Intelligent spectroscopy for fruit and vegetable quality assessment: From biological complexity to robust modeling

Spectroscopic techniques provide rapid, non-destructive means for fruit and vegetable quality assessment, yet the biological complexity of fruits and vegetables introduces high spectral variability and limits conventional model robustness. This review systematically examines the transition from traditional spectroscopic modeling to intelligent analysis strategies for enhanced robustness in fruit and vegetable quality detection. The fundamental optical properties of fruit and vegetable tissues are first analyzed, highlighting how cellular structures and chemical compositions contribute to spectral complexity. Subsequently, the limitations of conventional modeling approaches are critically evaluated when applied to complex biological systems, particularly their susceptibility to overfitting and poor cross-scenario performance. The core contribution focuses on intelligent modeling strategies that address these limitations through advanced machine learning techniques including deep learning architectures, multi-modal fusion approaches, transfer learning and ensemble methods. Additionally, applications across various fruit and vegetable categories demonstrate the practical effectiveness and performance improvements achieved through these intelligent strategies, with future research directions focusing on model interpretability and sustainable deployment strategies for real-world applications. This review provides essential guidance for developing dependable spectroscopic quality assessment frameworks within intelligent food processing systems.