Lightweight Deep Learning Framework for Quality Assessment of Rolled Fondant

Key Points

• To develop a lightweight deep learning framework that assesses the quality of rolled fondant using color metrics.
• Utilized CIELAB-based lightweight convolutional neural networks (CNNs) to analyze food colorimetry.
• Conducted statistical validation to ensure metrological reliability with a 95% confidence interval.
• Implemented semantic segmentation techniques to improve handling of continuous textures.
• Parameter-efficient models with less than 800k parameters outperformed YOLO in reliability.
• Achieved successful segmentation of textures, overcoming bounding-box saturation issues.
• Validated edge-device capability for final-state quality control in food products.