are explored. To achieve these objectives, an extensive series of LIBS experiments was conducted on one of the largest sample collections analyzed to date. Over 1000 milk samples from different animal origins and extra virgin olive oil (EVOO) samples from several renowned olive oil-producing regions in Greece have been studied. Numerous machine learning algorithms have been employed to analyze the LIBS spectroscopic data, thoroughly assessing their performance and effectiveness. Comparative studies have also been performed, evaluating the capabilities of LIBS and comparing its results with those from other spectroscopic techniques, such as UV-Vis-NIR absorption spectroscopy. In addition, some data fusion strategies have been explored, demonstrating the benefits of the combination of LIBS with other spectroscopic techniques. In the first chapter of the present thesis, a comprehensive review of LIBS foods-related applications, emphasizing staple foods, such as olive oil, milk, and honey, is presented. A detailed discussion of the principles of LIBS, including the fundamental physical processes, technical aspects, and instrumentation-related issues, is given. A description of machine learning methods often combined with LIBS is also provided. Emphasis is given to the learning styles, along with the challenges and some major pitfalls associated with the application of machine learning in LIBS analysis. Following this theoretical background, the recent advancements in LIBS-based authentication studies for olive oil, milk, and honey are reviewed. The focus is placed on key applications, including the identification of geographical, cultivar, animal, and botanical origin, and the detection of adulteration. Next, in Chapters 2 and 3, the application of LIBS for milk analysis is presented. In more detail, in Chapter 2, the classification of liquid and powdered milk samples, based on their animal origin, i.e., cow, goat, and sheep milk, is examined, using nanosecond (ns) laser pulses, while in Chapter 3 only lyophilized powder milk samples were studied using femtosecond (fs) laser pulses. In all cases, the collected spectroscopic data were analyzed by several machine learning algorithms, including Linear Discriminant Analysis (LDA), Multilayer Perceptron (MLP) neural network, Support Vector Machine (SVMs), Gradient Boosting (GB), and Logistic Regression (LR). A thorough evaluation of the performance of these algorithms was conducted, focusing particularly on their predictive performance. Moreover, the impact of the spectral features of the major inorganic components of milk, i.e., magnesium (Mg), calcium (Ca), sodium (Na), and potassium (K) were also studied in terms of milk’s samples classification. In Chapters 4 and 5, the potential of LIBS and UV-Vis-NIR absorption spectroscopy for the discrimination of pure EVOOs from EVOOs adulterated with lower-quality edible oils (i.e., corn, pomace, soybean, and sunflower oils) and the identification of the type of adulterant are studied. In Chapter 4, the LIBS related results are presented, while Chapter 5 the results from the combined study, i.e., using both LIBS and UV-Vis-NIR absorption spectroscopy are presented and discussed. For the analysis of the spectroscopic data, the LDA, SVMs, GB, and LR algorithms were used and are thoroughly assessed. Finally, the comparison of the performance of the two spectroscopic techniques is presented.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Nanou, Eleni
- Νάνου, Ελένη
Subjects
dc:subject × 18Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58753