Research Of Vehicle Model Recognition Algorithm Based On Genetic Neural Network

As the rapid development of China's Highway & Transport, researchers pay much attention on Intelligent Transportation research.Vechicle classification occupies a very important position in Intelligent Transportation System,which is widely used in automatic toll collection system, traffic statistics, and other relating fields.Therefore, the development of inexpensive and suitable for China's national conditions vechicle classification system is of great practical significance.Based on the acoustic signal generated by moving vehicles, this article spread a research on vehicle recognition. The article first outlines the development of pattern recognition and sound detection technology, depicts the law of the acoustic propagation and the mechanism of automobile noise, and analyzes the feasibility of vehicle recognition based on the acoustic signals of vehicles. And then, a variety of acoustic signal samples about the types of the vehicles are collected by using sensor array. The signal samples are conducted a series pretreatments, such as wavelet denoising. As vehicle noise is a typical non-stationary signal, so this article puts the noise signal into different band decomposition by using the wavelet packet, reconstructs the wavelet packet coefficient, extracts of the different bands signals, and constructs eigenvector by using the band's energy as the vector elements. In this paper, a genetic BP neural network classification algorithm is proposed: first, through genetic algorithm, the paper conducts a training on artificial neural network weights; and then puts the trained weights value into the neural network to train and recognise the samples.The new BP neural network classification algorithm avoids the problem of traditional BP algorithm--such as local minimum, shorts the learning time and increases the capacity of network generalization.Using this feature extraction and classification algorithm, this article classifies the practical acoustic signals from the truck, car and bus. As different vehicles noise has different distributing of energy in different bands, in the simulation experiment, the signal samples are decomposed for three layers by using the wavelet packet, and extract different bands of energy from different type of vehicles to construct eigenvector. Then this article establishes genetic neural network model, trains and recognizes the signal samples using traditional BP neural network and genetic BP neural network, alternatively. The results show that, comparing with the method of traditional BP neural network, the convergence speed, the classification effect and the recognition rate are better for genetic BP neural network.From this paper's study, using the acoustic signal, generated from the moving vehicles, to recognize the type of vehicles is an feasible and effective method. It will have great prospect.