Model for improving predictive capability of optical fiber speckle sensor

Abstract

The invention discloses a model for improving the prediction capability of an optical fiber speckle sensor, relates to the technical field of speckle measurement and optical fiber sensing, and solves the technical problem of poor speckle calibration prediction capability of the optical fiber speckle sensor in the prior art. The method comprises the steps of obtaining a plurality of speckle patterns of an optical fiber sensor; performing feature extraction on each speckle pattern based on a feature extraction algorithm to obtain a texture feature vector; carrying out dimensionality reduction on the texture feature vector based on a data dimensionality reduction algorithm to obtain a low-dimensional feature vector; inputting the low-dimensional feature vector into a regression neural network for training, and marking feature extraction, data dimensionality reduction and the regression neural network as a speckle calibration model; and performing calibration prediction on the optical fiber sensor based on the speckle calibration model. According to the invention, dimension reduction is carried out on the texture feature vector of the speckle pattern through the data dimension reduction algorithm, the training difficulty of the speckle prediction model is reduced, the convergence speed of the speckle prediction model is accelerated, and the prediction capability of the optical fiber speckle sensor is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of speckle measurement and optical fiber sensing, and in particular is a model for improving the prediction capability of an optical fiber speckle sensor. Background Art

[0002] The calibration of fiber optic speckle sensors requires various algorithm models to implement. The important parameter that characterizes the prediction performance of fiber optic speckle sensors is the prediction accuracy within a certain error range, that is, the proportion of the number of unlearned samples in the model within a certain error range to the total number of prediction sets. Therefore, a high prediction ability means that the smaller the error range, the greater the prediction accuracy.

[0003] There are two existing speckle prediction schemes. The first one is to use the similarity between speckle patterns to calibrate the displacement of the fiber optic sensor. The similarity calculation selects the speckle pattern corresponding to the initial displacement as a reference, and the numerical value obtained by calculating the speckle pattern corresponding to the remaining displacements represents the similarity corresponding to the displacement. The closer the value is to 1, the closer the image is. The displacement is calibrated by a straight line fitted by the similarity. For uncalibrated points, the deviation is large and cannot be predicted well, resulting in weak prediction ability.

[0004] The second type is that the fiber optic speckle sensor uses a convolutional neural network to complete the calibration, using the displacement value to represent the label. The speckle pattern is divided into a training set and a test set according to a certain ratio. The classification function of the network is used to complete the calibration, and a one-to-one correspondence between the speckle pattern and the displacement is achieved. Then the test set can only correspond to the label of the training set. Therefore, it is impossible to predict the unlearned speckle pattern, resulting in poor prediction ability.

[0005] The present invention proposes a model for improving the prediction capability of a fiber optic speckle sensor to solve the above technical problems. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a model for improving the prediction ability of a fiber optic speckle sensor, which is used to solve the technical problem that the prior art has poor prediction ability for speckle calibration of a fiber optic speckle sensor.

[0007] To achieve the above object, a first aspect of the present invention provides a model for improving the prediction capability of a fiber optic speckle sensor, comprising:

[0008] S1: Acquire several speckle patterns of the optical fiber sensor;

[0009] S2: Extract features from each speckle pattern based on a feature extraction algorithm to obtain a texture feature vector;

[0010] S3: Reduce the dimension of the texture feature vector based on the data dimension reduction algorithm to obtain a low-dimensional feature vector;

[0011] S4: inputting the low-dimensional feature vector into the recurrent neural network for training, and labeling the feature extraction, data dimensionality reduction and recurrent neural network as a speckle calibration model;

[0012] S5: Calibration prediction of optical fiber sensor based on speckle calibration model.

[0013] Preferably, the extracting features of each speckle pattern based on a feature extraction algorithm comprises:

[0014] A1: extract the speckle pattern and grayscale it to obtain a speckle grayscale image;

[0015] A2: Set the calculation window size of the LBP operator;

[0016] A3: Obtain the grayscale value corresponding to each pixel in the speckle grayscale image;

[0017] A4: In a clockwise order, determine whether the grayscale value of the pixel point at the edge of the window is greater than the grayscale value of the pixel point at the center of the window; if yes, set the binary code s(x) to 1; if no, set the binary code s(x) to 0;

[0018] A5: According to the LBP operator Starting from the upper left corner of the speckle pattern, the LBP value corresponding to each calculation window is calculated row by row; where g i is the gray value of the pixel in the circular neighborhood with a radius of R, where R is a positive integer, g c is the gray value of the center pixel, x is g i Subtract g c The obtained difference, P is the number of pixels in the circular neighborhood of the central pixel; i is the arrangement order of the pixels in the circular neighborhood of the central pixel, i = 0, 1, 2, ..., P-1;

[0019] A6: Arrange the LBP values ​​in rows to obtain the texture feature vector of each speckle pattern.

[0020] The present invention improves the traditional LBP algorithm, directly uses the calculated LBP value as the value of the texture feature vector, and arranges them according to the moving direction of the calculation window. The texture information of the speckle patterns with different curvatures at different local positions is quite different. The LBP value calculated by the present invention can increase the difference between the feature vectors, highlighting the texture difference of different speckle patterns, which is beneficial to improving the prediction ability of the optical fiber speckle sensor.

[0021] Preferably, the dimension reduction of the texture feature vector based on the data dimension reduction algorithm includes:

[0022] B1: Extract the texture feature vector of the speckle pattern;

[0023] B2: Arrange the texture feature vectors row by row to form the first matrix;

[0024] B3: Subtract the average value of the corresponding row from each row of the first matrix to obtain the second matrix;

[0025] B4: By formula The covariance matrix C is calculated; where X is the first matrix, n is the number of rows of the first matrix, m is the number of columns of the first matrix, and X T is the transposed matrix of the first matrix;

[0026] B5: Solve the covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors;

[0027] B6: Determine the dimension reduction based on the eigenvalues ​​of the covariance matrix;

[0028] B7: Arrange the eigenvectors from large to small according to the corresponding eigenvalues ​​to obtain a eigenvector sequence;

[0029] B8: Select the first k eigenvectors from the eigenvector sequence to form a third matrix; where k is the dimension of dimensionality reduction;

[0030] B9: Multiply the first row of the third matrix by the first matrix to obtain a low-dimensional eigenvector.

[0031] The present invention reduces the dimension of the texture feature vector of the speckle image by using the PCA algorithm, reduces the difficulty of training the speckle prediction model, and accelerates the convergence speed of the speckle prediction model, which is beneficial to improving the prediction ability of the optical fiber speckle sensor.

[0032] Preferably, determining the dimension reduction dimension based on the eigenvalue of the covariance matrix includes:

[0033] C1: Calculate the cumulative contribution rate of each principal component based on the eigenvalue of the covariance matrix;

[0034] C2: The number of vectors corresponding to the principal components with cumulative contribution rates higher than 50% is selected as the dimension reduction dimension;

[0035] C3: Substitute the data after dimension reduction into the regression network for training to obtain the error result, and judge whether the error result reaches the expected result; if yes, it is successfully selected; if not, continue to select according to the cumulative contribution rate.

[0036] It should be noted that the eigenvalue of the covariance matrix can describe the amount of information contained in the direction of the corresponding eigenvector. The cumulative contribution rate refers to the ratio of the eigenvalue of each principal component to the sum of the eigenvalues ​​of all principal components. The cumulative contribution rate can reflect the contribution of each principal component to the overall data change, and can be used to indicate how much information retained in the low-dimensional space accounts for the total information in the high-dimensional space. The closer the cumulative contribution rate is to 1, the more information is retained in the low-dimensional space. Selecting an appropriate dimensionality reduction dimension can improve the accuracy of speckle prediction.

[0037] Preferably, the step of inputting the low-dimensional feature vector into a recurrent neural network for training comprises:

[0038] Extract low-dimensional feature vectors and integrate them into original data, and divide the original data into a training set and a test set; use the training set to train the recurrent neural network; use the test set to test the trained recurrent neural network, and adjust the recurrent neural network according to the test results; mark the feature extraction, data dimensionality reduction and recurrent neural network as a speckle calibration model; wherein the recurrent neural network includes a BP neural network model or an RBF neural network model and other types of recurrent neural networks.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention improves the traditional LBP algorithm, directly uses the calculated LBP value as the value of the texture feature vector, and arranges them according to the moving direction of the calculation window. The texture information of the speckle patterns with different curvatures at different local positions is quite different. The LBP value calculated by the present invention can increase the difference between the feature vectors, highlight the texture difference of different speckle patterns, and thus is beneficial to improving the prediction ability of the optical fiber speckle sensor. In addition, the PCA algorithm is used to reduce the dimension of the texture feature vector of the speckle pattern, which reduces the difficulty of training the speckle prediction model and accelerates the convergence speed of the speckle prediction model, and thus is beneficial to improving the prediction ability of the optical fiber speckle sensor. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is an overall flow chart of the model for improving the prediction capability of the optical fiber speckle sensor of the present invention;

[0043] Figure 2This is a flow chart of feature extraction of a speckle pattern in the present invention;

[0044] Figure 3 It is a schematic diagram of feature extraction of a speckle pattern in the present invention;

[0045] Figure 4 This is a schematic diagram of the system for training the speckle prediction model in the present invention. DETAILED DESCRIPTION

[0046] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] See also Figure 1-Figure 4 The first aspect of the present invention provides a model for improving the prediction capability of a fiber optic speckle sensor, including:

[0048] S1: Acquire several speckle patterns of the optical fiber sensor;

[0049] S2: Extract features from each speckle pattern based on a feature extraction algorithm to obtain a texture feature vector;

[0050] S3: Reduce the dimension of the texture feature vector based on the data dimension reduction algorithm to obtain a low-dimensional feature vector;

[0051] S4: inputting the low-dimensional feature vector into the recurrent neural network for training, and labeling the feature extraction, data dimensionality reduction and recurrent neural network as a speckle calibration model;

[0052] S5: Calibration prediction of optical fiber sensor based on speckle calibration model.

[0053] In this embodiment, feature extraction is performed on each speckle pattern based on a feature extraction algorithm, including:

[0054] A1: extract the speckle pattern and grayscale it to obtain a speckle grayscale image;

[0055] A2: Set the calculation window size of the LBP operator;

[0056] A3: Obtain the grayscale value corresponding to each pixel in the speckle grayscale image;

[0057] A4: In a clockwise order, determine whether the grayscale value of the pixel point at the edge of the window is greater than the grayscale value of the pixel point at the center of the window; if yes, set the binary code s(x) to 1; if no, set the binary code s(x) to 0;

[0058] A5: According to the LBP operator Starting from the upper left corner of the speckle pattern, the LBP value corresponding to each calculation window is calculated row by row; where g i is the gray value of the pixel in the circular neighborhood with a radius of R, where R is a positive integer, g c is the gray value of the center pixel, x is g i Subtract g c The obtained difference, P is the number of pixels in the circular neighborhood of the central pixel; i is the arrangement order of the pixels in the circular neighborhood of the central pixel, i = 0, 1, 2, ..., P-1;

[0059] A6: Arrange the LBP values ​​in rows to obtain the texture feature vector of each speckle pattern.

[0060] For example, the calculation window size of the LBP operator is set to 5×5, and the gray value g of the pixel at the center of the window is c =16, radius R=2, g0=17, g1=16, g2=16, g3=16, g4=16, g5=16, g6=17, g7=17, g8=18, g9=18, g 10 =17, g 11 =17, g 12 =18, g 13 =18, g 14 =17, g 15 = 17; start from the upper left corner of the speckle pattern and traverse the calculation window line by line to obtain The LBP values ​​are arranged in rows to obtain the texture feature vector of each speckle pattern.

[0061] The present invention improves the traditional LBP algorithm, directly uses the calculated LBP value as the value of the texture feature vector, and arranges them according to the moving direction of the calculation window. The texture information of the speckle patterns with different curvatures at different local positions is quite different. The LBP value calculated by the present invention can increase the difference between the feature vectors, highlighting the texture difference of different speckle patterns, which is beneficial to improving the prediction ability of the optical fiber speckle sensor.

[0062] In this embodiment, the dimension reduction of the texture feature vector is performed based on a data dimension reduction algorithm, including:

[0063] B1: Extract the texture feature vector of the speckle pattern;

[0064] B2: Arrange the texture feature vectors row by row to form the first matrix;

[0065] B3: Subtract the average value of the corresponding row from each row of the first matrix to obtain the second matrix;

[0066] B4: By formula The covariance matrix C is calculated; where X is the first matrix, n is the number of rows of the first matrix, m is the number of columns of the first matrix, and X T is the transposed matrix of the first matrix;

[0067] B5: Solve the covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors;

[0068] B6: Determine the dimension reduction based on the eigenvalues ​​of the covariance matrix;

[0069] B7: Arrange the eigenvectors from large to small according to the corresponding eigenvalues ​​to obtain a eigenvector sequence;

[0070] B8: Select the first k eigenvectors from the eigenvector sequence to form a third matrix; where k is the dimension of dimensionality reduction;

[0071] B9: Multiply the first row of the third matrix by the first matrix to obtain a low-dimensional eigenvector.

[0072] Exemplarily, the eigenvalues ​​of the covariance matrix C are set to λ1, λ2, λ3, λ4, λ5, and λ2>λ1>λ3>λ5>λ4, and the eigenvectors corresponding to the eigenvalues ​​are v1, v2, v3, v4, v5; the eigenvectors are arranged from large to small according to the size of the corresponding eigenvalues ​​to obtain the eigenvector sequence {v2, v1, v3, v5, v4}; the dimensionality reduction dimension k is set to 3, and the first 3 eigenvectors are selected from the eigenvector sequence to form a third matrix {v2, v1, v3}; the first row of the third matrix is ​​multiplied by the first matrix to obtain a low-dimensional eigenvector.

[0073] The present invention reduces the dimension of the texture feature vector of the speckle image by using the PCA algorithm, reduces the difficulty of training the speckle prediction model, and accelerates the convergence speed of the speckle prediction model, which is beneficial to improving the prediction ability of the optical fiber speckle sensor.

[0074] In this embodiment, determining the dimension reduction based on the eigenvalue of the covariance matrix includes:

[0075] C1: Calculate the cumulative contribution rate of each principal component based on the eigenvalue of the covariance matrix;

[0076] C2: The number of vectors corresponding to the principal components with cumulative contribution rates higher than 50% is selected as the dimension reduction dimension;

[0077] C3: Substitute the data after dimension reduction into the regression network for training to obtain the error result, and judge whether the error result reaches the expected result; if yes, it is successfully selected; if not, continue to select according to the cumulative contribution rate.

[0078] Exemplarily, the cumulative contribution rate of each principal component is calculated according to the eigenvalue of the covariance matrix, and the number of vectors corresponding to the principal components with larger cumulative contribution rates is set to 30 and marked as reduced dimensions; since the error result has reached the expected result, the selection is successful.

[0079] It should be noted that the eigenvalue of the covariance matrix can describe the amount of information contained in the direction of the corresponding eigenvector. The cumulative contribution rate refers to the ratio of the eigenvalue of each principal component to the sum of the eigenvalues ​​of all principal components. The cumulative contribution rate can reflect the contribution of each principal component to the overall data change, and can be used to indicate how much information retained in the low-dimensional space accounts for the total information in the high-dimensional space. The closer the cumulative contribution rate is to 1, the more information is retained in the low-dimensional space. Selecting an appropriate dimensionality reduction dimension can improve the accuracy of speckle prediction.

[0080] In this embodiment, the low-dimensional feature vector is input into a recurrent neural network for training, including:

[0081] Extract low-dimensional feature vectors and integrate them into original data, divide 50% of the original data into a training set and 50% into a test set; use the training set to train the recurrent neural network; use the test set to test the trained recurrent neural network, and adjust the recurrent neural network according to the test results; mark the feature extraction, data dimensionality reduction and recurrent neural network as a speckle calibration model; wherein the recurrent neural network includes a BP neural network model or an RBF neural network model and other types of recurrent neural networks.

[0082] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A model for improving the prediction capability of optical fiber speckle sensor, characterized in that: include: S1: Acquire several speckle patterns of the optical fiber sensor; S2: Extract features from each speckle pattern based on a feature extraction algorithm to obtain a texture feature vector; S3: Reduce the dimension of the texture feature vector based on the data dimension reduction algorithm to obtain a low-dimensional feature vector; S4: inputting the low-dimensional feature vector into the recurrent neural network for training, and labeling the feature extraction, data dimensionality reduction and recurrent neural network as a speckle calibration model; S5: Calibration prediction of optical fiber sensor based on speckle calibration model.

2. A model for improving the prediction capability of optical fiber speckle sensor according to claim 1, characterized in that: The feature extraction of each speckle pattern based on a feature extraction algorithm comprises: A1: extract the speckle pattern and grayscale it to obtain a speckle grayscale image; A2: Set the calculation window size of the LBP operator; A3: Obtain the grayscale value corresponding to each pixel in the speckle grayscale image; A4: In a clockwise order, determine whether the grayscale value of the pixel point at the edge of the window is greater than the grayscale value of the pixel point at the center of the window; if yes, set the binary code s(x) to 1; if no, set the binary code s(x) to 0; A5: According to the LBP operator Starting from the upper left corner of the speckle pattern, the LBP value corresponding to each calculation window is calculated row by row; where g i is the gray value of the pixel in the circular neighborhood with a radius of R, where R is a positive integer, g c is the gray value of the center pixel, x is g i Subtract g c The obtained difference, P is the number of pixels in the circular neighborhood of the central pixel; i is the arrangement order of the pixels in the circular neighborhood of the central pixel, i = 0, 1, 2, ..., P-1; A6: Arrange the LBP values ​​in rows to obtain the texture feature vector of each speckle pattern.

3. A model for improving the prediction capability of optical fiber speckle sensor according to claim 1, characterized in that: The method of reducing the dimension of the texture feature vector based on a data dimension reduction algorithm includes: B1: Extract the texture feature vector of the speckle pattern; B2: Arrange the texture feature vectors row by row to form the first matrix; B3: Subtract the average value of the corresponding row from each row of the first matrix to obtain the second matrix; B4: By formula The covariance matrix C is calculated; where X is the first matrix, n is the number of rows of the first matrix, m is the number of columns of the first matrix, and X T is the transposed matrix of the first matrix; B5: Solve the covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors; B6: Determine the dimension reduction based on the eigenvalues ​​of the covariance matrix; B7: Arrange the eigenvectors from large to small according to the corresponding eigenvalues ​​to obtain a eigenvector sequence; B8: Select the first k eigenvectors from the eigenvector sequence to form a third matrix; where k is the dimension of dimensionality reduction; B9: Multiply the first row of the third matrix by the first matrix to obtain a low-dimensional eigenvector.

4. A model for improving the prediction capability of optical fiber speckle sensor according to claim 3, characterized in that: The step of determining the dimension reduction based on the eigenvalue of the covariance matrix includes: C1: Calculate the cumulative contribution rate of each principal component based on the eigenvalue of the covariance matrix; C2: The number of vectors corresponding to the principal components with cumulative contribution rates higher than 50% is selected as the dimension reduction dimension; C3: Substitute the data after dimension reduction into the regression network for training to obtain the error result, and judge whether the error result reaches the expected result; if yes, it is successfully selected; if not, continue to select according to the cumulative contribution rate.

5. A model for improving the prediction capability of optical fiber speckle sensor according to claim 1, characterized in that: The step of inputting the low-dimensional feature vector into the recurrent neural network for training includes: Extract low-dimensional feature vectors and integrate them into original data, and divide the original data into a training set and a test set; use the training set to train the recurrent neural network; use the test set to test the trained recurrent neural network, and adjust the recurrent neural network according to the test results; mark the feature extraction, data dimensionality reduction and recurrent neural network as a speckle calibration model; wherein the recurrent neural network includes a BP neural network model or a RBF neural network model.

Images