Convolution Neural Network Vehicle Classification Algorithm Based On Label Correlation And Collaborative Training

In recent years,the convolutional neural network has become a research hotspot in the field of computer vision with its breakthrough achievements.At the same time as rapid development,the industry has also put forward higher requirements for tasks such as image processing.For example,the problem of poor label recognition accuracy in multi-label recognition of things can be fully described.For example,a large number of wasteful manpower labeling problems for data samples in the training process of convolutional neural networks.These problems are ubiquitous in real-life scenarios and restrict the application of computer vision technology to a large extent.According to the actual source of the project,after analyzing and analyzing the shortcomings of common convolutional neural network algorithms,this paper proposes a set of improved algorithms for training problems under multi-label classification and small sample data sets encountered in practical applications.In order to make the convolutional neural network can effectively use the information between tags in multi-tag classification,a convolutional neural network model based on Bayesian theory is proposed.According to the statistical information of the tag to obtain prior knowledge,the prior knowledge of the tag is combined with the prior knowledge of the tag to assist the multi-tag classification of the convolutional neural network,so as to improve the overall recognition accuracy of the label.This paper also summarizes the research results of the collaborative training algorithms of semi-supervised learning by domestic and foreign researchers.Aiming at the problem that the initial training data set is insufficient to make the model difficult to train,a semi-supervised learning algorithm for convolutional neural network collaborative training is proposed.Combined with the idea of ensemble learning,multiple base classifiers are integrated to predict and classify a large number of unlabeled samples,and the information generated by the unlabeled samples is used in the training process of the convolutional neural network,thereby indirectly enabling the number of training samples to be amplified.Avoid overfitting the model because of too few training samples.The experimental results show that the classification of tag-related information can effectively improve the accuracy of highly-recognized tag recognition,and increase the difficulty rate of the tag from about 30% to 53%,thus improving the overall recognition accuracy of the model.Cooperatively trained convolutional neural networks can effectively alleviate the overfitting caused by too few samples.Through analysis,we can see that the proposed algorithm has obvious effect on performance improvement.