基于最小二乘支持向量机的车型识别算法研究

Study on Vehicle Type Identification Algorithm Based on Least Squares Support Vector Machine

以感应线圈车辆检测器检测数据为分析基础,给出了基于Bayes理论的感应曲线自适应特征提取流程和方法,对选取的12个统计特征指标进行提取和优选。选择了曲线宽度、最大值、波峰数量、最小波谷值和波谷比组成车型识别模型的特征输入向量,不仅降低了输入向量的维数,缩短了最小二乘支持向量机的训练时间,同时也可加快车型识别的分类速度,增强特征值的分类辨别能力,提高车型分类的可靠性。在提出的基于最小二乘支持向量机的车型识别算法中,采用了修剪算法,加快了计算速度,同时保持了良好的回归性能。通过实例分析证明:基于最小二乘支持向量机的车型识别算法可提高自学习能力和识别准确率。

On the basis of data from inductive coil, the induction curve adaptive characteristic pick-up process and method based on Bayes theory were presented. After picking-up and optimizing 12 selected statistic characteristics according to the method, curve width, inaximum value, wave crest number, least trough value and trough ratio were chosen to constitute the characteristic input vectors of vehicle type identification model. This could not only reduce the dimensions of input vectors and cut the training time of least squares support vector machine （IS-SVM）, but also speed up the classification of vehicle type identification, improve eigenvalue＇ s identification capability and rise the reliability of vehicle type classification. The vehicle type identification algorithm based on IS-SVM, which adopted pruning algorithm, could speed up the calculation and keep well regression capability. Example analysis proves that the vehicle type identification algorithm based on IS-SVM would be able to enhance the self-learning capability and identify accuracy.