Low-cost machine learning-integrated optical spectrophotometer for non-destructive color and shelf-life analysis: A...

Monitoring physical color and spectral signatures is essential for the early detection of spoilage in perishable food products. This study introduces a cost-effective, Machine Learning (ML)-enabled spectrophotometer for nondestructive spoilage detection in sliced bread. An Artificial Neural Network was developed to accurately predict reflectance spectra (340-750 nm) from RGB values. Shelf life was evaluated using Total Plate Count (TPC) alongside color and spectral data. Results highlighted a significant TPC increase in Day 8 and Day 10 samples, marking them spoiled. Color analysis revealed substantial L* value reductions after Day 4, with a* and b* shifts post-Day 8, indicating microbial growth. The chromaticity plots showed a shift to a greenish hue by Day 10. Spectral data remained stable until Day 4 and declined between 525 and 619 nm from Day 6 onwards. ML models, particularly Gradient Boosting Trees and Support Vector Machines (SVM), effectively classified samples, with SVM achieving the highest F1 score (95.84 %) in testing. These results highlight the potential of integrating low-cost optical technologies with ML models for nondestructive food spoilage detection, with promising applications for food safety and shelf-life monitoring.