Optical fiber speckle sensing method and device based on mode decomposition

Abstract

The invention discloses an optical fiber speckle sensing method based on mode decomposition. Performing feature extraction on the speckle pattern of the optical fiber speckle sensor, inputting the extracted features into a pre-trained machine learning model, and outputting sensing parameters corresponding to the speckle pattern by the machine learning model; the feature extraction is carried out by using the following method: firstly, carrying out mode decomposition on a speckle pattern to obtain amplitude information and phase information of each intrinsic mode contained in the speckle pattern; and then constructing the features according to the obtained amplitude information and / or phase information of each intrinsic mode. The invention further discloses an optical fiber speckle sensing device based on mode decomposition. According to the invention, the rapid and accurate estimation of various sensing parameters according to the amplitude and / or phase information of each mode in the optical fiber is proposed for the first time, the high-precision and multi-parameter optical fiber sensing measurement is realized, the sensing precision and stability are improved, the equipment configuration cost and the dependence on a complex model are reduced, and the application range of optical fiber sensing is expanded.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber sensing, and particularly relates to an optical fiber speckle sensing method and device. Background Art

[0002] With the rapid development of information technology and sensing technology, multi-dimensional parameter sensing, as a bridge connecting the physical world and the digital world, significantly improves the perception ability of intelligent devices by simultaneously capturing and processing multi-dimensional environmental information. With the integration of microelectronics technology, Internet of Things, big data and artificial intelligence, multi-dimensional sensing technology shows great application potential and research value in the fields of intelligent wearable devices, human-computer interaction, industrial intelligent robots and healthcare. Among them, optical fiber sensing has obvious advantages such as light installation, low cost, anti-electromagnetic interference, strong concealment, chemical corrosion resistance, stable operation in harsh environments and easy large-scale multiplexing compared with other methods. It is widely used in the fields of security monitoring, environmental monitoring, medical equipment and industrial automation.

[0003] Existing solutions can simultaneously measure multiple physical quantities such as temperature, strain and curvature through technologies such as grating technology, optical fiber structure design, hybrid fusion splicing of different types of optical fibers and distributed sensing, meeting the requirements for multi-dimensional parameter sensing in complex application scenarios. Among them, the method based on inter-mode interference utilizes the linear relationship between measurement parameters and has the advantages of light weight, high sensitivity and good stability, but the limited linear interval restricts the measurement range of the sensor. Optical frequency domain reflectometry and Brillouin optical time domain analysis are also applied to distributed optical fiber sensing research to measure the bending of optical fibers, but the larger measurement dynamic range is limited by the mode splitting effect in the optical fiber and the wavelength shift resolution.

[0004] Existing fiber optic speckle sensors typically adopt structures such as multimode fibers, ring-core fibers, and few-mode-multimode hybrids, supporting the coupling of multiple different spatial modes. Each mode wave propagating in these fibers is easily affected by the surrounding environment, and finally forms a seemingly random but still deterministic speckle pattern at the output end, which lays the foundation for fiber optic speckle sensing. Such sensors usually use machine learning models to directly analyze and extract the speckle image features under different environmental conditions, so as to invert specific sensing parameters according to the fiber optic speckles. In addition, there is also a parameter sensing scheme based on the zero-mean normalized cross-correlation coefficient of the calculated fiber optic speckle pattern. In short, the fiber optic sensors based on speckles have a simpler structural configuration, do not require complex preparation processes, and can complete data acquisition only relying on an ordinary CCD camera, greatly reducing the dependence on expensive high-resolution measuring instruments. However, the main problem faced by the existing research on machine learning-assisted fiber optic speckle sensing is the low accuracy. In the classification task, the sensor is limited by the number of discrete label classifications, resulting in poor performance when dealing with continuously changing physical parameters; while in the regression task, the sensing accuracy highly depends on complex deep models, which means consuming higher training costs and more computing resources, and at the same time reducing the interpretability of the scheme.( Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the problems existing in the existing fiber optic speckle sensing technology, such as low accuracy when measuring continuously changing physical parameters, limited measurement range, complex equipment configuration, and high dependence on deep models. A fiber optic speckle sensing method based on mode decomposition is provided, which first proposes to realize the rapid and accurate estimation of various sensing parameters based on the amplitude and / or phase information of each mode in the fiber. While realizing high-precision and multi-parameter fiber optic sensing measurement, it improves the sensing accuracy and stability, reduces the equipment configuration cost and the dependence on complex models, and expands the application range of fiber optic sensing.

[0006] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0007] A fiber optic speckle sensing method based on mode decomposition extracts features from the speckle pattern of a fiber optic speckle sensor, and inputs the extracted features into a pre-trained machine learning model, and the machine learning model outputs the sensing parameters corresponding to the speckle pattern; the following method is used for the feature extraction: first, the speckle pattern is decomposed into modes to obtain the amplitude information and phase information of each eigenmode included in the speckle pattern; then, the features are constructed based on the amplitude information and / or phase information of each obtained eigenmode.

[0008] Preferably, the features are constructed based on the amplitude information of each obtained eigenmode.

[0009] Preferably, the fiber optic speckle sensor uses a few-mode fiber (FMF) as a sensing element.

[0010] Further, a photon lantern for coupling a detection optical signal from a single-mode fiber into the FMF is connected to the front end of the FMF.

[0011] Preferably, the method for mode decomposition is as follows: First, utilize the linear relationship between the modal coefficient vector and the optical field intensity vector to fit the eigenmode matrix related to the actual noise characteristics of the fiber speckle sensor; during actual mode decomposition, first calculate the initial modal coefficients of the input speckle pattern according to the fitted linear mapping relationship; then optimize through an optimization algorithm to obtain the accurate value of mode decomposition.

[0012] Based on the same inventive concept, the following technical solutions can also be obtained:

[0013] A fiber speckle sensing device based on mode decomposition, comprising a fiber speckle sensor, a feature extraction module for extracting features from the speckle pattern of the fiber speckle sensor, and a pre-trained machine learning model that takes the extracted features as input and the corresponding sensing parameters as output; the feature extraction module uses the following method for feature extraction: First, perform mode decomposition on the speckle pattern to obtain the amplitude information and phase information of each eigenmode included in the speckle pattern; then construct the features with the obtained amplitude information and / or phase information of each eigenmode.

[0014] Preferably, the feature extraction module constructs the features with the obtained amplitude information of each eigenmode.

[0015] Preferably, the fiber speckle sensor uses a few-mode fiber (FMF) as a sensing element.

[0016] Further, a photon lantern for coupling a detection optical signal from a single-mode fiber into the FMF is connected to the front end of the FMF.

[0017] Preferably, the method for mode decomposition is as follows: First, utilize the linear relationship between the modal coefficient vector and the optical field intensity vector to fit the eigenmode matrix related to the actual noise characteristics of the fiber speckle sensor; during actual mode decomposition, first calculate the initial modal coefficients of the input speckle pattern according to the fitted linear mapping relationship; then optimize through an optimization algorithm to obtain the accurate value of mode decomposition.

[0018] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0019] For the first time, based on the continuity and predictability of the evolution of optical fiber modes with respect to changes in external physical parameters, this invention uses optical fiber mode information to characterize external physical parameters such as optical fiber curvature, angle, and position. It breaks through the limitations of the traditional linear response range, achieving a larger measurement range and higher resolution, enabling optical fiber sensors to adapt to a wider range of application scenarios. Compared with existing optical fiber speckle sensing technologies, this invention features optical fiber mode information, significantly reducing the feature dimension. Only a small amount of mode field data is required for training and estimation, significantly reducing the required training time and memory resources. At the same time, it reduces the dependence on complex deep models, lowering the training cost and consumption of computing resources. In addition, the continuous change of modal data with the sensing parameter is easier to understand and analyze by humans and machine learning models, having better interpretability. The mode decomposition method proposed in this invention has good noise robustness and can maintain a high decomposition accuracy in an environment with obvious noise, thereby further improving the stability of optical fiber sensors working in actual complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 FIG. is a schematic structural principle diagram of an optical fiber speckle sensing device based on mode decomposition according to this invention;

[0021] Figure 2 FIG. is an optimization process of the correlation between the reconstructed image and the original image in the mode decomposition algorithm of this invention;

[0022] Figure 3 FIG. is the correlation result of decomposing 1000 intensity sample images by the mode decomposition algorithm of this invention;

[0023] Figure 4 FIG. is an example of several typical mode decomposition results;

[0024] Figure 5 FIG. is a schematic flow diagram of an optical fiber speckle sensing method based on mode decomposition according to this invention;

[0025] Figure 6 FIG. is the evolution curve of the modal weights in a six-mode optical fiber with respect to the optical fiber curvature;

[0026] Figure 7 FIG. is the optical fiber curvature distribution estimated based on the sensing scheme of this invention and the fitting line;

[0027] Figure 8 FIG. is the importance of different modes for sensing the curvature, bending position, bending angle, and rotation angle of a FMF;

[0028] Figure 9 FIG. is the result of reconstructing different tactile patterns based on the sensing scheme of this invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] Aiming at the deficiencies of existing fiber optic speckle sensing technology, the solution idea of the present invention is based on the continuity and predictability of the evolution of fiber optic modes with the change of external physical parameters. By means of fiber optic mode information, external physical parameters such as fiber curvature, angle, and position are characterized. While realizing high-precision and multi-parameter fiber optic sensing measurement, the sensing accuracy and stability are improved, the equipment configuration cost and the dependence on complex models are reduced, and the application range of fiber optic sensing is expanded.

[0030] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0031] A fiber optic speckle sensing method based on mode decomposition, which extracts features from the speckle pattern of a fiber optic speckle sensor and inputs the extracted features into a pre-trained machine learning model, and the machine learning model outputs the sensing parameters corresponding to the speckle pattern; the following method is used for the feature extraction: first, the speckle pattern is decomposed into modes to obtain the amplitude information and phase information of each eigenmode contained in the speckle pattern; then, the features are constructed with the obtained amplitude information and / or phase information of each eigenmode.

[0032] A fiber optic speckle sensing device based on mode decomposition, including a fiber optic speckle sensor, a feature extraction module for extracting features from the speckle pattern of the fiber optic speckle sensor, and a pre-trained machine learning model that takes the extracted features as input and the corresponding sensing parameters as output; the feature extraction module uses the following method for the feature extraction: first, the speckle pattern is decomposed into modes to obtain the amplitude information and phase information of each eigenmode contained in the speckle pattern; then, the features are constructed with the obtained amplitude information and / or phase information of each eigenmode.

[0033] The present invention can construct the features based on the amplitude information and / or phase information of each eigenmode. For example, all amplitude information and all phase information are used, amplitude information alone or phase information alone, or a combination of partial amplitude information and partial phase information; through a large number of studies by the inventor, it is found that using the amplitude information of each eigenmode alone to construct the features, that is, using the modal weights as the input features of the model, can obtain basically equivalent sensing accuracy while reducing the feature dimension by half, and the stability and noise robustness of sensing are better; therefore, the present invention preferably constructs the features with the obtained amplitude information of each eigenmode.

[0034] The present invention can be directly implemented based on various existing fiber optic speckle sensors. However, most of the existing fiber optic speckle sensors use multimode fibers with hundreds or thousands of eigenmodes as sensing elements. On the one hand, it is difficult to achieve mode decomposition. On the other hand, the feature dimensions constructed from their modal information are still relatively high, requiring machine learning models with high complexity, and higher costs and more time are needed for training and estimation. The inventor's research found that when using few-mode fiber (FMF) with only a few eigenmodes as the sensing element, compared with traditional multimode fibers, it is simpler to achieve mode decomposition. While significantly reducing the feature dimensions, higher sensing accuracy can be obtained, and the sensing stability and noise robustness are better. Therefore, preferably, the fiber optic speckle sensor uses few-mode fiber (FMF) as the sensing element.

[0035] For the convenience of public understanding, the technical solution of the present invention will be described in detail below through a specific embodiment in conjunction with the accompanying drawings:

[0036] The basic structure of the fiber optic speckle sensing device in this embodiment is as Figure 1 shown. The narrow linewidth laser emits a laser with a wavelength of 1550 nm, which enters the variable optical attenuator and polarization controller through the fiber optic connector. The variable optical attenuator is used to adjust the optical power of the system, and the polarization controller can select the polarization components of the light beam, thereby effectively simplifying the polarization state and avoiding errors caused by polarization crosstalk when collecting the near-field image of the light beam. The probe light is coupled into the FMF from the single-mode fiber with the help of a six-mode photonic lantern. The FMF used in this embodiment is a standard six-mode commercial step-index fiber with a length of 10 m. At the working wavelength, the FMF with a core diameter of 18.5 μm and a cladding diameter of 125 μm supports LP 01 、LP 02 、LP 11e 、LP 11o 、LP 21e and LP 21o six linearly polarized modes. The light output from the end face of the FMF passes through the lens and is emitted into space. The CCD camera is used to collect the beam intensity pattern (i.e., the speckle pattern) of the near field of the FMF, and the pixel pitch is 20 μm. The obtained beam intensity pattern undergoes mode decomposition to obtain the amplitude information and phase information of each eigenmode. The amplitude information (modal weight) is input into a pre-trained machine learning model (the LightGBM model is used in this embodiment), and the machine learning model can output sensing parameters such as the curvature, bending angle, rotation angle, and bending position of the optical fiber.

[0037] The FMF can transmit optical signals of multiple different spatial modes, and the propagation field distribution at the end face of the optical fiber is composed of the linear superposition of each eigenmode. For any position (x, y) on the end face of the optical fiber, the actually measured optical field intensity can be expressed as:

[0038]

[0039] where E is the electric field distribution at the end face of the optical fiber, N is the number of eigenmodes, and ρ j and θ j are the normalized amplitude and relative phase of each mode respectively, and satisfy ρ j ∈[0, 1], θ j ∈[-π, π]. Under the approximation conditions based on weakly guiding fibers and step-index profiles, Ψ j can be described as the normalized field distribution of the j-th order linearly polarized (LP) mode, which is related to the numerical aperture, transverse characteristic size, and operating wavelength of the optical fiber.

[0040] Mode decomposition is to obtain 2N - 1 modal coefficients including amplitude and phase (θ 1 = 0) from the measured beam profile image. Existing mode decomposition methods can be roughly divided into two categories. The first category is experimental measurement methods including frequency-domain cross-correlation imaging [1] , spatial and spectral resolution imaging [2] and wavefront measurement techniques [3] etc. These methods can usually provide accurate and intuitive measurement results, but the high equipment cost, complex operation process, and strict experimental conditions limit their wide deployment and application. The second category is numerical analysis methods, including the classical stochastic parallel gradient descent method (SPGD) [4] and advanced inverse matrix solution [5] etc. However, the former is prone to being affected by the initial value for optimizing non-convex problems and falling into local optimal solutions; the latter has the disadvantage of being highly sensitive to noise, and the decomposition accuracy will decrease significantly with the increase in the number of modes and noise level. To solve the above problems, the present invention further proposes a mode decomposition method as follows: First, utilize the linear relationship between the modal coefficient vector and the optical field intensity vector to fit the eigenmode matrix related to the actual noise characteristics of the optical fiber speckle sensor; during actual mode decomposition, first calculate the initial modal coefficients of the input speckle pattern according to the fitted linear mapping relationship; then optimize through an optimization algorithm to obtain the accurate value of mode decomposition.

[0041] Next, still taking the mode decomposition of this embodiment as an example, the mode decomposition method proposed by the present invention will be further described in detail:

[0042] Divide the mode decomposition problem into a linear part and a non-linear part. The linear part is to write the optical field intensity I in the form of a linear equation system:

[0043] HX = I

[0044] where

[0045]

[0046] H is the eigenmode matrix composed of each eigenmode field in the FMF, with a dimension of M 2 ×N(N + 1) / 2, Ψ n m represents the m-th element of the n-th eigenmode field. M×M is the pixel size of the intensity image captured by the CCD or CMOS camera. X is the amplitude-phase vector containing modal coefficient information, with a dimension of N(N + 1) / 2×1. Therefore, each near-field intensity image of the FMF can be expressed by a linear equation as the product of the eigenmode matrix H and the amplitude-phase vector X.

[0047] In the preparation stage, according to the equation fit the eigenmode matrix H related to the actual noise characteristics z , where I z is the intensity matrix containing actual noise, and each column of it is the captured FMF beam profile image. X is the modal coefficient matrix obtained by performing mode decomposition on the K captured images through the classical SPGD algorithm. () -1 represents the Moore - Penrose pseudoinverse. Here, H z contains the mapping relationship from the actual intensity image to the true modal coefficients, which is usually related to the noise characteristics of the experimental system, especially the noise of the CCD camera. It should be emphasized that although this process requires collecting a large amount of intensity data for decomposition using the SPGD algorithm and will take some time, it can be completed before the formal mode decomposition and does not need to be repeated.

[0048] In the mode decomposition stage, first calculate the initial modal coefficients according to X = (H z ) -1 I, where I is the actually collected light field intensity data for decomposition. The calculation results include N mode amplitudes and N - 1 phases:

[0049]

[0050] where sgn is the sign function, and without loss of generality, it can be assumed that θ 2 ∈[0, π]. In addition, it is necessary to normalize the image data before decomposition, which can improve the decomposition speed and ensure the stability of the algorithm at the same time.

[0051] Then, based on this initial value, through the BFGS quasi - Newton optimization algorithm, combined with the Wolfe update step - size technique, further quickly optimize to obtain the accurate value of the mode decomposition, and achieve the accurate characterization of each mode in the few - mode fiber. The objective function for algorithm optimization is the correlation between the measured image and the reconstructed image of the beam (J ∈ [0, 1]):

[0052]

[0053] where (j = me or re), represents the measured image or the reconstructed image minus their respective means. J is also used to evaluate the accuracy of mode decomposition. The larger J is, the smaller the error between the measured image and the reconstructed image, and the higher the decomposition accuracy.

[0054] The performance of the above mode decomposition method is verified by simulation below. Based on the beam propagation principle in FMF, 30,000 beam profile images are randomly generated for decomposition and fitting the eigenmode matrix, and an additional 1,000 beam profile images are generated for testing. Among them, the signal-to-noise ratio of the intensity image is 24 dB, and the normalized image size is 256×256. In the mode decomposition stage, according to the fitted eigenmode matrix and the linear relationship, the initial modal coefficients are calculated using the measured intensity data, and then the BFGS algorithm is used for rapid optimization to obtain the final mode amplitudes and relative phases. The BFGS algorithm combines the gradient vector at the current position, the approximate second derivative matrix, and a more appropriate update step size in each iteration to calculate the new position, so it is fast. Usually, an accurate solution can be obtained after dozens of updates, as Figure 2 shown. Decompose 1,000 six-mode fiber beam images with noise. The average correlation between the original image and the reconstructed image is 0.998, as Figure 3 shown. The part with a correlation greater than 0.998 accounts for 79%, and the correlation of almost all reconstructed images is not less than 0.99. Some typical results are as Figure 4 shown, including the original beam profile image, the reconstructed image, the reconstruction difference, and the correlation. The highest correlation is 0.999. The difference in the reconstructed image comes from the background noise added to the original image. Therefore, the high similarity between the original intensity pattern and the reconstructed intensity pattern shown in the figure verifies that the mode decomposition method in the solution of the present invention has good noise robustness and can maintain a high decomposition accuracy in an obvious noise environment.

[0055] The flow of the fiber optic sensing method proposed by the present invention is as Figure 5 shown. Under different external parameter conditions, such as different fiber curvatures, bending positions, bending angles, and rotation angles, etc., the modal weight data corresponding to the state are obtained through the mode decomposition algorithm as training samples.

[0056] The evolution curves of the modal weights in the six-mode fiber with the fiber curvature in the range of 0 - 3.83 m -1 are as Figure 6 shown. It can be seen that there are differences in the correlations between different modes and the curvature. For example, modes LP 01 , LP 02 and LP 21oIt has a strong correlation with the curvature change, while the correlations of other modes are weak. This difference is reflected in the machine learning model as the modes with stronger correlations have a greater information gain and more significant contributions to the sensing results.

[0057] The LightGBM algorithm is used to construct a machine learning model. The collected modal weight data is used as input features, and the corresponding external sensing parameters are used as output values to train the machine learning model. The LightGBM algorithm has the advantages of fast training speed, low memory occupancy, and high accuracy, and can effectively learn the complex non-linear relationship between modal weights and external parameters. The mean square error (MSE) is used as the loss function for model training, where N is the number of samples, y i is the true value of the sensing parameter, and

[0058]

[0059] Using the trained machine learning model, the sensing parameters such as the curvature, bending angle, rotation angle, and bending position of the optical fiber can be estimated based on the modal weights obtained from real-time measurement. Taking curvature sensing as an example, the curvature sensing effect of the model is tested on the untrained FMF modal weight data set, and the results are as Figure 7 shown. The R2 of the estimated data is 0.9987, and the fitting line is y = 0.9929x + 0.023, showing a strong linear relationship between the curvature estimated value and the true value. This indicates that the trained LightGBM model can accurately estimate the optical fiber curvature within the sensing range of 0 - 3.83 m-1 based on the modal weight information obtained from the optical fiber mode decomposition, meeting the requirements of high-precision sensing. The high resolution of this curvature measurement scheme mainly benefits from the fact that the optical fiber modes change continuously and predictably with the external curvature.

[0060] Different external physical parameters such as the curvature, bending position, bending angle, and rotation angle of the optical fiber have different effects on the refractive index distribution inside the optical fiber, resulting in different responses of the modal weights. This phenomenon is an important basis for multi-dimensional parameter sensing of FMF. In the machine learning model, different types of modes can be regarded as features in different dimensions, and the feature importance in the LightGBM algorithm can be used to analyze the relationship between different modes and different sensing quantities. Figure 8 Shows the importance of different modal weights in the model for FMF curvature, bending position, bending angle, and rotation angle sensing. The higher the importance of the mode, the greater the information gain it has for the final sensing, and at the same time, it reflects the good correlation between the change of modal weight and the external sensing parameter. The spatial distributions of different modes in the optical fiber are different. For example, in curvature sensing, LP 01 、LP 02 and LP 21oThe importance of the mode is relatively high, which is related to the uneven radial refractive index distribution caused by fiber bending. In rotational angle sensing, when the optical fiber rotates around the axis, the change in refractive index may be the reason why the LP 01 、LP 11o and LP 21e modes are more important. Therefore, for specific sensing parameters, methods such as principal component analysis (PCA) and linear discriminant analysis (LDA) can be used for feature dimensionality reduction to further reduce the complexity of the machine learning model and reduce the required computing and memory resources.

[0061] The proposed solution of the present invention can measure different sensing parameters simultaneously. The bending position and angle change of the optical fiber will result in different modal responses, and the two are coupled to form an interference superposition speckle pattern on the output end face of the optical fiber. Changing the bending angle will significantly affect the modal weights, and the linear correlation between the bending position and the change in modal weights is stronger. Using a machine learning model can estimate and decouple the bending position and angle information from the modal weight data after mode decomposition, so as to realize multi-parameter sensing of the optical fiber.

[0062] The sensing solution proposed by the present invention can realize an intelligent tactile sensor, which can real-time sense the pressure distribution on a 30mm×30mm silicone sensing plane and reconstruct any two-dimensional pattern of 10×10 pixels with high resolution. As Figure 9 shown are the real images of different tactile patterns and their reconstruction results. All four patterns have been effectively reconstructed, which shows the ability of the present invention in two-dimensional tactile sensing and also provides a reference for the development of low-cost intelligent tactile sensors.

[0063] References

[0064] [1]Demas J,Ramachandran S.Sub-second mode measurement of fibers usingC^2imaging[J].Optics Express,2014,22(19):23043.

[0065] [2]Nguyen D M,Blin S,Nguyen T N,et al.Modal decomposition techniquefor multimode fibers[J].Applied Optics,2012,51(4):450.

[0066] [3] Paurisse M, Lévèque L, Hanna M, et al. Complete measurement of fibermodal content by wavefront analysis[J]. Optics Express, 2012, 20(4): 4074.

[0067] [4] Lü H, Zhou P, Wang X, et al. Fast and accurate modal decomposition ofmultimode fiber based on stochastic parallel gradient descent algorithm[J]. Applied Optics, 2013, 52(12): 2905.

[0068] [5] Manuylovich E, Donodin A, Turitsyn S. Intensity-only-measurement modedecomposition in few-mode fibers[J]. Optics Express, 2021, 29(22): 36769。

Claims

1. A fiber optic speckle sensing method based on mode decomposition, which extracts features from a speckle pattern of a fiber optic speckle sensor, and inputs the extracted features into a pre-trained machine learning model, which outputs a sensing parameter corresponding to the speckle pattern; characterized in that: The feature extraction is performed using the following method: firstly, the speckle pattern is pattern decomposed to obtain the amplitude information and phase information of each eigenmode contained in the speckle pattern; then, the feature is constructed with the obtained amplitude information and / or phase information of each eigenmode.

2. The optical fiber speckle sensing method based on mode decomposition as claimed in claim 1, characterized in that: The feature is constructed with the obtained amplitude information of each eigenmode.

3. The optical fiber speckle sensing method based on mode decomposition as claimed in claim 1, characterized in that: The optical fiber speckle sensor uses a few-mode optical fiber FMF as a sensing element.

4. The optical fiber speckle sensing method based on mode decomposition as claimed in claim 3, characterized in that: A photon lantern is connected to the front end of the FMF for coupling the detection light signal from the single-mode optical fiber into the FMF.

5. The optical fiber speckle sensing method based on mode decomposition as claimed in claim 1, characterized in that: The method for mode decomposition is specifically as follows: the linear relationship between the modal coefficient vector and the light field intensity vector is used in advance to fit the intrinsic mode matrix of the optical fiber speckle sensor related to the actual noise characteristics; when performing actual mode decomposition, the initial modal coefficients of the input speckle pattern are first calculated according to the fitted linear mapping relationship; and then the optimization algorithm is used to optimize to obtain the accurate value of the mode decomposition.

6. A fiber optic speckle sensing device based on mode decomposition, comprising a fiber optic speckle sensor, a feature extraction module for extracting features from a speckle pattern of the fiber optic speckle sensor, and a pre-trained machine learning model with the extracted features as input and the corresponding sensing parameters as output; characterized in that: The feature extraction module uses the following method to extract the features: firstly, the speckle pattern is pattern decomposed to obtain the amplitude information and phase information of each eigenmode contained in the speckle pattern; then, the feature is constructed with the obtained amplitude information and / or phase information of each eigenmode.

7. The optical fiber speckle sensing device based on mode decomposition as claimed in claim 6, characterized in that: The feature extraction module constructs the feature with the obtained amplitude information of each eigenmode.

8. The optical fiber speckle sensing device based on mode decomposition as claimed in claim 6, characterized in that: The optical fiber speckle sensor uses a few-mode optical fiber FMF as a sensing element.

9. The optical fiber speckle sensing device based on mode decomposition as claimed in claim 8, characterized in that: A photon lantern is connected to the front end of the FMF for coupling the detection light signal from the single-mode optical fiber into the FMF.

10. The optical fiber speckle sensing device based on mode decomposition as claimed in claim 6, characterized in that: The method for mode decomposition is specifically as follows: the linear relationship between the modal coefficient vector and the light field intensity vector is used in advance to fit the intrinsic mode matrix of the optical fiber speckle sensor related to the actual noise characteristics; when performing actual mode decomposition, the initial modal coefficients of the input speckle pattern are first calculated according to the fitted linear mapping relationship; and then the optimization algorithm is used to optimize to obtain the accurate value of the mode decomposition.

Images