ABSTRACT
This chapter constructs a multiclass classification model that is fast as well as accurate when there are a large number of classes (food categories). It discusses a study that was aimed to apply convolutional neural network (CNN) with the use of data augmentation techniques to a set of five categories with a medium-scaled food image dataset. The chapter proposes that it would suffice to recognize a generalized version of a particular food item, based on which its dietary value can be approximately estimated, for instance, calories. Three different conventional algorithms, such as Support Vector Machine (SVM), K Nearest Neighbor (KNN), and random forest classification techniques, were used to train their respective models on a dataset of five categories, and their accuracy on the test images were recorded for the justification of using conventional neural networks.
