基于动载的路面不平度识别的小波特征提取

Wavelet feature extraction in road roughness recognition based on dynamic load

提出了基于车轮垂直动载的路面不平度识别的小波特征提取方案.对垂直动载采用bior 1.5小波,做4层小波分解,求4层细节系数和第4层近似系数幅值的均值和方差作为特征参数.使用Fisher判据和自组织映射(SOM)神经网络对小波参数、快速傅立叶变换(FFT)分段系数参数、全部数值的均值和方差、全变差、过零率、共振峰幅值参数以及实倒谱系数参数的分类能力进行了判定和比较.试验结果表明在路面不平度识别方面小波参数组合优于其他参数组合.

Wavelet feature extraction scheme is proposed on Vehicle Vertical Dynamic Load （VVDL）. VVDL is d for Road Roughness Recognition （RRR） based ecomposed to 4 layers with wavelet named bior 1.5. Means and variances of amplitude of wavelet coefficient （4 layers detailed coefficients and the 4th layer approximate coefficient） are taken as feature parameters. Fisher criterion and Self-Organizing Feature Map net are comparatively employed to determine the classification abilities of wavelet parameters, Fast Fourier Transform （FFT） subsection coefficient parameters, mean and variance of whole data, total variation, zero-cross ratio and formant amplitudes. Experimental results indicate that the wavelet parameter combination is superior to other combinations in RRR.