Threshold de-noising method based on wavelet transformation is an effective approach to reduce white noise in digit signals. Considering different characters of signal and white noise in wavelet domain, a novel algorithm to determine wavelet threshold was proposed with multiple hypothesis test. Since a wavelet de-noising transformation process could be regarded as a multiple hypothesis test process and both step-up and step-down procedures could control false discovery rate (FDR) , the new method FDR step-up-down procedure formed by combining the above-mentioned procedures was used to determine the wavelet threshold. An attractive advantage of this method was that it could obtain the desired result by adjusting the FDR level flexibly. The simulated numerical results showed that this method works as effectively as the hearsure method and gives better signal to noise ratio (SNR) gains and MSE performance than both the traditional BH FDR and sqtwolog method. The selection of the significance level was also discussed, and it was pointed out that the relationship between significant level and improvement of SNR is nonlinear. Then, the tactic of selecting proper significant level was proposed.%小波阈值滤波是信号处理领域的重要方法,根据信号和白噪声在小波空间上传播的特性,提出了一种基于多假设检验确定小波滤波阈值的新算法.将小波阈值处理过程看作一个多重假设检验过程,FDR( False Discovery Rate)准则的step-up和step-down过程均能控制FDR在给定的显著性水平,综合这两个过程形成了FDR step-up-down过程并应用于确定小波滤波阈值.仿真实验表明,算法能够灵活调整显著性水平的大小来达到滤波后所希望的效果,以信噪比和均方误差作为衡量指标,该方法滤波效果与hearsure方法相当,优于BH FDR及sqtwolog方法.讨论了显著性水平的选取对滤波效果的影响,指出显著性水平的大小与信噪比的改善并非线性关系,提出了合理选取显著性水平的思路.
展开▼
