Multi-Source Imaging and Machine Learning for Non-Destructive Quality Assessment of Edible Root and Tuber Crops: A Review

Edible roots and tubers, such as potato, sweetpotato, cassava, and carrot, are vital to global food security; however, postharvest quality control still relies largely on costly, subjective manual inspection. Computer vision and machine learning applied to multiple imaging modalities have emerged as non-destructive approaches for automated inspection and quality assessment. This review synthesizes 141 studies (2016–2026) that applied these methods to the non-destructive quality assessment of edible roots and tubers. The reviewed studies span eight species, with potato (77.3%) and sweetpotato (17.0%) being dominant. RGB imaging is the most common acquisition modality (79%), alongside the growing use of hyperspectral and near-infrared spectroscopy. Deep learning is now the default approach, with its share rising from 18% (2016–2020) to 71% (2024–2026). YOLO-family detectors lead real-time sorting and defect detection, while classical image processing and machine learning remain competitive for narrowly defined tasks such as size grading. Several gaps stand out. Evidence beyond potato and sweetpotato remains thin. Public benchmarks are almost nonexistent, with only 5 of 141 studies using a public dataset. Foundation and zero-/few-shot models were almost absent from the reviewed studies, and most studies measure benchtop accuracy rather than the full inspection-to-actuation pipeline. This review maps where the field has matured and where it has not, offering a structured reference point and a research agenda for non-destructive quality evaluation of these crops.