Artificial Intelligence and Multispectral Imaging for Common Bean Seed Quality Analysis: A Critical Review of Evidence, Translation Pathways and Research Priorities

Abstract

Seed quality testing in common bean must resolve several distinct questions: whether seed is genetically and physically authentic, whether it will germinate and establish reliably, whether it carries damaging defects or pathogens, and whether its composition supports intended food-processing uses. Conventional assays remain indispensable, but many are destructive, labour-intensive or too slow for high-throughput decisions. Artificial intelligence applied to multispectral and hyperspectral imaging offers a non-destructive route to combine surface morphology, spatially resolved reflectance and, in some systems, autofluorescence. This critical narrative review evaluates the evidence for these approaches in common bean and examines what can reasonably be inferred from studies in other crop species. The eligible literature window extended from 1 January 2000 to 26 May 2026, with verified foundational sources included where necessary. Direct common bean evidence supports proof-of-concept discrimination of hard seeds, ageing-defined vigour classes, varieties and accessions, and selected processing traits. Yet the most impressive accuracies were generally obtained under controlled acquisition conditions, with restricted genetic or environmental domains, and with limited evidence of external validation across laboratories, seasons, instruments or commercial seed lots. Cross-species research shows that reduced-band multispectral systems can capture physiologically relevant variation, particularly through visible, near-infrared and chlorophyll-related signals, but it also demonstrates weak transfer across cultivars and production contexts. The central limitation is therefore not model capacity; it is the validity and transportability of the reference labels, sampling design and validation scheme. Multispectral imaging is presently best positioned as a rapid triage, phenotyping and decision-support tool that complements confirmatory germination, health and identity testing rather than replacing them. Progress towards routine use in common bean requires lot- and season-aware external validation, prospective multi-site studies, transparent preprocessing and feature-selection pipelines, model calibration monitoring, and benchmark datasets that preserve biological hierarchy. A staged translation pathway from hyperspectral discovery to purpose-built multispectral devices is proposed, with priorities for robust seed-lot inference, pathogen-specific testing and climate-resilient quality surveillance.