A Novel Dish Recognition Method Using Deep Learning

Understanding digital media dishes is a fascinating issue, but it also involves a lot of challenge. The dish's complicated ingredient list presents a hurdle. Due to the growth of deep learning, a number of efficient tools can partially resolve the issue. The job of dish recognition in this work is thought about. Based on the EfficientNet architecture and transfer learning, a unique dish recognition algorithm is proposed. First, add a number of significant layers to the EfficientNet-B0. Next, employ transfer learning to retrain the model using the best parameters that were learned during the first pre-training on ImageNet on the UEH-VDR dataset, a fresh batch of dish pictures. The UEH-VDR dataset includes pictures of Vietnamese food gathered from a variety of sources. According to experimental findings, the suggested approach can identify a dish with an accuracy of 92.33%. Additionally, it performs better than models built on well-known Space Invariant Artificial Neural Networks (SIANN) like VGG and residual neural network. On the basis of the training data, a mobile application is also created to assist tourists who wish to learn about Vietnamese cuisine.