Food Research Institute

Yusun, S. – Yılmaz, O. – Tarlak, F.
Deep learning algorithm for dessert recognition and nutritional evaluation
Journal of Food and Nutrition Research, 64, 2025, No. 1, s. 61-71

Fatih Tarlak, Department of Bioengineering, Faculty of Engineering, Gebze Technical University, Cumhuriyet Street 2254, 41400 Gebze, Kocaeli, Turkey. E-mail: ftarlak@gtu.edu.tr

Original article
Received 14 August 2024; 1st revised 29 November 2024; accepted 13 January 2025; published online 14 March 2025.

DOI: https://doi.org/10.64122/OKGF7329

This article contains supplementary data.

Summary: Food recognition systems are crucial for healthcare and the food industry, aiding in diet tracking, personalised meal planning, and promoting nutritional awareness. This work develops a software interface that recognises food products using deep learning algorithms, and announces their nutritional values and gastronomic characteristics. Specifically, photographs of various desserts were captured in a restaurant setting, and the classification performance of two deep learning models, GoogleNet and ResNet-50, was analysed. Both models achieved high accuracy rates exceeding 99.6 %, with ResNet-50 demonstrating superior performance due to its lower error rates, higher accuracy, and faster learning capabilities. Based on these results, the interface was developed using ResNet-50 to provide consumers with detailed gastronomic information about desserts and support healthier dessert choices. At present, the resulting software is limited to the 23 dessert items on the menu of Healin restaurant (Nisantasi, Istanbul, Turkey), and the way they look in that particular restaurant, but the scope could be expanded in the future. This innovative approach enhances consumer awareness about healthy eating while offering a competitive edge for the food industry by effectively meeting consumer expectations.

Keywords: deep learning algorithms; food recognition; nutritional evaluation; dessert classification

Download:
  (pdf, 994.5 Kb, 542x)
  (pdf, 111.62 Kb, 635x)