基于子带谱熵法和PSO-GA-SVM的汽车鸣笛识别

Car Whistle Recognition Based on Sub-Band Spectral Entropy Method and PSO-GA-SVM

针对鸣笛抓拍系统会产生误判的问题,提出了一种基于子带谱熵法和支持向量机的汽车鸣笛识别算法.首先,使用子带谱熵法对声音样本进行初判,将子带谱熵高于阈值的样本直接判定为非鸣笛样本.然后,对初判为鸣笛的样本中的疑似鸣笛部分进行分割,并提取Mel频率倒谱系数作为声音的特征.最后,使用支持向量机对分割结果进行进一步分类,并使用粒子群算法与遗传算法的融合来优化支持向量机的参数.仿真结果表明,该算法具有较好的鲁棒性.在对实际采集样本的鸣笛识别中,该算法也取得了较高的准确率.

In order to solve the problem of misjudgment in car whistle capture system,an algorithm based on sub-band spectral entropy and support vector machine is proposed in this paper.Firstly,sub-band spectral entropy method is used to preliminarily judge the sound samples.The samples whose sub-band spectral entropies are higher than a threshold value are directly determined as non-whistle samples.Then,the suspected whistle parts are segmented in the samples which are initially judged as whistle,and Mel frequency cepstrum coefficient is extracted as sound feature.Finally,the segmentation results are classified by support vector machine,and the parameters of support vector machine are optimized by the combination of particle swarm optimization and genetic algorithm.Simulation results show that the algorithm has good robustness.In the whistle recognition of actual collected samples,the algorithm also achieves a higher accuracy.