Concept of Real-Time Prediction and Evaluation System of Robotic Food Chewing Using Machine Vision and Deep Learning.

The evaluation and prediction of food state during chewing are important for understanding and imitating the chewing process, which has implications for the food industry. This study aims to develop a prediction and evaluation system deep learning vision system to analyze changes in Particle Size Distribution (PSD) and features of solid food during mastication. A stereoscopic vision system with multiple cameras will be designed to capture all food samples from different angles. A chewing robot will be used under the supervision of this vision system to collect images of food samples, creating a database. This database will be used to train a deep learning module capable of understanding PSD and bolus feature changes. An application tool will be developed to automate the manual PSD processing method, providing real-time analysis of images captured by the vision system and instant display of the processed outcomes. This enables real-time assessment and prediction of food PSD and bolus features during chewing. The predicted results can be compared with actual chewing outcomes to evaluate the robot's performance and potentially assess human chewing health. Future work will focus on improving the model's precision, expanding its applicability to various food categories, and refining the training dataset. This investigation has the potential to advance flavorful and nutritious food production, benefiting the food industry and consumers.