Integration of ResNet50 Architecture in Food Image Detection Systems for Dynamic Nutrition Estimation

Icha Maulidya (Sekolah Vokasi IPB University)
Siti Farah Fakhirah (Sekolah Vokasi IPB University)
Jonser Steven Rajali Manik (Sekolah Vokasi IPB University)
Asa Yuaziva (Sekolah Vokasi IPB University)

Journal of Applied Science, Technology & Humanities
Vol. 3 No. 3 (2026): June 2026

The development of modern lifestyles demands accurate food intake management to maintain health quality. This study aims to assess NutriScan, an online tool for identifying the type of citra and calculating calories automatically. The AICrowd Food Recognition Challenge dataset is used for quantitative modeling utilizing the Convolutional Neural Network (CNN) ResNet50 architecture based on transfer learning, and nutritional data is integrated via Application Programming Interface (API). Data augmentation, cropping, scaling, and normalization are used in the processing of Citra. According to the study's findings, NutriScan has the potential to be an effective and adaptable web-based solution with an accuracy of roughly 75%, precision of 76%, recall of 75%, and F1-score of 75%