Publication

Abstract : Food classification is a difficult challenge because there are many distinct categories, different foods look quite similar to one another, and there aren't enough datasets to train cutting-edge deep models. It will make improvements in computer vision models and datasets to test these models to solve this issue. This paper introduces Food10, a dataset of 10 Traditional Indian food categories with 5000 photos gathered from the web and concentrates on the second component of this study. We employ 4000 photos as a training set and 1000 images with human-validated labels for testing and validation. In the current study, we describe the steps involved in producing this dataset and offer pertinent baselines with deep learning models used in the Food10 dataset. Indian food is naturally oily and sweet hence contains lot of calories. Managing calorie intake is crucial for preventing obesity and mitigating the risk of numerous other diseases. The analysis of food images and calorie estimation can serve as a valuable tool to assist individuals in adhering to a healthy diet. Moreover, it can be beneficial for the general population in maintaining their everyday dietary choices. To calculate the calories food classification is the first step. In this research, a novel model was introduced with the aim of achieving enhanced accuracy and efficiency in the identification of Indian food, surpassing the performance of existing methodologies. The conventional models, such as AlexNet, VGG, and GoogleNet, were trained alongside the proposed model. On FOOD10 dataset, the proposed model Shallow Convolutional Neural Network (SCNN) gives a remarkable result with an average accuracy of 96%.