Meat Dish Image Recognizer: Automatic Categorization of Various Meat Dish Images in Malaysia using Deep Learning Techniques

Malaysia is a gourmet treat, offering a diverse range of meals and delicacies. Here, we are describing the major types of Malaysia meat dish cuisine food. Beef is the most prevalent meat in Malaysian cuisine and it can be prepared in various methods. Depending on the dish, the beef dish can be cooked with different methods, and ingredients and comes with a different flavour. Automatic image recognition is a convenient option for labelling and categorizing dishes. In this paper, the five types of beef dishes classification approaches are presented with a dataset that contains approximately 1843 images captured with different positions, viewpoints, lighting, colours, scale, etc. The proposed model classifies the type of beef dishes from the input images containing daging kunyit, masak merah, rendang, sate and sup gearbox. For this project, Convolutional Neural Network (CNN) algorithms, a deep learning approach widely utilised in image recognition, were used. Additionally, transfer learning approaches in the categorization of meat dishes were examined. By using the parameters of a pretrained dish classification model as the new beginning parameters of the proposed neural network, transfer learning was achieved. The transfer learning model outperforms CNN in terms of accuracy compared with the same model when trained with random initial weights. The final classification accuracy of the transfer learning model by ResNet50 was 89.40% which is higher compared to the proposed CNN deep learning architecture at 54.35% accuracy. The accuracy level of 89.40%, demonstrates the feasibility of this approach in the classification of meat dishes