Research On Image Classification Based On Active Learning In Chinese Dishes

The algorithm commonly used in image classification is based on supervised learning.However,the main problem is that the classification performance of supervised learning algorithm is highly dependent on the number and quality of training samples.For the task of Chinese dish image classification,the data is large and the number of labeled samples available is limited.How to make full use of a small number of labeled samples and a large number of unlabeled samples to improve the classification performance of the model is of great significance.The advantage of the active learning method is that it can select unlabeled samples in the training set,mark the samples with large amount of information,add the newly marked samples to the original marked sample set,and train the classifier with the expanded marked sample set.The model is iterated until the set stop condition is met or all the samples in the training set are marked.Therefore,the active learning can improve the classification accuracy under the condition that the number of marks of the training samples is small.In this paper,the idea of active learning is introduced into the classification research of Chinese dish images.Based on the existing active learning classification models,through the research and analysis of active learning methods and the characteristics of Chinese dish images,aiming at the two problems in the classification of Chinese dish images,namely,the number of labeled training samples and the sampling cost,models suitable for the classification of Chinese dish images are proposed respectively,and the combined models are analyzed and verified for classification performance.The main contributions of this paper are as follows:(1)The active learning method is combined with the sampling strategy based on the committee,and the convolution neural network is used to extract image features to minimize the number of sample image labels in the learning process.Through the classification experiments of Chinese dish images by active learning and supervised learning,the results show that the algorithm can significantly reduce the number of sample marks in image classification,and at the same time can obtain higher classification accuracy,which can be applied to the classification of Chinese dish images with small number of marks.(2)A Chinese dish image classification algorithm based on the combination of active learning and unsupervised learning is proposed,and an unlabeled sample buffer pool is obtained by using a sampling strategy based on the entropy value ofactive learning bagging,and then the unlabeled sample with the most representative representation is selected from the candidate sample pool for labeling by using the unsupervised clustering algorithm.Through the Chinese dish image classification experiment,the results show that the algorithm can effectively select image samples with large amount of information,can quickly converge,and obtain relatively high classification accuracy.