Portuguese Meals Image Recognition Using CNN Models

Recent research in deep learning has played a crucial role in many areas especially in image classification. Processing and analyzing large volumes of visual data has proven to be a huge success for deep learning techniques, which imitate the neural networks found in the human brain. This paper has an objective on identifying Portuguese cuisine using deep learning techniques, by creating and comparing three different models such as Multi-Layer Perceptron (MLP) as an Artificial Neural Network (ANN), Long Short-Term Memory (LSTM) as a Recurrent Neural Network (RNN), and ResNet-50, a pre-trained Convolutional Neural Network (CNN). Each model is trained using a carefully chosen dataset of Portuguese meals that represents the many types of culinary. We evaluate the models in detail using accuracy and loss performance criteria. Our experimental results indicate that the ResNet-50 model outperforms the others with a 90% test accuracy and a 97% training accuracy. Still, more investigation is needed.