Leftovers Food Recognition using Deep Neural Network and Regression Approach for Objective Visual Analysis Estimation

Abstract

Understanding the nutritional intake is essential for basic life since every human being must have insight into what food they have eaten. A nutritionist can help in guiding what the body should consume, where each patient may have different diet and treatment patterns. One indicator used by dietitians or nutritionists is by estimating the leftovers consumed by the patient. They measure it by visually named Comstock method, which is divided into scales. This method's drawback is subjective from one another dietitians or nutritionists so that an objective assessment with a machine learning-based approach is acquired. This paper proposes a novel stage of defining food recognition and measuring its leftovers using visual analysis. The food image recognition method used CNN to estimate food waste using pixel-based AFLE and regression approach to fit into six scales. The best result of food image recognition was 92.5% using dropout 0.3 with image augmentation and ReLu activation function, while the accuracy result of visual estimation application compared to experts was 85%. It is proved that the combined proposed algorithm is robust for the application of recognizing and estimating leftovers.