Research On Chinese Food Dishes Image Classification Algorithm Based On Deep Learning

As people's health awareness increases,people pay more attention to their own health problems.There is an old Chinese saying: “Medicinal supplements are not as good as food supplements”.Therefore,diet is at the core of people's health management.The classification technology of dish images can be used as the technical support for intelli-gent health management systems such as nutritional balance,blood sugar,and calories,and it has a wide range of application scenarios.With the continuous improvement of deep learning methods,it is used in various scenarios,and of course,the classification of dishes images is no exception.At present,the technical research related to dish images is mainly aimed at Western food and Japanese cuisine,and there is little technical research on Chinese food dishes.The forms of Chinese food dishes are endless.Different types of dishes may be highly similar in shape and color,and the same types of dishes may be quite different.Therefore,the classification of Chinese food image is still a challenging topic.This paper studies the image classification technology of Chinese food dishes from two perspectives of multi-scale and attention,and proposes two new algorithms for food image classification.The main contents of this article are as follows:Firstly,in order to solve the limitation of the input size of the fully connected layer,a multi-scale sampling module is proposed,which is used to sample the image at multiple scales before the fully connected layer of the deep convolutional neural network.There-fore,images of any size are allowed to be input into the network model,while reducing the number of neurons in the fully connected layer,and improving the training speed of the network model on the basis of ensuring the accuracy of the network model.In order to solve the problem of imbalance in the number of samples of each category in the Chinese food image data set,the loss function trained by the network model uses a weighted cross entropy loss function,and in order to reduce the degree of overfitting,a weighted cross entropy loss function is added to The regularization mechanism verified the effectiveness of the multi-scale sampling module through multiple sets of comparative experiments.Secondly,there are many fine-grained features with small inter-class distance and large intra-class distance in the dish image.In order to better extract these fine-grained features,a bilinear network based on attention mechanism is proposed,and from the chan-nel and space Attention network construction in two directions.The channel attention network can learn the influence coefficient of each channel's features on the classification accuracy according to a large number of training samples,and then can locate the key channel features,so that the network model pays more attention to the classification re-sults of the key channel features during training And inhibited some of the information that interfered with the classification accuracy? inspired by the idea of bilinear networks,the channel attention network was used as a branch network of bilinear networks to achieve the extraction of fine-grained features and improve the accuracy of classification The purpose of the rate? subsequently,the mixed attention network composed of space and channel is added to the bilinear network to accurately express fine-grained features from both the space and the channel? at the same time,two branches in the bilinear network are con-sidered Different combination methods of the network will make the network express the characteristics differently,and select the best combination method by combining different networks.Finally,through multiple sets of comparative experiments,the effectiveness of the fusion of bilinear network and attention network is verified.Finally,the experimental results of the convolutional neural network model trained by the author are compared horizontally and the classification effect of the bilinear net-work based on the attention mechanism is the best,which proves the effectiveness of the classification algorithm.