Non-Destructive Quality Assessment of Fruit and Vegetables: Advances in Spectroscopy, Imaging Sensors and Artificial Intelligence
Sonali Parekh *
Department of Food Safety and Quality Assurance, College of FPT & BE, Anand Agricultural University, Anand, Gujarat, India.
Divyang Prajapati
Department of Food Processing Technology, College of FPT & BE, Anand Agricultural University, Anand, Gujarat, India.
Jerul Halpati
Department of Food Business Management, College of FPT & BE, Anand Agricultural University, Anand, Gujarat, India.
Ravi Prajapati
Department of Dairy Engineering, G N Patel College of Dairy Science, Kamdhenu University, Sardarkrushinagar, Dantiwada, Gujarat, India.
*Author to whom correspondence should be addressed.
Abstract
Non-destructive assessment can extend fruit and vegetable quality control beyond destructive subsampling, but the usefulness of a prediction depends on what the sensor observes and how its performance is validated. This critical narrative review examines spectroscopy, imaging and artificial intelligence for estimating composition, texture, maturity and defects in fresh produce. Literature published between 1 January 2000 and 27 July 2026 was identified through targeted scholarly-index searches and citation tracing, with methodological papers included to evaluate calibration, independence and uncertainty. The synthesis distinguishes direct optical sensitivity from predictions mediated by ripening, cultivar or storage conditions. Visible and near-infrared spectroscopy provide an established basis for selected compositional measurements, whereas spatially resolved optical methods, hyperspectral imaging, Raman spectroscopy, fluorescence and X-ray imaging offer complementary information with different sampling depths and operational constraints. Deep learning can improve prediction within a defined domain, yet external-season and instrument-transfer studies show that architectural complexity does not remove calibration dependence. Higher spectral resolution, additional sensors and larger numbers of image pixels do not necessarily increase independent information. Evidence is particularly vulnerable when repeated observations cross data partitions, sample-level reference values are interpreted as validated pixel maps, or laboratory classifications are presented as commercial readiness. Fruits, leafy vegetables and underground storage organs require different measurement geometries and reference strategies; they cannot be treated as interchangeable prediction tasks. A decision-centred approach is proposed in which observability, biological sampling, independent validation and operational consequences are assessed together. Priorities include prospective multi-site testing, spatially matched reference measurements, explicit uncertainty and rejection policies, and evaluations of sorting outcomes rather than model accuracy alone. The most defensible advance is therefore not a universally superior sensor or algorithm, but a traceable measurement system whose limitations remain visible when products, instruments and operating conditions change.
Keywords: Chemometrics, postharvest phenotyping, calibration transfer, optical properties, hyperspectral imaging, uncertainty quantification, external validation