Abstract:
In the area of culinary culture, the classification of desserts provides a fascinating
problem, particularly due to the complicated nature and vast assortment of dessert
styles found throughout many regions. This work offers a rigorously curated dataset of
dessert photographs, specifically optimized for Bangladeshi dessert classification. Our
dataset features a comprehensive range of high-quality photos showing the lively
diversity of traditional Bangladeshi sweets, showcasing the richness and complexity of
the country's culinary heritage. The project intends to construct strong dessert
recognition models by leveraging multiple visual processing techniques and deep
learning algorithms, including MobileNet. Through comprehensive examination
utilizing established parameters, our models achieve an astonishing 98% overall test
accuracy. This work offers as a significant resource for scholars and practitioners
digging into culinary picture classification, with a special focus on Bangladeshi
desserts. By offering access to such a rich dataset, I intend to drive improvements in
machine learning techniques targeted to culinary applications, thereby contributing to
the preservation of culinary legacy and the promotion of cultural diversity via the lens
of technology.
