Article citationsMore>>

Marin, J., Escalante, H.J., Hernández, C.A., Gonzalez, J.A., Lopez-Lopez, A., Sucar, L.E., Guyon, I., et al. (2019) Recipe1M: A Dataset for Learning Cross-Modal Embed-dings for Cooking Recipes and Food Images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41, 1352-1362.

has been cited by the following article:

• TITLE: Predicting Categories and Ingredients of Traditional Dishes Using Deep Learning and Cross-Attention Mechanism

  AUTHORS: Ima Sokolo, Chidiebere Ugwu, Friday E. Onuodu

  KEYWORDS: Traditional Dishes, Cross-Attention Mechanism, Convolutional Neural Network

  JOURNAL NAME: Open Access Library Journal, Vol.12 No.3, March 19, 2025

  ABSTRACT: The food recognition systems are available for foreign dishes but much work has not been done for our traditional dishes making it difficult to classify the traditional dishes and the ingredients they are made up of. From extensive literature reviews conducted, the existing models on food recognition are not robust enough to handle classification and identification of ingredients in traditional dishes. This study developed an improved food recognition system for the classification and identification of ingredients in traditional dishes. The food image dataset used to build the model was gotten from Kaggle, which was not standardized. It was preprocessed and standardized for consistency across datasets in eighteen different classes for model building. The standardized dataset was split into two; 80% for training and 20% for testing, convolutional neural network and cross attention mechanism were used to build the model. The cross-attention mechanism was used to selectively pick features across the multiple classes in the food dataset. ReLU was used as activation function and Adam optimizer was used as optimization function in building the model. The object oriented analysis methodology was used in the design, while python programing language was used in the development of the system. The result obtained shows an accuracy of 93.57% for training and 90.0% for validation and error loss of 0.062% and 0.001% respectively and interestingly, during testing the model gave 99% accuracies to traditional food images inputted on it. The results from application were able to detect and classify traditional dishes into different classes and outline the ingredients used to prepare them which shows tremendously performance of the system.