Multi-modal zero-shot prediction of color trajectories in food drying
View Full Paper
Ask AI
Bookmark
Share
Discussion
Loading...
Member takes
Overview
Machine learning study demonstrates accurate zero-shot color trajectory prediction in drying foods, indicating potential for real-time automated quality control.
Key Points
To develop and evaluate a multi-modal data-driven method capable of continuous zero-shot color trajectory prediction during food drying under unseen process conditions.
Represented color trajectories as weighted sums of predefined basis functions to parameterize continuous evolution into a compact coefficient set.
Extracted and fused features from initial sample images and process settings, using a similarity-informed data selection approach to train on condition-relevant samples.
Evaluated the framework in a zero-shot setup across cookie drying and apple drying datasets, benchmarking performance against a standard long short-term memory (LSTM) model.
Achieved root mean square errors (RMSEs) of 2.12 for cookie drying and 1.29 for apple drying under completely non-overlapping zero-shot test conditions.
Reduced prediction error by greater than 90% relative to the baseline LSTM model across evaluated food drying tasks.
Demonstrated via ablation experiments that multi-modal fusion and similarity-informed training independently enhance accuracy, yielding the greatest improvements when combined.