Abstract
Evaluating the transfer of culinary skills from educators to students is essential but challenging due to the subjective nature of traditional assessment methods like direct observation. This study proposes using deep learning and image analysis, particularly convolutional neural networks (CNNs) such as VGG-16, to objectively and automatically evaluate the skill transfer by identifying and quantifying visual differences between student and instructor-prepared dishes. The results show that CNNs can effectively capture critical visual features, offering a more consistent and scalable assessment approach. However, challenges remain, including sensitivity to image quality and discrepancies between automated evaluations and human judgments. These findings highlight the need for further refinement of models and expanding datasets to better capture the diversity of real-world culinary outputs. This research lays the foundation for integrating advanced analytical techniques into culinary education, with future work focusing on developing specialized datasets, fine-tuning models, and standardizing protocols to enhance the accuracy and reliability of automated culinary assessments.
