基于多分辨率显著性滤波的微动信号增强方法

微动信号是典型的非平稳信号, 时频分析能够获得微动信号的联合时间-频率分布图像, 是微动信号分析的主要工具之一, 良好的时频图像质量能保证后续特征提取和参数估计的准确性。然而在实际场景中, 时频图像通常受到噪声污染, 使得微动信号难以分辨, 严重制约了后续特征提取和参数估计。根据显著性检测和图像金字塔的基本原理, 本文在多分辨率表示图像上分别计算显著性并滤波, 最后进行加权融合获得增强的时频图像, 有效抑制了噪声, 提升了低信噪比(signal to noise ratio, SNR)下时频图像的质量和微动信号的显著性。实验结果表明, 对于仿真信号以及暗室测量信号, 在-7~7 dB SNR下, 采用该方法均能显著提升时频图像质量, 且-3 dB以下时能大幅提高周期估计的准确率, 是一种有效的微动信号增强方法。

Micro-motion signal enhancement method based on multi-resolution saliency filtering

Micro-motion signal is a typical non-stationary signal. Time-frequency analysis can obtain the joint time-frequency distribution image of micro-motion signal, and it is one of the main tools for micro-motion signal analysis. Good time-frequency image quality can ensure the accuracy of the subsequent feature extraction and parameter estimation. However, in the actual scene, time-frequency images are usually polluted by noise, which makes it difficult to distinguish micro-motion signals, which seriously restricts the subsequent feature extraction and parameter estimation. According to the basic principles of saliency detection and image pyramid, we compute the saliency on the multi-resolution images separately, and then perform the filtering process. Finally, we perform weighted fusion of them to obtain the enhanced time-frequency image. This method effectively suppresses noise, improves the quality of time-frequency images and the saliency of micro-motion signals under low signal to noise ratio (SNR). Experimental results show that for simulated signals as well as darkroom measured signals, the proposed method can significantly improve the time-frequency image quality when the SNR is from -7 dB to 7 dB, and the accuracy of period estimation can be greatly improved when the SNR is below -3 dB. Therefore, it is an effective method for micro-motion signal enhancement.