Recent advances in spectroscopy and machine learning for non-destructive and real-time detection of mycotoxins in cereals

Abstract

Background

The crucial role of cereals in the food chain is devastated by mycotoxins that cause a harmful impacts on animals and humans. For their detection, the conventional approaches require burdensome pretreatment, are time-consuming, and destructive in nature. To overcome this issue, AI-driven (machine learning – ML, and deep learning – DL) spectroscopic techniques have shown potential as a groundbreaking tool, offering optimal solutions, accuracy, and precision through optimization. However, its understanding of practical implications is still limited and necessitates further exploration.

Scope and approach

This study synthesizes the applications of ML and spectroscopic techniques (multi and hyperspectral imaging and non-imaging, raman spectroscopy, visible-infrared spectroscopy, fluorescence spectroscopy, and nuclear magnetic resonance), considering mycotoxins detection in cereals (wheat, maize, and rice). Moreover, this review also encompasses the functioning principles, interaction of spectroscopic lights, data pre-processing, feature optimization, ML-based predictive modeling, and validation of results for decision-making and their applications.

Key findings and conclusions

Developing a viable spectroscopic based mycotoxins detection system driven by ML requires a comprehensive optimization process. This includes fine-tuning the ML model itself and carefully selecting and balancing several components: dataset size, preprocessing approaches, features’ selection and extraction strategies, model architecture, and hyperparameter tuning through validation. Furthermore, while ML algorithms are advancing rapidly, designing a specialized and robust model specifically for spectroscopic mycotoxin detection remains an active and evolving area of research.