天体光谱信号去噪的小波域复合阈值新算法

A Novel Wavelet Multiple Thresholding Algorithm for Astronomical Spectral Signal Denoising

利用谱线和噪声在小波域内的不同相关特性,提出了一种小波域复合阈值去噪算法。首先将小波系数作NeighShrink阈值处理,然后对得到的小波系数进行二值化,在此基础上定义了每一小波系数所对应的横向相关性指数和纵向相关性指数,最后确定出决定小波系数取舍的决策系数。由于该决策系数是通过多重判据得到的,因此该方法克服了简单阈值方法过保留或过扼杀的缺点,同时可以去除大脉冲噪声,实验结果表明了该方法的有效性。

By using the different relativities of spectral lines and noises in the wavelet domain, a novel wavelet multiple threshol- ding algorithm is presented in the present paper for astronomical spectra with low signal-noise-ratio （SNR）. Firstly the wavelet coefficients are estimated by NeighShrink approach and then 0-1 coefficients are obtained. Based on the above binary coefficients, two kinds of relativity exponents at each scale and across scales respectively are defined for each wavelet coefficients. Finally decision coefficients are determined according to the magnitudes of the relativity exponents. This algorithm overcomes the overreserving or over-shrinking disadvantages of the simple threshold method in that the decision coefficients are obtained by multiple criteria. Besides, large pulse noises can be removed by the presented algorithm because it takes spectral line features into consideration. The experimental results show that the proposed algorithm is computationally efficient and practical.