Research On Vehicle Logo Recognition Method Based On Deep Learning

With the rapid development of artificial intelligence system and the promotion of automation in all areas of the urban life,the convenience of traffic intelligence is also increasingly prominent.While people are paying attention to the ever-changing technological breakthroughs,they are also beginning to think about what practical problems these technologies can be used to solve.In order to solve the problems of vehicle management,traffic control,traffic accidents and other aspects,scientists have studied and realized the technologies for real-time detection and recognition of license plates.However,due to the influence of irresistible factors such as light and stains in complex traffic scenes,by the license plate recognition alone people still cannot guarantee that all vehicles can be accurately identified,and by using license plate recognition alone people also cannot cope with the situation of sets of vehicles.Therefore,it is necessary to study to identify vehicles from more dimensions.As the second ID card of a car,the vehicle logo not only contains manufacturer information and model information that the license plate does not have,but also the vehicle logo is difficult to be replaced.Therefore,how to efficiently recognize the vehicle logo efficiently has become a new research topic in the current intelligent transportation system.The main work of this paper is as follows:1)A two-stage vehicle logo positioning method based on segmentation experience search rectangles and Ada Boost classification is proposed.This method introduces the concept of "segmentation of empirical search rectangles" in the coarse positioning stage to divides the empirical search area of the vehicle logo and classifies it through Ada Boost.And through a experience search rectangle comparison experiment,it is verified that the proposed method has better compatibility and higher overall recognition accuracy for large vehicles and special vehicles.2)A vehicle logo recognition model based on similarity support vector machine with convolutional neural network is proposed.Thispaper analyzes the instability of the support vector machine classification method for directed acyclic graphs,and then uses the similarity between samples and the similarity between categories to design the optimal support vector machine classification network,and analyzes the effectiveness of the network.On this basis,this paper constructs a similarity-based support vector machine network model,and proposes a similarity support vector machine-based convolutional neural network vehicle logo recognition method.Through comparative experiments with some classic methods,it is shown that the proposed method has the advantages of shorter training time and higher overall classification accuracy than other vehicle logo recognition methods.