Traditional Bengali Food Classification Using Convolutional Neural Network

Abstract

Image classification is turning into a significant and promising perspective in the fields of object recognition using computer vision. However, researchers have barely scratched the superficials of food image classification till now. To evaluate the dietary aptitudes of people from various ethnicities, the classification of their traditional foods makes a huge impact. That’s what steered us into the classification of seven traditional foods in Bangladesh. In this regard, our key contribution to this aspect is the development of a dataset of Traditional Bengali Food Image (TBFI) including images of seven different classes of traditional Bengali foods: Biriyani, Panta Ilish, Khichuri, Fuchka, Roshogolla, Dim Vuna & Kala Vuna. For this, a scratch model incorporating Convolutional Neural Network (CNN) has been developed, rectifying to another vital contribution. As conventional Neural Network doesn’t perform well in case of image datasets, the CNN approach has been followed in view of its high accuracy, computational power with efficiency and automatic recognition of important features without any human oversight. Moreover, transfer learning approach with fine tuned VGG16 has also been used for TBFI classification. The proposed model in this paper has generated a culminated outcome upon our TBFI dataset with an average accuracy of 98% in classifying the traditional Bengali food images.