Research On Vehicle Type Recognition Method Based On Multi-Source LSSVM

Vehicle identification is an important part of intelligent transportation system.Due to the incompleteness and inaccuracy of the information obtained by a single sensor,it is difficult to correctly classify.Therefore,this paper uses infrared and visible light sensors to image the target,and uses canonical correlation analysis to extract feature vectors to obtain a consistent description of the target.Among many classification algorithms,the least square support vector machine is widely used because of its high recognition accuracy,fewer parameters and simple.This paper focuses on the application of least squares support vector machine in vehicle type identification,the main research contents and innovations are summarized as follows:1.Research on improved least squares support vector machine algorithm.This paper introduces a differential evolution algorithm to optimize the parameters after considering that the parameters of the least squares support vector machine cannot be determined by mathematical theory and affect the final classification result.On the basis of analyzing the theory of traditional differential evolution algorithm,it is improved for the problem that it is easy to fall into premature convergence.The results show that the improved algorithm has the ability of jumping out of the local best and overcome the shortcomings of traditional algorithm.Using the improved differential evolution algorithm to optimize the least squares support vector machine can improve the correct classification rate.2.Extract the feature vector of the image.The infrared image cannot reflect the target details for its contrast is not enough high.Median filtering and histogram equalization are selected as the image denoising and enhancement algorithms,the kernel principal component analysis is used to extract the feature vector of the infrared image.Compared with the traditional algorithm,the convolutional neural network can input the original image,avoiding complicated preprocessing steps,and the algorithm is simple.However,it requires a large amount of sample data training.This paper proposes a CNN-LSSVM algorithm to extract feature of visible light images,avoiding complex pre-processing and simulation experiments to prove that the improved algorithm still has a good classification rate in small samples.3.Apply the previous study to vehicle identification.This paper adopts canonical correlation analysis to realize feature level fusion after the analysis of various structural models and fusion processes of information fusion.The least squares support vector machine needs to be multi-classified in the actual multi-classification.This paper compares several commonly used multi-classification methods,and chooses the coding method for multi-classification.Based on the above research,a vehicle identification experiment based on multi-source least squares support vector machine is designed and the effectiveness of the proposed algorithm is verified.