Deep Learning-Based Approach for Calorie Estimation in Indian Foods

People in the modern era are increasingly concerned with their diet and food choices to prevent developing chronic diseases like high blood pressure and diabetes. With the increasing usage of smartphones and technology, an application to automatically monitor their diet would be a great boon, especially for the younger generation. This research work proposes automatic food calorie estimation from the food images using deep learning technology. The proposed work has three main phases. In the first phase, the food image is categorised as one of the types, for example, boiled, fried in oil or junk. A novel deep neural network namely MobileNetV2 which is a 15-layer model architecture is used for this purpose. In the second phase, the quantity of the labelled food is estimated. The third and final stage involves computing the calories present in the food using the Google search API. The details of ingredients present in each type of food are used for this purpose. Classification accuracy, food amount calculation precision, and calorie estimation accuracy are used assessing the effectiveness of the proposed approach. The proposed model is producing results with an overall accuracy of 95.21 percent which is better than the existing architecture.