Food Recognition And Recipe Recommendation For Cebuano Cuisine Using An Ensemble Of Modified CNN And Web Scraping

Knowledge surrounding Cebuano cuisine, an intangible cultural heritage, continues to diminish due to the effects of globalization and the reliance on imports. Existing food computing solutions that preserve culinary heritage primarily focus on major cuisines such as Japanese, Chinese, and American. Thus, there is a gap for lesser-known cuisines like Cebuano cuisine, which has no existing food image datasets and deep learning models. To address these challenges and simultaneously promote tourism, there is a need to develop a food recognition and recipe recommendation system that shares and preserves local culinary knowledge of Cebuano cuisine culture. To achieve this, the researchers proposed an ensemble of pre-trained Convolutional Neural Network (CNN) models—specifically VGG16, DenseNet201, and InceptionV3—enhanced by a Self-Attention Mechanism (SAM) to identify Cebuano cuisine. Due to the lack of a dataset for Cebuano cuisine, the researchers created one based on the book “Hikay: The Culinary Heritage of Cebu” and a web article on famous Cebu delicacies, which is gathered via web scraping and data augmentation techniques such as rotation, flipping, shifting, saturation, and exposure adjustment. The dataset comprised 32,400 images across 27 classes. The ensemble learning attained an accuracy of 96.96%, suggesting a high rate of correct Cebuano cuisine predictions. The finding suggests that the developed system is effective in recognizing and promoting Cebuano culinary heritage. By leveraging advanced CNN models and enhancing them with a Self-Attention Mechanism, the study demonstrates the potential of AI in preserving cultural knowledge.