¿µ¹®³»¿ë (English Abstract) |
Food recognition is essential in nutritional assessment for eating more healthily, avoiding some diseases, and managing weight loss. However, due to the nature of the food, it is a challenging task to recognize some types of food. Deep learning has been widely used for food recognition, as it outperformed other approaches. In this paper, we propose an enhanced densely connected convolutional network (DenseNet) approach for food image recognition. For performance tests, we compared our model with Resnet152 using UEC-FOOD256 dataset. The test results show that our model outperforms other methods with an accuracy of 92.23% on the training set and 65.30% on the validation set. |