Convolutional Neural Network with Transfer Learning for Classification of Food Types in Tray Box Images

∗Food is one of the primary needs that cannot be separated from people’s lives. The various food types tend to increase every day, directly proportional to the human need for food increases as well. However, the lack of knowledge of the diversity of types of food is still common. It will cause nutrition and food problems for the community. Therefore, artificial intelligence with image processing techniques and computer vision that can recognize types of food can help people recognize various types of food. So, in this study, an image processing technique based on an artificial neural network, namely a Convolutional Neural Network (CNN) with transfer learning methods is used to classify the types of food in the tray box images. The use of the transfer learning method aims to overcome the problem of lack of training data for food images by utilizing the weights from the training model using the ImageNet dataset so that the model can be retrained using training data with minimum duration. The transfer learning models used in this research are ResNet-18 and GoogLeNet. The image data used are 304 food image data segmented from food images in 76 tray boxes divided into four classes: blank, fried rice, egg, and cucumber with 242 training data, 29 validation data, and 33 test data. The test is carried out using each model with the epoch having the best validation accuracy of 1. The 3rd epoch of the ResNet-18 model gets test results with an accuracy of 1, precision of 1, recall of 1, and F1-score of 1. The 12th epoch of the GoogLeNet model gets test results with an accuracy of 0.9845, precision of 0.9722, recall of 0.9808, and F1-score of 0.9752. Based on these results, it can be concluded that the CNN model with the transfer learning method can be used to classify the types of food in the tray box images.