Deep Learning-Aided Spectroscopic Approaches for Assessment of Beef Freshness and Authenticity

Abstract

The overarching aim of this dissertation is to develop and validate novel methodologies for assessing meat quality, with a specific focus on beef freshness, authenticity, and the identification of counterfeited beef. Given the increasing consumer demand for high-quality meat products and the prevalent issue of food fraud, particularly in the meat industry, this study employs spectroscopic approaches combined with deep learning algorithms to overcome the limitations of traditional meat evaluation methods. Chapter 1 of this study provides a comprehensive overview of the current trends in global meat consumption, emphasizing the escalating demand for high-quality meat products due to rising incomes and population growth. It highlights the critical role of freshness in influencing consumer purchasing decisions, underscoring the need for innovative and reliable methods to assess meat quality and safety. The chapter methodically outlines factors leading to meat spoilage. Additionally, it elaborates on the dynamic changes in myoglobin content within meat over storage time, knowledge that is critical for the comprehensive discussions in the ensuing chapters. Chapter 2 delves into the principles of diffuse reflectance spectroscopy (DRS), the cornerstone technique of this study. It begins by elucidating the fundamentals of light absorption and scattering within biological tissues, subsequently progressing to the formulation of a steady-state diffuse reflectance model through diffusion approximation derived from a radiative transfer model. This model is pivotal for extracting myoglobin content from beef samples by addressing an inverse problem. Chapter 3 of this study provides an overview of the machine learning and deep learning techniques employed in developing classification models for assessing beef freshness and authenticity using diffuse reflectance spectrum and myoglobin information. Chapter 4 presents a detailed exploration of advanced methodology for assessing beef freshness, employing deep learning algorithms and diffuse reflectance spectroscopy (DRS). This chapter elucidates the integration of DRS with deep learning, particularly convolutional neural networks (CNNs), to analyze spectral data for meat evaluation. It discusses the use of DRS in quantifying myoglobin redox forms in beef, a critical factor in determining meat color. It also examines the impact of integrating myoglobin information into the classification process, utilizing gradient-weighted class activation mapping (Grad-CAM) to visualize influential spectral regions. This approach underscores the potential of combining spectroscopic data with deep learning techniques in enhancing the accuracy and efficiency of beef freshness assessment. Chapter 5 of this study addresses the critical issue of food fraud in the meat industry, focusing on the development and validation of an innovative approach to detect counterfeited beef using DRS and deep learning. The chapter then introduces a CNN model that, combined with DRS, effectively distinguishes between fresh and counterfeited beef based solely on spectral data. This novel method, validated with various beef samples, demonstrates a high level of accuracy in detecting counterfeited beef. By leveraging Grad-CAM for identifying crucial spectral regions, this approach marks a significant advancement in ensuring meat authenticity and protecting consumer health, showcasing the potential of integrating spectroscopy with deep learning in food safety. Lastly, Chapter 6 concludes the study, summarizing the development of a deep learning-aided spectroscopic approaches for meat freshness and authenticity assessment and addressing its superiority over traditional methods. The chapter also outlines future research directions for enhancing the scope and application of this method in the meat industry, emphasizing the need for broader experimental validation and technological refinement. Overall, this study contributes to the fields of food science and technology by presenting advanced and practical solutions for meat freshness and authenticity assessment. The combination of DRS and deep learning opens new avenues for rapid, non-destructive, and reliable meat freshness and authenticity assessment, providing valuable insights for both industry practitioners and researchers.