Research On Vehicle Information Recognition Based On Deep Learning

With the continuous improvement of people's living standards,cars have become a common means of transportation for people to travel.With the continuous increase of the number of vehicles,traffic control and public safety issues have become increasingly prominent,and image detection and automatic identification of vehicles are important links.In practical applications,most of the image detection of vehicles adopts traditional methods such as optical flow method and interframe difference method.The detection results are not satisfactory.In the aspect of vehicle attribute recognition,the license plate recognition method is used to identify vehicles,but the current local fakes.Cards,occluded license plates,and deck ruts have brought a lot of resistance to the detection of cases involving vehicles as a crime tool,increasing the pressure on traffic and public security departments.Traditionally,relying on the identification of license plates to handle cases has been ineffective.Therefore,it is particularly important to identify other attributes of the vehicle.Based on the research results of predecessors,this paper proposes two vehicle detection methods,combined with deep learning technology to deal with vehicle detection and vehicle identification,and achieved good practical results.(1)For the target detection problem of vehicles,two improved vehicle detection methods are proposed.One is the SUSAN algorithm based on traditional edge corner detection.By improving the double threshold of the algorithm,the threshold t and the threshold g are improved to The adaptive acquisition method from the original image and the pseudo corner elimination method are used to detect the vehicle target.Experiments show that the improved SUSAN edge detection algorithm can effectively detect the vehicle targets in the road,and can effectively avoid the influence of the guardrails and vegetation on both sides of the road to extract the target edge features of the vehicle.It has high robustness.The second is the Faster-RCNN vehicle detection method based on convolutional neural network.The method is to add the RPN network based on the RCNN series detection algorithm,and share it with the convolutional layer of the convolutional network to realize the extraction of candidate frames and target features.In this way,the road vehicle can be detected quickly and accurately in the original image,and the target frame is marked.(2)For the problem of vehicle sign recognition,a vehicle sign recognition method based on deep migration learning is proposed for the problem of large demand and over-fitting of sample data in deep convolutional neural networks.The method performs feature migration in a homogeneous space through the deep network VGG-19 pre-trained on the ImageNet dataset;combined with the improved model loss function Softmax and the fully connected layer of the network,and freezes the middle and low layer convolutional layers,using different the learning rate is used to fine tune the high-level convolutional layer and the fully connected layer parameters.Under the small sample vehicle data set,the experimental results show that the method can have a high accurate recognition rate in training accuracy and test accuracy,and the accurate recognition rate of the test reaches 97.73%.The possibility of training the depth network of the small sample data set is solved.Sexuality also solves the problem of over-fitting of the model caused by insufficient sample size,and achieves the effect of accurately and quickly identifying the vehicle identification.