Food Type Detection Using the Convolutional Neural Network Method for Calorie Counter Applications

Nutrition in the diet significantly impacts human health and quality of life. Despite its importance, many individuals need to be made aware of the nutritional content of their food. This lack of awareness can contribute to a variety of health issues, including malnutrition, obesity, diabetes, heart disease, and cancer. Therefore, there is a need for an application that allows users to quickly, easily, and inexpensively determine food nutrition by taking photos. This study aims to develop an image-based food-type detection application using the Convolutional Neural Network (CNN) method with the ResNet-50 architecture, which will be compared with Inception v3. The study utilizes three datasets, combining 15 of Indonesia's most commonly found food types into a single dataset. Nutritional information for the food is sourced reliably from the FatSecret API, which provides verified nutritional data. Experimental results demonstrate that ResNet-50 consistently achieves higher accuracy than Inception v3, with ResNet-50 achieving 85.364% accuracy on validation data and 86.328% on testing data, establishing its suitability for a food-type detection system.