基于人耳听觉特性的信号显著图计算方法研究

Calculation method for signal saliency map based on human auditory characteristics

信号的幅值突变往往蕴含着设备的故障信息,为明确产生幅值突变的时频区域,基于听觉系统的信息处理机制,提出了一种信号显著图计算方法。首先,对信号进行带通滤波、相位调整、半波整流等听觉外周处理,然后,提取处理结果的一次或多次包络信息,并对包络信息进行多尺度二维滤波。继而,利用中心-周边差算子得到信号在不同尺度下的局部显著度,最后,对局部显著度进行跨尺度整合和线性合并得到信号的时频显著图和全局显著图。仿真和实验结果表明,该方法适用于振动信号和语音信号,可以清晰表征幅值突变的时频区域,并可抵抗一定的噪声干扰,具有一定的有效性和实用性。

A signal amplitude saltation often implies fault information of an equipment. In order to recognize the time-frequency domain where a signal amplitude saltation appears, a calculation method of signal saliency map based on the information processing mechanism of an auditory system was proposed. Firstly, auditory peripheral processings, such as, band-pass filtering, phase adjustment and half wave rectification were performed for a signal. Secondly, extracting the envelope information from the results of processings one time or multiple times, they were then processed with the multi-scaled two dimensional filtering. Nextly, the local saliency of the signal under different scales was obtained using the centre-rim difference operator. Finally, the global saliency map and time-frequency saliency map of the signal were gained by means of the multi-scaled integration and the linear combination of local saliencies. The results of tests and simulations showed that the proposed method is applicable to both vibration signals and voice signals, it is effective and practical, it can clearly represent amplitude saltation time-frequency domain of a signal and resist a certain noise interference.