Post-harvest Quality Control of White Quinoa on Smartphones Using Deep Learning

Salazar-Campos O.; Tafur-Vera D.; Lecca-Pino P.; Moran-Ruiz J.; Salazar-Campos J. · Food and Bioprocess Technology (2026) · DOI: 10.1007/s11947-026-04443-6

Resumen

The global expansion of quinoa requires optimized, non-invasive, objective, and standardized methodologies to ensure uniform post-harvest quality. In this study, we developed and validated a lightweight, on-device artificial vision system (AVS) for the regulatory classification of white quinoa in accordance with the Peruvian standard NTP 011.462:2025. A total of 3000 sample units (SUs) from Puno (Perú) were analyzed. Individual grains were segmented into 64 × 64 px regions of interest (ROIs) and classified using a convolutional neural network (CNN) based on MobileNet_v1_0.50_224, evaluated through fivefold GroupKFold cross-validation at the SU level. Performance metrics were estimated with 95% confidence intervals (CIs), and probabilistic calibration was applied via temperature scaling (TS). The system classified grains into four size classes and five morphological categories, enabling type and regulatory category assignment via a rule-based engine aligned with NTP thresholds. The AVS achieved 97.14% accuracy, 94.15% F1_macro, and a Matthews correlation coefficient (MCC) of 0.94, with specificities above 99% for size and 98% for category classification. Calibration substantially improved reliability, reducing the expected calibration error (ECE) to 0.023 for category classification, enabling stable, risk-aware decision thresholds without loss of discrimination. On-device inference required 0.345 s per sample and 0.188 s for image capture on a mid-range Android device. Overall, the proposed AVS provides an objective, reproducible, and auditable decision-support tool aligned with regulatory criteria for quinoa quality control, offering a practical reference for other grains and agro-food matrices under standardized morphological and chromatic criteria. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2026.

Instituciones: Universidad San Ignacio de Loyola; Universidad Privada del Norte