Research on Automatic Dish Recognition Algorithm Based on Deep Learning
Hongmin Shao, Jiong Mu, Rong Tang, Xiao Chen, Mingxin Liu · 2020
Deep learning can automatically extract and learn multi-layer feature representations hidden between data, and has been successfully applied in image recognition and segmentation, semantic analysis and other fields. In order to improve restaurant settlement efficiency, save time and cost, this paper proposes a method that uses Convolution Neural Network and Huff transform to joint recognition, conducts preliminary detection of images taken with multiple dishes, uses Huff transform to extract and label individual dishes according to the shape of the plate. Then, the article constructs a convolutional neural network in the classification algorithm. Under different network parameters, this paper experiments on 10 dishes images in the cleaned VireoFood172 dataset. Through continuous optimization of network parameters, the classification accuracy of dishes Top2 has reached 87.2%. Under the fixed environment, the plate can be effectively extracted. It is proved that the method has a certain improvement in performance compared with the traditional dish recognition method, and can provide effective help for dish recognition.