Research On License Plate And Vehicle Type Information Recognition Based On Deep Learning

With the rapid increase of the number of vehicles on the road,the cases of using fake license plate vehicle、unlicensed vehicle and other vehicle crimes also increased,so the research of vehicle information recognition is imminent in Intelligent Traffic System.License plate recognition and vehicle type recognition are the most important technologies in vehicle information recognition.Due to the influence of weather,complex background and other factors,license plate recognition and vehicle type recognition still have many difficulties.In this paper,the deep learning theories are studied deeply,the license plate recognition network and vehicle type recognition network model are designed,and experiments are carried out on the corresponding data sets.The experimental results show that the recognition accuracy of license plate and vehicle type can be improved by the recognition models that were proposed in this paper.The main work of this paper is as follows:In the aspect of license plate recognition: the license plate character recognition model that is based on the improved Le Net-5 network is designed and implemented.In the traditional LeNet-5 network model,part of the layers are improved to be the Inception-SE module,and the full connection layer is replaced by the global average pooling layer,and the BN and Dropout are used to optimize the network structure for license plate character recognition.Experiments are carried out on the license plate character data set that was obtained by image preprocessing and segmentation,Compared with the original network,the recognition accuracy and speed of license plate characters are improved by the improved LeNet-5 network significantly,and the recognition accuracy can reach 99.88%.Furthermore,the effects of SE module and Dropout are explored,and the recognition results of GUI interface intuitive feedback model are designed.In the aspect of vehicle type identification: a vehicle type identification model that is based on improved EfficientNet network is designed and implemented.The EfficientNet network model is simplified reasonably,the attention mechanism SE module is improved to be a better effect SK module,and the Adam optimization algorithm is used to improvethe network speed,experiment on the BIT-Vehicle data set.Compared with the original EfficientNet network and other vehicle recognition methods,the improved EfficientNet network in this paper has a higher recognition accuracy 96.9%,and the recognition speed is 1.87 times faster than the original network.