Combination Of Wavelet Transform And Mutual Information For Vehicle Recognition

Vehicle classification is an important part of the intelligent transportation system,which has important significance in traffic statistics,express highway toll collection,improving road utilization,road system and transportation planning.In order to achieve vehicle type classification automaticly,this paper researchs vehicle identification and classification by using its sound characteristics.According to the current international advanced vehicle classification criteria,three different types of vehicle has been chosen,vehicle acoustic signal acquisition experiments are carried out in different road environment.Because of the collected sound signals are accompanied by a large number of random background noise,the adaptive noise reduction method based on wavelet per-levelthresholding is implemented on the collected vehicle acoustic signals.The main frequency band associated with the vehicle classification is cleared by using spectrum analysis for vehicle acoustic signals.To avoid losing minor but also important acoustic characteristics,which is minor in energy relatively but contribute most tothe discriminatory information and is also important for vehicle type classification,a fusion approach to combine wavelet transform method which based on signal energy conservation and mutual information method which doesn’t based on signal energy conservation is applied in feature extraction process.Taking into account the high dimensionality of feature vectors will reduce the performance of the classifier,genetic algorithm is employed to reduce the dimensionality of the fusion feature vector that combined wavelet coefficients energy feature and key frequency component feature,not only removes the redundancy of the fusion feature vector,but also improves the accuracy of the classifier.After feature extraction,this paper selects BP neural network,LVQ neural network and support vector machine for pattern recognition.In order to take advantage of the complementary and discriminatory information between wavelet coefficients energy feature and key frequency component feature,D-S evidence theory is applicated for decision level information fusion,which combined the classification outcome of BP neural network with wavelet transform and support vector machine with mutual information.The relevant experimental results indicate that: the feature extraction method that combines wavelet transform and mutual information will use the complex information of vehicle acoustic signals,in response to the conflict situation of classification results obtained from two different classifiers when input the same vehicle sound signal sample,D-S evidence theory increases the confidence to the unknown correct sample category,eliminating the uncertainty of classification results and improving the accuracy of vehicle type classification.