Few-Shot Ordinal Learning for Day-Wise Freshness Estimation with Hyperspectral Fish Images
A research paper introduces a few-shot learning framework for hyperspectral imaging-based food quality estimation, specifically day-wise freshness of salmon fillets. The method uses a CORAL-style ordinal prediction head and biologically grounded constraints. On a 16-day salmon HSI dataset under an unseen-fillet protocol, it achieves a mean absolute error of 1.58 days and 2-day accuracy of 72.3% with only three labelled days per fillet, outperforming scalar regression.
Researchers propose the first few-shot learning framework for hyperspectral imaging (HSI) based food quality estimation. Each fillet is treated as a distinct episodic task, and a CORAL-style ordinal prediction head models the ranked nature of freshness progression. Biologically grounded monotonicity and embedding smoothness constraints guide predictions. Evaluated on a 16-day salmon HSI dataset with a strict unseen-fillet protocol, the method achieves a mean absolute error of 1.58 days and 2-day accuracy of 72.3% using only three labelled days per fillet, substantially outperforming scalar regression.
The approach leverages episodic few-shot learning to address inter-fillet variability and scarce labels. The ordinal prediction head with cumulative threshold modelling captures the ordered nature of freshness days. Monotonicity and smoothness constraints inject domain knowledge, improving generalization to unseen fillets. The reported metrics suggest that few-shot ordinal learning can reduce annotation costs while maintaining accuracy in HSI-based quality assessment.
This research could lower the barrier for deploying HSI-based freshness monitoring in food processing by reducing the need for extensive per-product labelled data. It may enable faster adaptation to new fish species or products. The method's success on salmon suggests potential for broader application in perishable food supply chains, where non-destructive, day-wise freshness estimation is valuable.
The technology could reduce labour and time for data annotation in food quality control, enabling more scalable deployment of HSI systems. It may improve inventory management and reduce waste by providing accurate day-wise freshness estimates. Potential applications include seafood processing, retail quality assurance, and supply chain monitoring.
Next signals include validation on other food types, integration with real-time HSI systems, and comparison with semi-supervised or self-supervised methods. Commercial adoption may depend on demonstrating robustness across varying storage conditions and species. Further research could explore extending the framework to other ordinal regression tasks in quality control.