Research On Application Of Convolutional Neural Network In Vehicle Identification

With the increasing use and penetration of automobiles and related security issues,people have higher requirements for recognition speed and accuracy.In recent years,the theory of deep learning has gradually matured,bringing new ideas to this issue.Deep learning can also have obvious advantages in recognition speed,flexibility,and accuracy because of its adaptive learning.Convolutional neural networks,as the most studied and in-depth category in deep learning,have also been widely used.The areas of image retrieval,speech recognition,face recognition,and translation have also achieved extraordinary results.Different from the full connection of the traditional neural network,the full and partial connections in the convolutional neural network are used interchangeably.This kind of network structure has high adaptability to changes in the operation of the image's translation,scaling,and tilting.Suitable for image data processing learning.The continuous development of deep learning has brought a lot of demand for the development framework,so a large number of deep learning frameworks including Caffe used in this article have emerged.The Caffe framework greatly facilitates everyone's research on deep learning with its advantages of not needing code to design networks,speeding up the training process,and having a large number of well-trained classical models,and is widely used in cutting-edge industry and academia.application.The VGG model is superior to other models in the performance of multiple migration learning tasks,and it is the preferred algorithm for extracting CNN features from images.Therefore,it is the preferred model for research.The main research in this paper is that in the static image,the vehicle model is identified by the convolutional neural network in the deep learning,and it is proved through experiment that it can be more suitable for the recognition research of the vehicle than other models and can achieve the speed There is a clear advantage in accuracy.Based on this point,the following work was mainly carried out.Firstly,the significance of vehicle identification and the development status at home and abroad were briefly introduced.Then the relevant theoretical knowledge required for experiments was introduced,including machine learning and deep learning.The related structures and functions of the convolutional neural network mainly studied in this paper are introduced,and the most suitable ReLU function is selected by comparison.Later,through the horizontal comparison of mainstream deep learning frameworks and models,the reasons of the Caffe framework and the VGG model used in this paper are analyzed.Finally,using Caffe on VGG to achieve their own data training,and adjust the sample data to achieve better experimental results.Compared with the traditional VGG model,the design of this paper is less network level,there is no large amount of computation,but it can also achieve a good recognition effect,the experiment shows that the accuracy rate can reach more than 95%.