-
2维小波变换用同一组滤波器分别对行和列进行1维小波变换。首先对2维矩阵的每一行提升分解得到低频信息和高频信息,再对列分解,则2维矩阵经过一层小波变换分解为近似系数,水平细节系数,垂直细节系数和对角细节系数。多层2维小波分解需要对近似系数重复分解过程直到分解到指定层数。
小波变换的运算量由低通滤波器和高通滤波器的长度决定。令低通滤波器长度为h,高通滤波器长度为g,基于MALLAT算法的传统小波分解一层所需的运算量为C=2(h+g)+2。设待分解原始信号的长度为L,进行1维k层小波分解所需的运算量为LC(1-2-k)。进行m列n行(L=m×n)2维小波分解时,每列进行列变换所需计算量为nC(1-2-k),对于m列则为mnC(1-2-k)。同理,行变换的计算量为mnC(1-2-k),总计算量为2mnC(1-2-k)。2维小波分解的计算量是1维小波分解的2倍。为了减少计算量,采用提升格式的2维小波变换。
-
传统的小波变换以傅里叶变换为基础,通过对函数的伸缩平移实现多分辨率分析,被称为第1代小波变换。提升小波变换是对传统小波变换的改进,摆脱了傅里叶变换,避免了传统小波中基于卷积算法的冗余计算,可实现原位计算,对内存需求量小。正反变换的架构完全相同,反变换只是正变换的逆向操作,算法简单、速度快,适合并行处理,易于硬件实现,在信号处理领域有广阔的应用前景[20]。
提升算法将小波变换过程分为分解、预测和更新。
(1) 分解。将输入的离散信号x[n]按照奇偶性分解xe[n]和xo[n]两个子集,即:
$ \left\{ \begin{matrix} {{x}_{\text{e}}}[n]=x[2n]\ \ \ \ \ \\ {{x}_{\text{o}}}[\mathit{n}]=x[2n+1]~ \\ \end{matrix} \right. $
(1) (2) 预测。利用信号x[n]相邻采样点间具有相关性,可以通过偶数采样点预测奇数采样点,即xo′[n]=F(xe[n]),F为预测算子,预测误差作为小波系数:
$ d[\mathit{n}]={{x}_{\text{o}}}[\mathit{n}]-{{x}_{\text{o}}}^{\prime }[\mathit{n}] $
(2) (3) 更新。构造更新算子U与预测的小波系数d[n]作用,叠加到原偶数序列xe[n],得到原信号的近似,即尺度系数:
$ c[\mathit{n}]={{x}_{\text{e}}}[\mathit{n}]~+\mathit{U}\left( d[\mathit{n}] \right) $
(3) 提升方案的重构过程与分解互逆。本文中选用系数简单的bior2.2小波函数,低通滤波器长度为5,高通滤波器长度为3。
基于提升格式的bior2.2小波对信号的分解重构表达式为:
$ \left\{ \begin{matrix} d\left( n \right)=s\left( 2n+1 \right)-\frac{1}{2}\left[ s\left( 2n \right)+s\left( 2n+2 \right) \right]\text{ } \\ a\left( n \right)=s\left( 2n \right)+\frac{1}{4}\left[ d\left( n-1 \right)+d\left( n \right) \right]~\ \ \ \ \ \ \ \ \ \\ \end{matrix} \right. $
(4) $ \left\{ \begin{matrix} s\left( 2n \right)=a\left( n \right)-\frac{d\left( n-1 \right)+d\left( n \right)}{4}\ \ \ \ \ \ \ \ \ \\ s\left( 2n+1 \right)=\frac{a\left( n \right)+a\left( n+1 \right)}{2}-\ \ \ \ \ \ \ \ \ \ \\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \frac{d\left( n-1 \right)-6d\left( n \right)+d(n+1)}{8} \\ \end{matrix} \right. $
(5) 式中,d(n)是高频细节系数,a(n)是低频近似系数; s(2n)和s(2n+1)是重构信号的偶数部分和奇数部分。
-
瑞利BOTDA系统检测光纤布里渊频移时,通过扫描微波频率,得到一系列不同扫频信号下的1维信号,见下式:
$ \left\{ \begin{matrix} {{S}_{1}}=f({{\nu }_{1}}, {{z}_{i}}) \\ {{S}_{2}}=f({{\nu }_{2}}, {{z}_{i}}) \\ \vdots \\ {{S}_{m}}=f({{\nu }_{m}}, {{z}_{i}}) \\ \end{matrix} \right., \left( i=1, 2, \ldots , n \right)~ $
(6) 式中,ν1, ν2, …, νm为扫描频率; z1, z2, …, zn为光纤位置; S1, S2, …, Sm为扫描频率ν1, ν2, …, νm下得到的m组1维信号。将m组1维信号按列存储,得到m列n行的2维矩阵M,见下式:
$ \mathit{\boldsymbol{M=}}\left[ \begin{matrix} f\left( {{\nu }_{1}}, {{z}_{0}} \right)&f\left( {{\nu }_{2}}, {{z}_{0}} \right)&\cdots &f\left( {{\nu }_{m}}, {{z}_{0}} \right) \\ f\left( {{\nu }_{1}}, {{z}_{1}} \right)&\text{ }f\left( {{\nu }_{2}}, {{z}_{1}} \right)&\cdots &f\left( {{\nu }_{m}}, {{z}_{1}} \right) \\ {}&\vdots &{}&{} \\ \text{ }f\left( {{\nu }_{1}}, {{z}_{n}} \right)&f\left( {{\nu }_{2}}, {{z}_{n}} \right)&\cdots &f({{\nu }_{m}}, {{z}_{n}}) \\ \end{matrix} \right] $
(7) 设扫描频率步进为Δf,脉冲宽度和采样频率共同确定的光纤位置间隔为Δz,矩阵的行坐标i表示第i个光纤位置,列坐标j表示第j个扫描频率,则:
$ \left\{ {\begin{array}{*{20}{c}} {i = \frac{z}{{\Delta z}}\;\;\;\;\;\;}\\ {j = \frac{{{v_j} - {v_1}}}{{\Delta f}}} \end{array}} \right. $
(8) 在坐标为(i, j)位置处的元素值为在扫描频率νj下光纤z处的布里渊时域信号功率值P,即M((νj-ν1)/Δf, z/Δz)=P。2维矩阵的列向量代表布里渊功率值随光纤长度的分布,即实际测量信号,包含空间信息,行向量代表布里渊功率值随扫描频率的分布,即布里渊增益谱信号,包含时间信息。布里渊功率值与光纤长度和扫描频率间均存在非线性关系,即2维矩阵的行向量和列向量间均存在相关性。因此采用2维小波变换,可以同时去除行列向量的相关性,达到良好的降噪效果。
在10.81GHz~10.92GHz范围内扫描微波频率,获得不同扫频下的测量曲线如图 2a所示;结合布里渊功率与光纤长度、扫描频率的关系得到3维布里渊增益谱如图 2b所示; 3维图的俯视图即为结合了测量信号空间信息和时间信息的2维矩阵,如图 2c所示。
-
对上述预处理后的2维信号降噪步骤如下:(1)采用提升格式的双正交小波bior2.2对2维图像信号进行小波分解,得到近似系数和细节系数; (2)保留近似系数,对细节系数软阈值量化,得到新的阈值系数;(3)对阈值处理后的细节系数和最高分解层的近似系数利用提升算法重构,得到降噪后的信号。
-
采用2维提升小波阈值降噪后结果如图 3所示。
Figure 3. a—denoised signals at different scan frequencies b—3-D graph of denoised signals c—top view of 3-D figure
由图 3可以看出,降噪后测量曲线波动减小,图像更加光滑。
通过信噪比(signal-to-noise ratio, SNR)和算法运算量指标将传统1维小波和2维提升小波降噪进行对比。分别在分解层数为2~8时计算传统1维小波和2维提升小波降噪信号的信噪比,如图 4所示。
由图 4可知,在分解层数为2~7时,随着分解层数增加,信噪比提高,当分解层数为8时,信噪比下降。这是因为分解层数过多时,低频域有用信号的特性被当作高频噪声滤除,实际信号丢失严重。2维提升小波降噪和1维传统小波降噪相比,信噪比提高了10dB左右,降噪效果明显优于1维传统小波。
bior2.2小波低通滤波器长度为5,高通滤波器长度为3。对1维信号基于MALLAT算法一层分解的运算量为18, 基于提升格式分解运算量为6(见(4)式),2维信号基于提升格式分解运算量为12。因此2维提升分解比传统1维小波分解运算量减少了1/3。
瑞利BOTDA系统的2维提升小波降噪方法
2-D lifting wavelet de-noising method for Rayleigh BOTDA system
-
摘要: 为了解决基于瑞利散射的布里渊光时域分析系统(BOTDA)中传感信号受噪声干扰严重的问题,采用2维提升小波变换算法,将测量信号从1维空间转换到2维空间,进行阈值降噪处理。通过理论分析和实验验证,取得了传统小波与2维提升小波降噪数据。结果表明,2维提升小波变换比传统小波变换信噪比提高约10dB,运算量减少了1/3;2维提升小波充分利用测量信号时间上的相关性,变换结构简单、运算速度快、降噪效果优于传统小波,适用于瑞利BOTDA系统降噪。该结果对光纤传感系统中信号降噪的研究有一定参考价值。
-
关键词:
- 光纤光学 /
- 信噪比 /
- 2维提升小波 /
- 瑞利-布里渊光时域分析系统
Abstract: In order to solve the problem that the sensor signal is seriously disturbed by noise in a Brillouin optical time domain analysis (BOTDA) system based on Rayleigh scattering, the 2-D lifting wavelet transform algorithm was used to convert the measured signal from 1-D space to 2-D space, and the noise was reduced by threshold. Through the theoretical analysis and experimental verification, the traditional wavelet and 2-D lifting wavelet denoised data were obtained. The results show that the signal-to-noise ratio of the 2-D lifting wavelet transform is about 10dB higher than that of the traditional wavelet transform, and the computation amount is reduced by 1/3. The 2-D lifting wavelet makes full use of the time correlation of the measured signal, the transformation structure is simple, the operation speed is quick and the noise reduction effect is superior to the traditional wavelet. It is suitable for noise reduction in a Rayleigh BOTDA system. The results of this paper are of great reference to the research of signal denoising in optical fiber sensing systems. -
-
[1] CUI Q S, PAMUKCU S, LIN A, et al. Distributed temperature sensing system based on rayleigh scattering BOTDA[J]. Sensors Journal IEEE, 2011, 11(2):399-403. doi: 10.1109/JSEN.2010.2066558 [2] ZHANG C, YU W F, XIA M, et al. Analysis of optical signal characteristics of optical fiber stimulated Brillouin scattering[J]. Laser Technology, 2016, 40(3):363-366(in Chinese). [3] WANG Y Q, HE J, MA Y, et al.Effect of random noise on measurement of marine brillouin laser radar[J]. Laser Technology, 2015, 39(1):6-12(in Chinese). [4] DAKIN J P, PRATT D J, BIBBY G W, et al. Distributed optical fibre Raman temperature sensor using a semiconductor light source and detector[J]. Electronics Letters, 1985, 21(13):569-570. doi: 10.1049/el:19850402 [5] FARAHANI M A, WYLIE M T V, CASTILLO-GUERRA E, et al. Reduction in the number of averages required in BOTDA sensors using wavelet denoising techniques[J]. Journal of Lightwave Technology, 2012, 30(8):1134-1142. doi: 10.1109/JLT.2011.2168599 [6] KE T B, LIN L, LI Y Q, et al.Study on denoising method of time-domain reflected signal in brillouin optical cable[J]. Laser Technology, 2014, 38(3):311-315(in Chinese). [7] WANG W, XIANG Z, WANG Z D.Denoising for optic gyroscope based on wavelet full frequeney-domain threshold processing[J]. Infrared and Laser Engineering, 2012, 41(12):3401-3405(in Chinese). [8] CHEN G, ZHU X F, XU Q Q, et al.Application of wavelet and sparse decomposition in discontinuous thin film denoising[J]. Laser Technology, 2014, 38(4):546-550(in Chinese). [9] QUAN Z, ZHANG B H.Application of an improved wavelet threshold algorithm in laser interrogation[J]. Laser Technology, 2014, 38(2):218-224(in Chinese). [10] CHEN Y, CHENG Y, LIU H. Application of improved wavelet a-daptive threshold de-noising algorithm in FBG demodulation[J]. Optik-International Journal for Light and Electron Optics, 2017, 132(3):243-248. [11] ZHAO L, LIU H, XU SH C, et al.Comparative study on the analysis of fiber vibration signals by HHT and CWT[J]. Laser Technology, 2017, 41(2):260-264(in Chinese). [12] LIU L, YU M, YANG R J, et al. Wavelet de-noising for fiber Raman temperature sensing system[J]. Chinese Journal of Lasers, 2013, 40(6):0605005. doi: 10.3788/CJL [13] CHEN S Y, LIU J X, DING Y.Study on fusion method of infrared and X-ray image based on wavelet transform[J]. Laser Technology, 2015, 39(5):685-688(in Chinese). [14] ZHAO R, GU G H, YANG W.Implementation of visible light image based on polarization imaging[J]. Laser Technology, 2016, 40(2):227-231(in Chinese). [15] ZHAO L J, LI Y Q.Research on temperature feature extraction of BOTDR system based on wavelet transform[J]. Electrical Measurement and Instrumentation, 2015, 52(1):19-23(in Chinese). [16] CAI Z H, CHEN W J, ZHONG M.Application of several wavelets in 3-Dimensional surface survey[J]. Laser Technology, 2015, 39(5):610-616(in Chinese). [17] FARAHANI M A, CASTILLO-GUERRA E, COLPITTS B G. Acce-leration of measurements in BOTDA sensors using adaptive linear prediction[J]. IEEE Sensors Journal, 2013, 13(1):263-272. doi: 10.1109/JSEN.2012.2213153 [18] ZHAO SH Q, PANG F F, HE M T, et al.Study on signal processing of optical fiber coherent optical time domain reflectometer based on gray image[J]. Chinese Journal of Lasers, 2015, 42(3):0305001(in Chinese). doi: 10.3788/CJL [19] SOTO M A, RAMI'REZ J A, THÉVENAZ L.Intensifying Brillouin distributed fibre sensors using image processing[J].Proceedings of the SPIE, 2015, 9634:96342D. [20] LIU A L. Dual filtering algorithm for infrared image based on lifting wavelet transform[J]. Laser Technology, 2015, 39(4):545-548(in Chinese).