Distributed sensing method for image denoising based on nonlocal Haar transform and optical frequency domain reflectance.

Abstract

This invention relates to a distributed optical frequency domain reflectance sensing method for image denoising based on nonlocal Haar transform, comprising the following steps: Under the triggering of a tunable laser, Rayleigh scattering signals at different temperatures are continuously acquired using an OFDR system during the heating process of a high-temperature furnace; adjacent sets of signals are processed separately to obtain a local reference Rayleigh scattering spectrum and a locally measured Rayleigh scattering spectrum; DC component is removed; cross-correlation calculations are performed separately, and the results are normalized, with each set of normalized cross-correlation results corresponding to a position on the optical fiber, resulting in an original two-dimensional cross-correlation intensity map y; the rows with the smallest Euclidean distance from the reference row are formed into a matrix; a separable two-dimensional Haar transform is performed to obtain the transform coefficients; double hard thresholding is applied; a preliminary denoised map is obtained; Wiener filtering is performed; and the image corresponding to temperature changes T0 to T1 is obtained. n The spectral frequency shift distribution.

Description

Technical Field

[0001] This invention relates to the field of fiber optic distributed sensing technology, and in particular to a distributed sensing method for optical frequency domain reflection using nonlocal Haar transform image denoising. Background Technology

[0002] Optical fibers, due to their resistance to electromagnetic interference, corrosion, and extreme temperatures, are used for sensing and measurement in harsh environments. In fields such as the steel industry, solid oxide fuel cells, and nuclear power plants, distributed optical fiber sensors can be deployed on test structures through embedding, insertion, or direct contact to achieve accurate and rapid distributed high-temperature measurements. Optical fiber distributed high-temperature sensing shows a very broad application prospect. Among existing optical fiber temperature sensing technologies, distributed optical fiber temperature sensing based on Raman scattering and Brillouin scattering typically employs optical time-domain reflectometry (OTDR), making it unsuitable for applications requiring high spatial resolution. Fiber Bragg gratings (FBGs) are also used for temperature sensing and can achieve centimeter-level spatial resolution; however, their high manufacturing cost and the grating erasure phenomenon above 700°C limit their application in higher-temperature scenarios. Optical frequency domain reflectometry (OFDR) based on Rayleigh scattering is another optical fiber distributed high-temperature sensing scheme. [1] Others reported that measurements at 800℃ were achieved using ZrO2-doped optical fibers, with a test distance of 40 cm and a spatial resolution of 1 cm; AKSang [2] Other researchers used gold-plated optical fibers for distributed temperature measurement, achieving a maximum measurable temperature of 850°C, a test distance of 1 meter, and a spatial resolution on the order of centimeters. However, the above-mentioned OFDR high-temperature sensing methods based on Rayleigh scattering do not use ordinary optical fibers, resulting in higher manufacturing costs and an inability to achieve high temperature resolution, long distances, and temperature sensing at even higher temperatures.

[0003] Rayleigh scattering in optical fibers originates from random fluctuations in the fiber's refractive index. When the temperature changes at a certain point in the fiber, it causes a change in the Rayleigh scattering light at that location, resulting in a wavelength shift in the Rayleigh backscattering spectrum (RBS). Temperature is measured by using the spectral shift obtained through cross-correlation calculations of the Rayleigh scattering signals before and after the temperature change. However, near 1000℃, the correlation between two sets of Rayleigh scattering signals with long time intervals decreases sharply, causing a shift in the cross-correlation peak position and thus demodulating an incorrect temperature value. This is evident in Chen... [3] It was also mentioned in the reports.

[0004] In recent years, image denoising methods have been widely applied in OFDR distributed sensing. Since the information measured by OFDR systems often exhibits high similarity and redundancy, image processing methods can be used to denoise two-dimensional images. Examples include wavelet denoising and Gaussian filtering. [4] Non-local mean [5] Median filtering [6] Shape Adaptive Principal Component Analysis BM3D [7] All of these methods have been used in OFDR distributed strain sensing to improve sensing performance and measurement accuracy. Among these methods, shape adaptive principal component analysis (BM3D) has the best performance, but its long computation time due to the introduction of numerous complex wavelet transforms and discrete cosine transforms limits its use in practical scenarios. Furthermore, the use of image denoising to improve the performance of OFDR distributed high-temperature sensing has not yet been reported.

[0005] The Non-Local Haar Transform (NLH) image denoising method was developed by Yingkun Hou. [8] The algorithm proposed by [Author Name] improves image denoising by extending the block-level non-local self-similarity (NSS) of the image to the pixel-level NSS, making greater use of NSS priors. It achieves effective image denoising by performing a simple Haar transform on the most similar pixel groups while preserving image details well. The entire algorithm requires no matrix multiplication operations, significantly reducing computational complexity compared to the shape adaptive principal component analysis (BM3D) method.

[0006] References:

[0007] [1]Bulot, P., Bernard, R., Cieslikiewicz-Bouet, M., Laffont, G., and Douay, M. "Performance study of a zirconia-doped fiber for distributed temperaturesensing by OFDR at 800℃." Sensors, 21(11), article no.3788.

[0008] [2]A.K.Sang,D.K.Gifford,B.D.Dickerson,B.F.Fielder,and M.E.Froggatt,“One centimeter spatial resolution temperature measurements in a nuclearreactor using Rayleigh scatter in optical fiber,”Proc.SPIE 6619,66193D(2007).

[0009] [3]Chen,C.;Chen,L.;Bao,X.,“Distributed temperature profile inhydrogen flame measured by telecom fiber and its durability under flame byOFDR.”Opt.Express 2022,30,19390.

[0010] [4]S.Qu et al.,"High Spatial Resolution Investigation of OFDR Basedon Image Denoising Methods,"IEEE Sensors Journal,vol.21,no.17,pp.18871-18876,1Sept.1,2021.

[0011] [5]S.Zhao,et al.,“Accuracy improvement in OFDR based distributedsensing system by image processing,”Opt.,Lasers Eng.,vol.124,p.105824,Jan.2020.

[0012] [6]Q.Wang et al.,"Improving OFDR Distributed Fiber Sensing by FibersWith Enhanced Rayleigh Backscattering and Image Processing,"IEEE SensorsJournal,vol.22,no.19,pp.18471-18478,1Oct.1,2022.

[0013] [7] M.Pan et al., "Long Distance Distributed Strain Sensing in OFDR byBM3D-SAPCA Image Denoising," Journal of Lightwave Technology, 2022.

[0014] [8]Y.Hou et al., "NLH: A Blind Pixel-Level Non-Local Method for Real-World Image Denoising," IEEE Transactions on Image Processing, vol.29.

[0015] [9] I.Daubechies and W.Sweldens, "Factoring wavelet transforms into lifting steps," J.FourierAnal.Appl., vol.4, no.3, pp.247–269, 1998.

[0016]

[10] W.Sweldens, "The lifting scheme: A custom-design construction of biorthogonal wavelets," Appl.Comput.Harmon.Anal., vol.3, no.2, pp.186–200, Apr.1996. Summary of the Invention

[0017] This invention provides a distributed high-temperature measurement method using NLH image denoising and optical frequency domain reflectance, the technical solution of which is as follows:

[0018] A distributed optical frequency domain reflectance sensing method for image denoising based on nonlocal Haar transform includes the following steps:

[0019] The first step involves using an OFDR system to continuously collect data at different temperatures T0, T1, ..., T during the heating process of a high-temperature furnace, triggered by a tunable laser. n The Rayleigh scattering signal is given by n, where n is the total number of measurement groups, and the temperature step is Δt > 0, i.e., Δt = T1 - T0 = ... = T n -T n-1Two adjacent sets of signals are processed separately, for a total of n-1 sets. The set with the lower temperature is used as the reference signal, and the set with the higher temperature is used as the measurement signal. For Rayleigh scattering signals at temperatures T0 and T1, where T1 > T0, the Rayleigh scattering signal at temperature T0 is used as the reference signal S. r The Rayleigh scattering signal at temperature T1 is the measurement signal S. m The following steps 2 to 12 are for processing the Rayleigh scattering signals at temperatures T0 and T1. The same processing is performed on the remaining n-2 groups of signals.

[0020] The second step is to process the reference signal S at temperature T0. r and the measurement signal S at temperature T1 m Perform a Fast Fourier Transform to obtain the distance domain signal R at each location of the optical fiber at temperatures T0 and T1. r and R m ; For R r and R m Perform window filling and zero padding operations respectively to obtain R ri and R mi The window size is N, and the number of zero-padding points for each window is M, i∈[1,D], where D is the number of segments, and M and D satisfy M = number of valid points / D; the zero-padding process is to fill each segment R ri and R mi The data was filled into segments with (M+N) points, each segment R ri and R mi Each corresponds to Rayleigh scattering information at a specific location on the optical fiber at temperatures T0 and T1; subsequently, a fast inverse Fourier transform is used to convert each segment of R... ri and R mi The inverse transform back to the wavelength domain yields the local reference Rayleigh scattering spectrum S. ri And local measurement of Rayleigh scattering spectrum S mi Take S ri and S mi The amplitude was obtained by removing the DC component to get S. ri 'and S mi ';

[0021] The third step is to obtain group D S ri 'and S mi Perform cross-correlation calculations separately and normalize the results. Each set of normalized cross-correlation results corresponds to a position on the optical fiber. Arrange the D sets of normalized cross-correlation results along the corresponding positions on the optical fiber to obtain the three-dimensional cross-correlation map along each position on the optical fiber. Then project it onto a two-dimensional plane to obtain the original two-dimensional cross-correlation intensity map y.

[0022] The fourth step involves extracting an image patch of size N1×N1 from the original two-dimensional cross-correlation intensity map y obtained in the third step, using a specified step size N_step, as a reference patch. Then, an image patch of size N is extracted from this reference patch. s ×N s Within the neighborhood of the reference block, perform Euclidean distance-based block matching to obtain N2 similar image blocks, including the reference block itself. Then, perform column scanning on these N2 similar image blocks of size N1×N1 to transform them into N2 smaller blocks of size N1×N1. 2 A vector of size N1 × 1, which will be N2 vectors of size N1. 2 ×1 vectors are concatenated to form a vector of size. matrix Y l In Y l The above method calculates the Euclidean distance ED between each row and all other rows, using each row as a reference row. The N3 rows (including the reference row) that have the smallest Euclidean distance ED with the reference row are then combined to form a matrix Y of size N3 × N2. l ';

[0023] Fifth step, take the matrix Y of size N3×N2 obtained in the fourth step. l Perform a separable two-dimensional Haar transform, that is, perform Haar transforms in the vertical and horizontal directions respectively. Apply a lifting Haar transform during the transformation process, that is, perform a weighted average only between rows or columns to obtain the transform coefficients.

[0024] Step 6, double hard thresholding: In step 5, for a matrix Y of size N3×N2... l The result of applying the lifting Haar transform to the two rows with the smallest Euclidean distance is equivalent to a weighted difference between these two rows. The coefficients of these weighted differences are almost entirely from noise, so we set all such coefficients to zero. For the remaining Haar transform coefficients, we use a hard threshold to shrink the coefficients, with the hard threshold parameter being Thr.

[0025] Step 7: Perform an inverse Haar transform on the Haar transform coefficients after the double hard thresholding shrinkage in step 6 to obtain a denoised two-dimensional matrix. Then, scan this two-dimensional matrix back to the original image blocks. Finally, weighted summaries of these image blocks are placed back into their original positions to obtain a preliminary denoised image. To best preserve image details, an iterative strategy is adopted with K iterations. The preliminary denoised image obtained after K iterations is denoted as y. K The noisy image used in the k-th iteration, and the denoised image y obtained through the (k-1)-th iteration. k-1 According to y k =λy k-1 +(1-λ)y is calculated, where k=1,…,K,λ is the regularization parameter;

[0026] Step 8: The initial denoised image y from step 7, after K iterations...K Simultaneously perform block matching and row matching operations on the original noisy image y obtained in step 3; perform block matching and row matching operations on y at a specified step size N_step. K Extract an image patch of size N1×N1 as a reference patch, and then extract an image patch of size N centered on this reference patch. s ×N s Within the neighborhood of the reference block, perform Euclidean distance-based block matching to obtain N2 similar image blocks, including the reference block. Then, perform column scanning on these N2 similar image blocks of size N1×N1 to transform them into N2 smaller blocks of size N1×N1. 2 A vector of size N1 × 1, which will be N2 vectors of size N1. 2 ×1 vectors are concatenated to form a vector of size. matrix Y K In Y K The Euclidean distance ED between each row and all other rows is calculated. The N3 rows with the smallest Euclidean distance to the reference rows (including the reference rows) are then used to form a matrix Y′ of size N3×N2. K The original noisy image y is subjected to synchronous, same-parameter operations to obtain a matrix Y of size N3×N2. l After block matching and row matching, two matrices Y′ of size N3×N2 are obtained. K and Y l ';

[0027] Step 9: Combine the two matrices Y′ of size N3×N2 obtained in step 8. K and Y l Perform the same separable two-dimensional Haar transform separately, i.e., perform Haar transforms vertically and horizontally respectively, and apply a lifting Haar transform during the transformation process, i.e., perform a weighted average only between rows or columns; where Y′ K The transformation coefficients are θ1, Y l The transformation coefficient of ' is θ2;

[0028] Step 10: Perform Wiener filtering according to formula (1):

[0029]

[0030] Step 11: To improve noise reduction performance, Y′ K The transform coefficients θ1 and the coefficients obtained by Wiener filtering Perform the Wiener filtering as shown in step 10 again, i.e., perform 2 iterations; perform the inverse Haar transform on the coefficients obtained after 2 iterations to obtain the denoised 2D matrix, then scan this 2D matrix back to the original image blocks, and finally weight these image blocks and place them back into their original positions in the image to obtain the final denoised image y. denoised ;

[0031] Step 12: [The text appears to be incomplete and contains several grammatical errors and inconsistencies. A more accurate translation would require the full context.] denoised By converting it to three dimensions, we can obtain the cross-correlation three-dimensional plot of each position in the fiber corresponding to the temperature change range T0 to T1 after noise reduction; then, by finding the peak position of the cross-correlation result of each position in the fiber, we can obtain the spectral frequency shift distribution of the temperature change range T0 to T1.

[0032] Step 13: After performing steps 2 through 12 on the Rayleigh scattering signals of n-1 groups of adjacent temperature values, n-1 spectral frequency shift distributions are obtained, corresponding to the n-1 temperature variation intervals T0~T1, T1~T2,…,T n-1 ~T n Then, by summing the n-1 spectral frequency shift distributions, we can obtain the distribution corresponding to the temperature change T0 to T1. n The spectral frequency shift distribution.

[0033] The beneficial effects of the technical solution provided by this invention are:

[0034] 1. Distributed high-temperature sensing with a test distance of 100m and a spatial resolution of 2cm was achieved using thin-diameter optical fiber;

[0035] 2. A distributed high-temperature measurement system with a temperature resolution of 1℃ and a temperature measurement range of 950℃~1050℃ was achieved. After processing with the NLH image denoising algorithm, a high linearity OFDR system with an R-square value of 0.9999 was obtained.

[0036] 3. It overcomes the problem that when optical fibers are used for 1000℃ measurements, the large temperature interval leads to poor signal similarity, which in turn makes measurement impossible. Attached Figure Description

[0037] Figure 1 This is a distributed high-temperature sensing device based on optical frequency domain reflection;

[0038] Figure 2 Comparison of 3D cross-correlation distribution before and after NLH denoising, and comparison of spectral frequency shift results obtained by finding peaks before and after NLH denoising;

[0039] Figure 3 The temperature range is 1000℃~1010℃, with a temperature interval of 1℃. The cumulative results before and after NLH noise reduction are compared.

[0040] Figure 4 With a spatial resolution of 2cm and a temperature measurement range of 950℃~1050℃, the comparison before and after NLH denoising and the linear fitting graph after NLH denoising are presented.

[0041] Figure 5 Flowchart of a distributed high-temperature sensing method based on NLH image denoising OFDR;

[0042] The attached diagram lists the components represented by each number as follows:

[0043] 1: Tunable laser; 2: First balanced detector;

[0044] 3: 20:80 polarization-maintaining coupler; 4: 1:99 polarization-maintaining beam splitter;

[0045] 5: 50:50 coupler; 6: Delay fiber;

[0046] 7: First Faraday rotation mirror; 8: Second Faraday rotation mirror;

[0047] 9: First circulator; 10: Second circulator;

[0048] 11: Reference arm; 12: Test arm;

[0049] 13: Light mixer; 14: High-temperature furnace;

[0050] 15: Analog-to-digital data acquisition device; 16: Second balance detector;

[0051] 17: Third balanced detector; 18: Sensing fiber optic cable;

[0052] 19: Computer; 20: USB control cable;

[0053] 21: Main interferometer; 22: Clock triggering device based on auxiliary interferometer Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below.

[0055] In this patent, during the heating process of the high-temperature furnace, the temperature interval was 1℃, and the measurement range was from 950℃ to 1050℃. An OFDR system was used to collect backscattered Rayleigh signals from a thin-diameter optical fiber at these 101 temperature values. During this process, due to the high temperature, the similarity between two sets of backscattered Rayleigh signals with a long time interval would decrease. Therefore, the high-temperature furnace needed to heat up as quickly as possible. We used a 1℃ temperature interval, with a time interval of approximately 4 seconds. This ensured both high similarity between the two sets of backscattered Rayleigh signals at adjacent temperature values ​​and sufficient time to collect signals at each temperature value. We grouped the backscattered Rayleigh signals at adjacent temperature values ​​(i.e., with a 1℃ interval) into one group (a total of 100 groups) for subsequent processing. Here, we only use 950℃ and 951℃ as examples; signals at other adjacent temperature values ​​were processed using the same method. We denote the signal with the lower temperature value (950℃) as the reference signal and the signal with the higher temperature value (951℃) as the measurement signal. We process these two signals using FFT, windowing and zero-padding, IFFT, DC removal, and cross-correlation to obtain the local cross-correlation distribution of the optical fiber. We then arrange the local cross-correlation distributions at each window position sequentially along the optical fiber to obtain a three-dimensional cross-correlation distribution. We project the obtained three-dimensional cross-correlation distribution onto a two-dimensional plane to convert it into a two-dimensional image. Subsequently, we use the NLH image denoising algorithm to process this two-dimensional image, thereby effectively removing random noise. We then convert this two-dimensional image into a three-dimensional image. By finding the peak positions of each window, we can obtain the spectral frequency shift distribution of 950℃ to 951℃. After processing 100 sets of backscattered Rayleigh signals from adjacent temperature values ​​using the above method, the spectral frequency shift distributions from 950℃ to 951℃ and from 951℃ to 952℃ are added together to obtain the spectral frequency shift distribution from 950℃ to 952℃. Similarly, by adding the spectral frequency shift distributions obtained after processing 100 sets of backscattered Rayleigh signals from adjacent temperature values, the spectral frequency shift distribution from 950℃ to 1050℃ can be obtained. Based on this method, we achieved a temperature measurement range of 950℃ to 1050℃ using a narrow-diameter optical fiber, with a spatial resolution of 2 cm and a temperature resolution of 1℃. The fitted spectral frequency shift values ​​showed a linear relationship with temperature changes, with an R-square of 0.9999.

[0056] Example 1:

[0057] The OFDR device used in this invention is basically the same as the OFDR device used in the patent previously applied for by Ding Zhenyang's research group. However, in order to keep this case easy to understand, the device used will be described below.

[0058] This example includes a distributed fiber optic sensing device based on an optical frequency domain reflectivity system.

[0059] The distributed fiber optic sensing device for optical frequency domain reflection includes: a tunable laser 1, a 1:99 polarization-maintaining beam splitter 4, a computer 19, a USB control cable 20, an analog-to-digital acquisition device 15, a clock triggering device based on an additional interferometer 22, and a main interferometer 21.

[0060] The clock triggering device 22 based on the auxiliary interferometer includes: a first balanced detector 2, a 50:50 coupler 5, a delay fiber 6, a first Faraday rotating mirror 7, a second Faraday rotating mirror 8, and a first circulator 9. The clock triggering device 22 based on the auxiliary interferometer is used to achieve equal optical frequency interval sampling, the purpose of which is to compensate for the nonlinear frequency sweep of the light source.

[0061] The main interferometer 21 includes: a 20:80 polarization-maintaining coupler 3, a second circulator 10, an optical mixer 13, a second balanced detector 16, a third balanced detector 17, a reference arm 11, a test arm 12, a high-temperature furnace 14, and a sensing fiber 18, which is a thin-diameter fiber. The main interferometer 21 is the core of the distributed fiber optic sensing device for optical frequency domain reflection, and it is an improved Mach-Zehnder interferometer.

[0062] The output of USB control line 20 is connected to the input of tunable laser 1; the input of USB control line 20 is connected to the output of computer 19; tunable laser 1 is connected to port a of 1:99 polarization-maintaining beam splitter 4; port b (1% splitting port) of 1:99 polarization-maintaining beam splitter 4 is connected to port a of first circulator 9; port c (99% splitting port) of 1:99 polarization-maintaining beam splitter 4 is connected to port a of 20:80 polarization-maintaining coupler 3; port b of first circulator 9 is connected to port a of 50:50 coupler 5; port c of first circulator 9 is connected to the input of first balanced detector 2; port b of 50:50 coupler 5 is connected to the input of first balanced detector 2; port c of 50:50 coupler 5 is connected to first Faraday mirror 7 via delay fiber 6; port d of 50:50 coupler 5 is connected to second Faraday mirror 8; the output of first balanced detector 2... The output of the optical mixer 10 is connected to the input of the analog-to-digital acquisition device 15; the c port of the 20:80 polarization-maintaining coupler 3, i.e., the 20% splitting port, is connected to the input a of the optical mixer 13 via the reference arm 11; the d port of the 20:80 polarization-maintaining coupler 3, i.e., the 80% splitting port, is connected to the a port of the second circulator 10 via the test arm 12; the c port of the second circulator 10 is connected to the sensing fiber 18; the b port of the second circulator 10 is connected to the input b of the optical mixer 13; the c and d outputs of the optical mixer 13 are connected to the two inputs of the second balanced detector 16; the e and f outputs of the optical mixer 13 are connected to the two inputs of the third balanced detector 17; the output of the second balanced detector 16 is connected to the input of the analog-to-digital acquisition device 15; the output of the third balanced detector 17 is connected to the input of the analog-to-digital acquisition device 15; and the output of the analog-to-digital acquisition device 15 is connected to the input of the computer 19.

[0063] When the device is working, the computer 19 controls the tunable laser 1 via the USB control line 20, controlling the tuning speed, center wavelength, tuning start, etc. The output light of the tunable laser 1 enters through port a of the 1:99 polarization-maintaining beam splitter 4, enters through port b (1% splitting port) of the 1:99 polarization-maintaining beam splitter 4, enters through port a of the first circulator 9, enters through port b of the first circulator 9, enters through port a of the 50:50 coupler 5, 50% exits through port c of the coupler 5, and 50% exits through port d of the coupler 5. The output light from port c of the 50:50 coupler 5... The delayed fiber 6 is reflected by the first Faraday rotator 7 and returns to port c of the 50:50 coupler 5. The outgoing light from port d of the 50:50 coupler 5 is reflected by the second Faraday rotator 8 and returns to port d of the 50:50 coupler 5. The two beams interfere in the 50:50 coupler 5 and are output from port b of the 50:50 coupler 5. The output light enters the first balanced detector 2, which converts the detected light signal into an interference beat frequency signal and transmits it to the analog-to-digital acquisition device 15 as an external clock signal for the analog-to-digital acquisition device 15.

[0064] The light enters from port C (99% splitting port) of the 1:99 optical beam splitter 4 to port A of the 20:80 polarization-maintaining coupler 3; after passing through port C (20% splitting port) of the 20:80 polarization-maintaining coupler 3, it enters from port D (80% splitting port) into the reference arm 11, and from port A (80% splitting port) into the test arm 12. Light enters from port A of the second circulator 10, and from port C of the second circulator 10 into the sensing fiber 18. The backscattered light from the sensing fiber 18 enters from port C of the second circulator 10 and exits from port B of the second circulator 10. The reference light output from the reference arm 11 enters from port A of the optical mixer 13 and from port B of the second circulator 10 into the optical... The test light at port b of the mixer 13 is combined to form a beat frequency interference. The optical mixer 13 is input to the second balanced detector 16 through ports c and d. The second balanced detector 16 converts the detected optical signal into an interference beat frequency signal and transmits it to the analog-to-digital acquisition device 15. The optical mixer 13 is input to the third balanced detector 17 through ports e and f. The second balanced detector 17 converts the detected optical signal into an interference beat frequency signal and transmits it to the analog-to-digital acquisition device 15. The analog-to-digital acquisition device 15 transmits the acquired analog electrical signal to the computer 19 under the action of the external clock signal generated by the clock triggering device 22 of the additional interferometer.

[0065] USB control cable 20 is used by computer 19 to control tunable laser 1.

[0066] Tunable laser 1 provides a light source for the optical frequency domain reflection system, and its optical frequency can be linearly scanned.

[0067] The first circulator 9 prevents reflected light from port b of the 50:50 coupler 5 in the additional interferometer from entering the laser.

[0068] 50:50 coupler 5 is used for optical interference.

[0069] Delayed fiber 6 is used to realize beat frequency interference with non-equidistant arms, and the optical frequency can be obtained from the beat frequency and the length of the delay fiber.

[0070] The high-temperature furnace 14 is used to cause temperature changes in the sensing optical fiber 18.

[0071] The first Faraday mirror 7 and the second Faraday mirror 8 are used to provide reflection for the interferometer and can eliminate the polarization fading phenomenon of the interferometer.

[0072] The optical mixer 13 performs polarization beam splitting on the signal, so that the light intensity of the reference light and the test light is basically the same in the two orthogonal directions when the polarization beam splitting is performed, eliminating the influence of polarization fading noise, realizing the beam combining of the reference light and the test light, and forming beat frequency interference.

[0073] Computer 19: Processes the interference signal acquired by analog-to-digital acquisition device 15 to realize fiber optic sensing based on optical frequency domain reflection and distributed measurement of high temperature using narrow-diameter optical fiber.

[0074] The present invention utilizes NLH image denoising and distributed high-temperature measurement method with optical frequency domain reflectance, the specific steps of which are as follows:

[0075] The first step involves using an OFDR system to continuously collect data at different temperatures T0, T1, ..., T during the heating process of a high-temperature furnace, triggered by a tunable laser. n The Rayleigh scattering signal is given by n, where n is the total number of measurement groups, and the temperature step is Δt > 0, i.e., Δt = T1 - T0 = ... = T n -T n-1 Two adjacent sets of signals are processed (n-1 sets in total). The set with the lower temperature is used as the reference signal, and the set with the higher temperature is used as the measurement signal. For example, if Rayleigh scattering signals at temperatures T0 and T1 are taken, where T1 > T0, the Rayleigh scattering signal at temperature T0 is used as the reference signal S. r The Rayleigh scattering signal at temperature T1 is the measurement signal S. m The following steps only use Rayleigh scattering signals at temperatures T0 and T1 as examples; the other n-2 groups of signals are processed in the same way.

[0076] The second step is to process the reference signal S at temperature T0. r and the measurement signal S at temperature T1 m Perform a Fast Fourier Transform to obtain the distance domain signal R at each location of the optical fiber at temperatures T0 and T1. r and R m For R r and R m Perform window filling and zero padding operations respectively to obtain Rri and R mi The number of window points is N, and the number of zero-padding points in each window is M, i∈[1,D], where D is the number of segments, and N and D satisfy M = number of valid points / D. The zero-padding process involves padding each segment R... ri and R mi The data was filled into segments with (M+N) points, each segment R ri and R mi These correspond to Rayleigh scattering information at specific locations on the optical fiber at temperatures T0 and T1. Subsequently, a fast inverse Fourier transform is used to convert each segment of R... ri and R mi The inverse transform back to the wavelength domain yields the local reference Rayleigh scattering spectrum S. ri And local measurement of Rayleigh scattering spectrum S mi Take S ri and S mi The amplitude was obtained by removing the DC component to get S. ri 'and S mi '.

[0077] The third step is to obtain group D S ri 'and S mi Perform cross-correlation calculations separately and normalize the results. Each set of normalized cross-correlation results corresponds to a position on the optical fiber. Arrange the D sets of normalized cross-correlation results along the corresponding positions on the optical fiber to obtain a three-dimensional cross-correlation map along each position on the optical fiber. Project this map onto a two-dimensional plane to obtain the original two-dimensional cross-correlation intensity map y.

[0078] The fourth step involves extracting an image patch of size N1×N1 from the original two-dimensional cross-correlation intensity map y obtained in the third step, using a specified step size N_step, as a reference patch. Then, an image patch of size N is extracted from this reference patch. s ×N s Within the neighborhood of the reference block, perform Euclidean distance-based block matching to obtain N2 similar image blocks (including the reference block itself). Then, perform column scanning on these N2 similar image blocks of size N1×N1 to transform them into N2 similar image blocks of size N1. 2 A vector of size N1 × 1, which will be N2 vectors of size N1. 2 ×1 vectors are concatenated to form a vector of size. matrix Y l In Y l The above method calculates the Euclidean distance ED between each row and all other rows, using each row as a reference row. The N3 rows (including the reference row itself) that have the smallest Euclidean distance ED with the reference row are then combined into a matrix Y of size N3 × N2. l '.

[0079] Fifth step, take the matrix Y of size N3×N2 obtained in the fourth step. lPerform a separable two-dimensional Haar transform, that is, perform Haar transforms on the vertical and horizontal sides respectively. During the transformation process, a lifting Haar transform is applied, that is, a weighted average is performed only between rows or columns to obtain the transform coefficients.

[0080] Step 6, double hard thresholding shrinkage. In step 5, for a matrix Y of size N³×N²... l The result of applying the lifting Haar transform to the two rows with the smallest Euclidean distance is equivalent to a weighted difference between these two rows. The coefficients of these differences are almost entirely derived from noise, so we set all such coefficients to zero. For the remaining Haar transform coefficients, we use a hard threshold to shrink the coefficients, with the hard threshold parameter being Thr.

[0081] Step 7: Perform an inverse Haar transform on the Haar transform coefficients after the double hard thresholding shrinkage in step 6 to obtain a denoised two-dimensional matrix. Then, scan this two-dimensional matrix back to the original image blocks. Finally, weighted summaries of these image blocks are placed back into their original positions to obtain a preliminary denoised image. Since the number of most similar pixels participating in denoising is limited each time, Thr cannot be too large in order to best preserve image details. Performing the above steps only once is unlikely to remove all image noise. Therefore, an iterative strategy is adopted, with K iterations. The preliminary denoised image obtained after K iterations is denoted as y. K The noisy image used in the k-th iteration, and the denoised image y obtained through the (k-1)-th iteration. k-1 According to y k =λy k-1 +(1-λ)y is calculated, where k=1,…,K,λ is the regularization parameter.

[0082] Step 8: The initial denoised image y from step 7, after K iterations... K Block matching and row matching operations are performed synchronously with the original noisy image y obtained in step three. The matching is performed at a specified step size N_step on y. K Extract an image patch of size N1×N1 as a reference patch, and then extract an image patch of size N centered on this reference patch. s ×N s Within the neighborhood of the reference block, perform Euclidean distance-based block matching to obtain N2 similar image blocks (including the reference block itself). Then, perform column scanning on these N2 similar image blocks of size N1×N1 to transform them into N2 similar image blocks of size N1. 2 A vector of size N1 × 1, which will be N2 vectors of size N1. 2 ×1 vectors are concatenated to form a vector of size. matrix Y K In Y KThe above method calculates the Euclidean distance ED between each row as a reference row and all other rows. The N3 rows (including the reference row itself) with the smallest Euclidean distance to the reference row are then combined to form a matrix Y′ of size N3 × N2. K The original noisy image y is subjected to synchronized, same-parameter operations to obtain a matrix Y of size N3×N2. l After block matching and row matching, two matrices Y' of size N3×N2 are obtained. K and Y l ′.

[0083] Step 9: Combine the two matrices Y′ of size N3×N2 obtained in step 8. K and Y l The same separable two-dimensional Haar transform is performed on each of the rows and columns, respectively. A lifting Haar transform is applied during the transformation process, meaning a weighted average is performed only between rows or columns. Where Y′ K The transformation coefficients are θ1, Y l The transformation coefficient of ′ is θ2.

[0084] Step 10: Perform Wiener filtering according to formula (1).

[0085]

[0086] Step 11: To improve noise reduction performance, Y′ K The transform coefficients θ1 and the coefficients obtained by Wiener filtering Perform the Wiener filtering as shown in step 10 again, i.e., perform 2 iterations. Apply the inverse Haar transform to the coefficients obtained after 2 iterations to obtain the denoised 2D matrix. Then scan this 2D matrix back to the original image blocks. Finally, weighted summaries of these image blocks are placed back into their original positions in the image to obtain the final denoised image y. denoised .

[0087] Step 12: [The text appears to be incomplete and contains several grammatical errors and inconsistencies. A more accurate translation would require the full context.] denoised Converting to three dimensions yields the denoised cross-correlation 3D plots corresponding to various locations in the fiber within the temperature range T0–T1. Then, by finding the peak position of the cross-correlation results at each location in the fiber, the spectral frequency shift distribution within the temperature range T0–T1 can be obtained.

[0088] Step 13: After performing steps 2 through 12 on the Rayleigh scattering signals of n-1 = 100 adjacent temperature values, n-1 = 100 spectral frequency shift distributions will be obtained, corresponding to n-1 = 100 temperature variation intervals T0 ~ T1 (950℃ ~ 951℃), T1 ~ T2 (951℃ ~ 952℃), ..., T n-1 ~T nThat is, 1049℃~1050℃. Then, by adding up n-1=100 spectral frequency shift distributions, we can obtain the temperature change range T0~T n That is, the spectral frequency shift distribution from 950℃ to 1050℃.

[0089] Example 2:

[0090] This invention provides a distributed high-temperature measurement method using NLH image denoising and optical frequency domain reflectance, the technical solution of which is as follows:

[0091] The first step involves continuously collecting data at different temperatures T0, T1, ..., T during the heating process of a high-temperature furnace, triggered by a tunable laser, using an OFDR system. n The Rayleigh scattering signal is given by n, where n is the total number of measurement groups, and the temperature step is Δt > 0, i.e., Δt = T1 - T0 = ... = T n -T n-1 Two adjacent sets of signals are processed (n-1 sets in total). The set with the lower temperature is used as the reference signal, and the set with the higher temperature is used as the measurement signal. For example, if Rayleigh scattering signals at temperatures T0 and T1 are taken, where T1 > T0, the Rayleigh scattering signal at temperature T0 is used as the reference signal S. r The Rayleigh scattering signal at temperature T1 is the measurement signal S. m In this patent, T0 = 950℃, T1 = 951℃, T n-1 =1049℃, T n =1050℃, n=101, Δt=1℃, T0, T1,…,T n-1 ,T n The selection of Δt needs to consider the temperature measurement range and accuracy. The selection of Δt needs to consider maintaining a high degree of similarity between adjacent signal groups and the temperature regulation accuracy of the high-temperature furnace. n is related to the temperature measurement range and the temperature measurement interval. In the following steps, only the Rayleigh scattering signals at the temperatures of T0 = 950℃ and T1 = 951℃ are used as examples. The other n-2 signal groups are processed in the same way.

[0092] The second step is to process the reference signal S at a temperature of T0 = 950℃. r The measurement signal S at temperature T1 = 951℃ m Performing a Fast Fourier Transform (FFT) yields the distance-domain signal R at various locations on the optical fiber at temperatures T0 = 950℃ and T1 = 951℃. r and R m For R r and R m Perform window filling and zero padding operations respectively to obtain R ri and R miThe number of window points is N, and the number of zero-padding points for each window is M, i∈[1,D], where D is the number of segments. N and D satisfy M = number of effective points / D. In this patent, N = 200, M = 5050. The selection of N and M needs to consider the spatial resolution and spectral fineness of the OFDR system. The number of effective points is 1,050,000, and D = 5250. The zero-padding process will pad each segment R... ri and R mi The data was filled into segments with M+N=5250 points, each segment R ri and R mi These correspond to Rayleigh scattering information at specific locations on the optical fiber at temperatures of T0 = 950℃ and T1 = 951℃. Subsequently, a fast inverse Fourier transform is used to convert each segment of R... ri and R mi The inverse transform back to the wavelength domain yields the local reference Rayleigh scattering spectrum S. ri And local measurement of Rayleigh scattering spectrum S mi Take S ri and S mi The amplitude was obtained by removing the DC component to get S. ri 'and S mi '.

[0093] The third step is to obtain D=5250 sets of S ri 'and S mi Cross-correlation calculations are performed separately, and the results are normalized. Each set of normalized cross-correlation results corresponds to a position on the optical fiber. By arranging the D=5250 sets of normalized cross-correlation results along the corresponding positions on the optical fiber, a three-dimensional cross-correlation map along each position on the optical fiber can be obtained. This map is then projected onto a two-dimensional plane to obtain the original two-dimensional cross-correlation intensity map y.

[0094] The fourth step involves extracting an image patch of size N1×N1 from the original two-dimensional cross-correlation intensity map y obtained in the third step, using a specified step size N_step, as a reference patch. Then, an image patch of size N is extracted from this reference patch. s ×N s Within the neighborhood of the reference block, perform Euclidean distance-based block matching to obtain N2 similar image blocks (including the reference block itself). Then, perform column scanning on these N2 similar image blocks of size N1×N1 to transform them into N2 similar image blocks of size N1. 2 A vector of size N1 × 1, which will be N2 vectors of size N1. 2 ×1 vectors are concatenated to form a vector of size. matrix Y l In Y l The above method calculates the Euclidean distance ED between each row and all other rows, using each row as a reference row. The N3 rows (including the reference row itself) that have the smallest Euclidean distance ED with the reference row are then combined into a matrix Y of size N3 × N2.l In this patent, N_step = 6, N1 = 9, N s =43, N2=16, N3=4, N_step, N1, N s The selection of N1, N2, and N3 needs to take into account the noise level of the original noisy image.

[0095] The fifth step is to process the matrix Y obtained in the fourth step, which is of size N3×N2, or 4×16. l Perform a separable two-dimensional Haar transform, that is, perform Haar transforms in both the vertical and horizontal directions, and apply a lifting Haar transform during the transformation process. [9]

[10] This means that only a weighted average is performed between rows or columns to obtain the transformation coefficients.

[0096] Step 6, double hard thresholding shrinkage. In step 5, the matrix Y of size N3×N2 (4×16) is... l The two rows with the smallest Euclidean distance are subjected to a lifting Haar transform, which is equivalent to a weighted difference between these two rows. The coefficients of these differences are almost entirely derived from noise, and these coefficients are all set to zero. The remaining Haar transform coefficients are subjected to a hard thresholding process, with the hard threshold parameter being Thr. In this patent, Thr = 1.65, and the selection of Thr needs to consider the noise level of the original noisy image.

[0097] Step 7: Perform an inverse Haar transform on the Haar transform coefficients after the double hard thresholding shrinkage in step 6 to obtain a denoised two-dimensional matrix. Then, scan this two-dimensional matrix back to the original image blocks. Finally, weighted summaries of these image blocks are placed back into their original positions to obtain a preliminary denoised image. Since the number of most similar pixels participating in denoising is limited each time, Thr cannot be too large in order to best preserve image details. Performing the above steps only once is unlikely to remove all image noise. Therefore, an iterative strategy is adopted, with K iterations. The preliminary denoised image obtained after K iterations is denoted as y. K The noisy image used in the k-th iteration, and the denoised image y obtained through the (k-1)-th iteration. k-1 According to y k =λy k-1 The formula is obtained by calculating +(1-λ)y, where k = 1, ..., K, and λ is the regularization parameter. In this patent, K = 3 and λ = 0.6. The selection of K and λ needs to take into account the noise level of the original noisy image.

[0098] Step 8: Simultaneously perform block matching and row matching operations on the preliminary denoised image y3 obtained from step 7 after K=3 iterations and the original noisy image y obtained from step 3. Extract an image block of size N1×N1 (9×9) from y3 at a specified step size N_step=6 as a reference block. Then, within this reference block, in a new image block of size N...s ×N s That is, block matching based on Euclidean distance is performed within a 43×43 neighborhood to obtain N2=16 similar image blocks (including the reference block itself). These N2=16 similar image blocks of size N1×N1 (i.e., 9×9) are then transformed into N2=16 similar image blocks of size N1 by column scanning. 2 ×1 is a vector of size 81×1, and N2 = 16 vectors of size N1 2 ×1, that is, 81×1 vectors concatenated to form a vector of size. That is, a matrix Y3 of size 81×16. Each row of Y3 is used as a reference row, and the Euclidean distance ED is calculated between each row and all other rows. The N3 = 4 rows (including the reference row itself) with the smallest Euclidean distance to the reference rows are combined to form a matrix Y3′ of size N3×N2 (4×16). The original noisy image y is then subjected to synchronized, same-parameter operations to obtain a matrix Y of size N3×N2 (4×16). l After block matching and row matching, two matrices Y3′ and Y′ of size N3×N2 (4×16) are obtained. l ′.

[0099] Step 9: Combine the two matrices Y′ of size N3×N2 (4×16) obtained in step 8. K and Y l The same separable two-dimensional Haar transform is performed on each of the rows and columns, respectively. A lifting Haar transform is applied during the transformation process, meaning a weighted average is performed only between rows or columns. Where Y′ K The transformation coefficients are θ1, Y l The transformation coefficient of ′ is θ2.

[0100] Step 10: Perform Wiener filtering according to formula (1).

[0101]

[0102] In this patent, σ = 25, and the selection of σ needs to take into account the denoising effect and the degree of detail preservation of the image.

[0103] In the eleventh step, to improve the denoising performance, the transform coefficients θ1 of Y′3 are compared with the coefficients obtained from Wiener filtering. Perform the Wiener filtering as shown in step 10 again, i.e., perform 2 iterations. Apply the inverse Haar transform to the coefficients obtained after 2 iterations to obtain the denoised 2D matrix. Then scan this 2D matrix back to the original image blocks. Finally, weighted summaries of these image blocks are placed back into their original positions in the image to obtain the final denoised image y. denoised .

[0104] Step 12: [The text appears to be incomplete and contains several grammatical errors and inconsistencies. A more accurate translation would require the full context.] denoisedConverting to three dimensions yields the denoised cross-correlation 3D plots of various locations on the optical fiber corresponding to the temperature range T0~T1 (950℃~951℃). Then, by finding the peak position of the cross-correlation results at each location on the optical fiber, the spectral frequency shift distribution within the temperature range T0~T1 (950℃~951℃) can be obtained.

[0105] Step 13: After performing steps 2 through 12 on the Rayleigh scattering signals of n-1 = 100 adjacent temperature values, n-1 = 100 spectral frequency shift distributions will be obtained, corresponding to n-1 = 100 temperature variation intervals T0 ~ T1 (950℃ ~ 951℃), T1 ~ T2 (951℃ ~ 952℃), ..., T n-1 ~T n That is, 1049℃~1050℃. Then, by adding up n-1=100 spectral frequency shift distributions, we can obtain the temperature change range T0~T n That is, the spectral frequency shift distribution from 950℃ to 1050℃.

[0106] Step 13: After performing steps 2 through 12 on the Rayleigh scattering signals of n-1 = 100 adjacent temperature values, n-1 = 100 spectral frequency shift distributions will be obtained, corresponding to n-1 = 100 temperature variation intervals T0 ~ T1 (950℃ ~ 951℃), T1 ~ T2 (951℃ ~ 952℃), ..., T n-1 ~T n That is, 1049℃~1050℃. Then, by adding up n-1=100 spectral frequency shift distributions, we can obtain the temperature change range T0~T n That is, the spectral frequency shift distribution from 950℃ to 1050℃.

[0107] The feasibility of NLH's noise reduction effect on the OFDR distributed high-temperature measurement system is verified through specific experiments, as detailed below:

[0108] Example 3:

[0109] The sensing fiber 18 used in this embodiment of the invention is a narrow-diameter fiber with a total length of approximately 102.5 m. In the experiment, the OFDR system was tuned to 8.33 nm, with a sweep rate of 100 nm / s and 5 M sampling points. The additional interferometer fiber length was 500 m. The number of window points N was 200, corresponding to a spatial resolution of 2 cm, and the number of zero-padding points M was 5050.

[0110] Figure 2 a is the original cross-correlation 3D image at 1℃ intervals. Figure 2b is the spectral shift result obtained by finding the peak position. It can be seen that, due to the influence of noise, although the ambient temperature changes at all locations of the optical fiber are the same, the magnitude of the spectral shift exhibits irregular fluctuations. Figure 2 c is the cross-correlation 3D image obtained after NLH image denoising. Figure 2 d represents the spectral frequency shift result obtained after NLH image denoising. It can be seen that after NLH image denoising, the temperature distribution on the optical fiber becomes more stable, indicating that NLH image denoising improves the performance of OFDR temperature measurement.

[0111] To demonstrate the effectiveness of this method for distributed measurements at 1000℃, as mentioned earlier, we add the denoised spectral frequency shift distributions of the two Rayleigh scattering signals from 1001℃ to 1002℃ to the same denoised spectral frequency shift distribution, thus obtaining the denoised spectral frequency shift distributions for the two Rayleigh scattering signals from 1000℃ to 1001℃. By analogy, we can obtain the cumulative results for the range of 1000℃ to 1010℃, with a temperature resolution of 1℃ and a spatial resolution of 2cm. Figure 3 a represents the original cumulative result at 1000℃~1010℃, with a temperature interval of 1℃ and a spatial resolution of 2cm. Figure 3 b represents the cumulative result after NLH denoising, with a temperature interval of 1℃ and a spatial resolution of 2cm, ranging from 1000℃ to 1010℃.

[0112] Example 4:

[0113] To verify the linear relationship between the magnitude of the spectral shift obtained by the OFDR system after denoising using NLH images and the temperature change, we obtained the spectral shift results for the following temperature ranges by summation: 950℃~960℃ (10℃ interval), 950℃~970℃ (20℃ interval), 950℃~980℃ (30℃ interval), 950℃~990℃ (40℃ interval), 950℃~1000℃ (50℃ interval), 950℃~1010℃ (60℃ interval), 950℃~1020℃ (70℃ interval), 950℃~1030℃ (80℃ interval), 950℃~1040℃ (90℃ interval), and 950℃~1050℃ (100℃ interval). The temperature distribution before NLH denoising is shown below. Figure 4 As shown in figure a, the temperature distribution after noise reduction is as follows: Figure 4 As shown in b, by Figure 4 The result of b is obtained by fitting. Figure 4 The linear fitting results for c. From... Figure 4 As can be seen from c, the linearity of the system after denoising is very high, with an R-square value of 0.9999, and the spectral frequency shift is linearly related to temperature change.

[0114] In summary, our NLH image denoising algorithm denoises the two-dimensional cross-correlation distribution map of distributed high-temperature measurements, enabling an OFDR system that measures the end of a 100m narrow-diameter fiber with a spatial resolution of 2cm, a temperature resolution of 1℃, and a temperature measurement range of 950℃ to 1050℃. The denoised OFDR system exhibits high linearity, with an R-squared value of 0.9999.

[0115] Unless otherwise specified, the model numbers of the various devices in this embodiment of the invention are not limited, and any device that can perform the above functions is acceptable.

[0116] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the above-described embodiment numbers are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. The above descriptions are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-local Haar transform based image denoising optical frequency domain reflectometry distributed sensing method comprising the following steps: Firstly, the Rayleigh scattering signals at different temperatures are continuously collected by the OFDR system during the heating process in the high-temperature furnace under the trigger of the tunable laser is the total number of measurement groups, and the temperature step is , the adjacent two groups of signals are taken for processing, a total of groups, the lower temperature group is the reference signal, and the higher temperature group is the measurement signal; for the Rayleigh scattering signals at and two temperatures, , the Rayleigh scattering signal at the lower temperature is the reference signal , the Rayleigh scattering signal at the higher temperature is the measurement signal , the following second to twelfth steps are the processing of the Rayleigh scattering signals at and two temperatures, and the remaining groups of signals are processed in the same way;​​ The second step is to separately... Reference signal at temperature and Measurement signal at temperature Perform a fast Fourier transform to obtain and Distance domain signals corresponding to various locations on the optical fiber at two temperatures and ;right and Perform window filling and zeroing operations respectively to obtain and The number of window points is The number of zero-point padding points for each window is , , For the number of segments, M and Between satisfy ; The zero-padding process will fill each segment and Filled with Each data segment contains points, and each segment... and Each corresponds to a specific location on the optical fiber. and Rayleigh scattering information at two temperatures; then, a fast inverse Fourier transform is used to analyze each segment. and The inverse transform back to the wavelength domain yields the local reference Rayleigh scattering spectrum. and local measurement of Rayleigh scattering spectrum ;Pick and The amplitude was obtained by removing the DC component. and ; Thirdly, the obtained groups and respectively carry out cross-correlation calculation, and normalize the results, each group of normalized cross-correlation results corresponds to a position on the optical fiber; arrange the group of normalized cross-correlation results along the corresponding position of the optical fiber, so as to obtain the cross-correlation three-dimensional graph along each position of the optical fiber, and then project it on the two-dimensional plane, so as to obtain the original two-dimensional cross-correlation intensity graph ; The fourth step is to obtain the original two-dimensional cross-correlation intensity map from the third step. Press up with a specified step size Extraction size is The image patch is used as a reference patch, and then a size of [missing information] is placed centered on that reference patch. Block matching based on Euclidean distance is performed within the neighborhood of the reference block to obtain the number of blocks including the reference block itself. Similar image patches, The size is Column scanning of similar image blocks to convert The size is The vector, will The size is The vectors are concatenated into a single vector of size. matrix ,exist The above calculates the Euclidean distance between each row as a reference row and all other rows. The Euclidean distance between the reference row and the reference row The smallest includes the reference line. The rows form a group of size matrix ; Fifth step, take the size obtained in the fourth step and set it to... matrix Perform a separable two-dimensional Haar transform, that is, perform Haar transforms in the vertical and horizontal directions respectively. During the transformation process, apply the lifting Haar transform, that is, perform a weighted average only between rows or columns to obtain the transform coefficients. Step 6, Double Hard Threshold Shrinkage: In step 5, for a size of matrix Applying the lifting Haar transform to the two rows with the smallest Euclidean distance is equivalent to taking a weighted difference between these two rows. The coefficients of this weighted difference result are almost entirely derived from noise; these coefficients are all set to zero. For the remaining Haar transform coefficients, a hard threshold is applied to shrink the coefficients. The hard threshold parameter for coefficient shrinkage is... ; Step 7: Perform an inverse Haar transform on the Haar transform coefficients after the double hard thresholding shrinkage in step 6 to obtain a denoised two-dimensional matrix. Then, scan this two-dimensional matrix back to the original image blocks. Finally, weighted summaries of these image blocks are placed back into their original positions to obtain the preliminary denoised image. To preserve image details, an iterative strategy is adopted, with the number of iterations being [number missing]. ,go through The initial denoised image obtained from the second iteration is denoted as... ; used for the The noisy image in the nth iteration, through the nth iteration The denoised image obtained in the second iteration according to The calculation yielded, where , For regularization parameters; Step 8, the steps from step 7... The initial denoised image of the next iteration And the original noisy image obtained in step three Perform block matching and line matching operations synchronously; at a specified step size. exist Extract the size as The image patch is used as a reference patch, and then a size of [missing information] is placed centered on that reference patch. Block matching based on Euclidean distance is performed within the neighborhood of [a] block to obtain [a] block, including the reference block. Similar image patches, The size is Column scanning of similar image blocks to convert The size is The vector, will The size is The vectors are concatenated into a single vector of size. matrix ,exist The above calculates the Euclidean distance between each row as a reference row and all other rows. The row with the smallest Euclidean distance from the reference row, including the reference row. The rows form a group of size matrix Original noisy image Performing synchronous operations with the same parameters yields a value of [size]. matrix After block matching and line matching, two values ​​of size are obtained. matrix and ; Step 9, the two matrices of size obtained in step 8 and are each subjected to the same separable two-dimensional Haar transform, i.e. a Haar transform is performed in each of the vertical and horizontal directions, in which the lifting Haar transform is applied, i.e. only a weighted average is performed between rows or columns; wherein the transform coefficients of , are ; Step 10. Perform Wiener filtering according to formula (1): (1) Step 11: To improve noise reduction performance, Transformation coefficients Coefficients obtained by Wiener filtering Perform the Wiener filtering as shown in step 10 again, i.e., perform 2 iterations; perform the inverse Haar transform on the coefficients obtained after 2 iterations to obtain the denoised 2D matrix, then scan this 2D matrix back to the original image blocks, and finally weight and place these image blocks back into their original positions in the image to obtain the final denoised image. ; twelfth step, the denoised two-dimensional cross-correlation intensity map converted into three dimensions, i.e. a denoised three-dimensional cross-correlation intensity map corresponding to the temperature variation range corresponding to the temperature variation range corresponding to the temperature variation range corresponding to the temperature variation range corresponding to the temperature variation range correspond The peak position of the cross-correlation result of each position through the optical fiber is searched, and the temperature change range is obtained Spectral frequency shift distribution; Thirteenth step, After the second step to the twelfth step are performed on the Rayleigh scattering signals of the group of adjacent temperature values, the following spectral frequency shift distributions are obtained, respectively corresponding to temperature change intervals . After the spectral frequency shift distributions are added, the spectral frequency shift distribution corresponding to the temperature change is obtained.

Images