Light-Weight Intelligent Egyptian Food Detector For Diabetes Management

Abstract

Diabetic patients need a management tool that combines multiple features and tracks and views detailed data time-efficiently. Effective food logging is an important element of health monitoring. In this paper, we propose “Suger.ly”, a lightweight mobile application with artificial intelligence food recognition for diabetes management. The system has been trained to recognize 101 distinct types of food, with a focus on Egyptian cuisine. The app can then get nutritional value and insulin calculations. The results obtained from the Single-Shot multibox Detection (SSD) MobileNet-V1 food detection model localization process, and a separate single-plates classifier achieved a mean average precision (mAP) of 0.592 and mean average recall (mAR) of 0.556. The average response time for the food object detection model in the case of a single food item detection is 74 ms.