Research On Classification And Recognition Methods Of Vehicles Based On Vehicle Acoustic Signals Using SVM And CNN

With the rapid development of intelligent transportation,the classification and recognition technology of vehicles as an important branch of the field has received a lot of attention.The vehicle classification and recognition method based on traditional machine learning is mainly based on the shallow model.The disadvantages are including the following two points.firstly,the number of layer is relatively small;modeling and expression capabilities are limited;secondly,the classification and recognition results depend on the shallow features extracted by traditional methods.The expression ability of shallow features is often limited,resulting in the classification result is not ideal.Therefore,it is still a difficult problem to use the aliased,intermittent,and multi-source noise signals generated during the motion of the vehicle and extract the traditional features of sound to analyze and study vehicle classification and recognition.Thesis focuses on the above research status and difficulties.Based on the vehicle acoustic signal processing and deep learning technology,analysis and research the vehicle classification and recognition methods based on convolutional neural network.The main research content is as follows:1)Acquire vehicle acoustic signals,and remove redundant information and condense experiment required feature set.The sound signals have the characteristics of wide frequency range,high sampling rate,and a large amount of quantized signal data.It can be seen that it is not suitable for direct put in classifiers and neural networks.Therefore,before classification and recognition of vehicle,signals need to be pre-processed and extracted traditional feature,and the signal is marked with corresponding labels.Through the experimental requirements,the experimental database needed for the research is finally established,and data preparation is made for the follow-up research work.2)Research on vehicle classification and recognition method based on convolutional neural network.Analysis by comparison the effectiveness of vehicle classification and recognition based on machine learning and deep learning models,combined with the high complexity of vehicle sound signals makes the signal characterization problem very well able to use the highly abstraction provided by deep learning,and convolutional neural network unique convolution and pooling operation can effectively express and process some typical features of audio signal hidden in frequency domain.This paper proposes a vehicle classification and recognition method based on convolutional neural network.Firstly,the classical Convolutional Neural Network Le Net-5 is used to classify the vehicles.The results show that Le Net-5 is not ideal for the classification performance of this data set of this paper,the value of the loss function does not drop,the model does not converge,and there is no accuracy.By analyzing the reasons,improved the Le Net-5 network to obtain three convolutional neural networks CNN1,CNN2,and CNN3.3)The effectiveness of the three convolutional neural network vehicle type classification recognition models was verified.The experimental results show that the convolutional neural network has excellent classification performance for vehicle type;and Compared with the classification performance of shallow model K-nearest neighbor and SVM.The results show that the accuracy of CNN1,CNN2 and CNN3 is greatly improved compared with shallow model K-nearest neighbors and support vector machines.Verify by comparison the classification performance of model on data set 1 and data set 2,the results show that the data volume increases,the accuracy of convolutional neural network has been improved,and the shallow model has decreased.It is proved that convolutional neural networks have certain advantages when dealing with large data sets,compared with shallow models that are good at handling small samples.