An optical fiber shape sensor error correction method, program, device and storage medium based on adaptive filtering

Abstract

The present application belongs to the technical field of optical fiber shape sensing, and particularly relates to an optical fiber shape sensing error correction method based on adaptive filtering, a program, a device and a storage medium. The present application adopts a reconstructed error conduction model to drive adaptive filtering. In view of the problem that errors are conducted from strain measurement to curvature and torsion in optical fiber shape reconstruction and accumulated and amplified in the integration process, an error amplification risk index based on the error conduction relationship is introduced, a spatial smoothing window length is determined point by point, and range constraints and stability constraints are applied to the window, so that the filtering process is based on the conduction of noise and the characteristics of the curve itself. The present application can be used for spatial domain smoothing processing of the arc length sampling sequence in the optical fiber measurement system, can be applied to the accuracy improvement of key information such as curvature and torsion, and finally improves the accuracy of the three-dimensional shape reconstruction result.

Description

Technical Field

[0001] This invention belongs to the field of optical fiber shape sensing technology, specifically relating to an error correction method, program, device, and storage medium for optical fiber shape sensing based on adaptive filtering. Background Technology

[0002] With the development of fiber optic sensing technology, reconstructing the shape of the fiber in three-dimensional space by calculating curvature and deflection information based on strain information obtained along the fiber has become an important research direction. However, the quality of the acquired data itself has a significant impact on the reconstruction accuracy during fiber reconstruction, necessitating data filtering. Existing methods typically employ median filtering, moving average, low-pass filtering, or fixed-window smoothing to suppress high-frequency noise and improve the smoothness of the reconstruction results. Their main limitation lies in the fact that the degree of curvature and deflection variation varies at different locations along the arc length, and noise levels may also vary with location. Furthermore, these methods generally use fixed windows or fixed parameters, making them unable to adapt to the differences in error sensitivity across different sections.

[0003] In addition, some studies have proposed improving the accuracy of distributed strain measurement by changing the sliding window length. However, this method adjusts the window length based on the observation of signal strength and the comparison of empirical thresholds, without taking into account the error propagation law in the shape reconstruction link. Therefore, it is difficult to suppress the amplification mechanism of torsion or direction-related errors.

[0004] In the spatial curve reconstruction process driven by curvature and deflection data, strain measurement errors are transmitted as curvature and deflection errors, which further cause tangent vector deviations and ultimately accumulate as reconstruction errors during integration along the arc length. Since the deflection error has an inversely proportional amplification effect with curvature, the sensitivity to noise is not uniform across different segments along the arc length, thus requiring an adaptive filtering strategy that can vary with the arc length position. Summary of the Invention

[0005] The purpose of this invention is to address the reconstruction instability caused by the error propagation link during fiber shape sensing and spatial curve reconstruction, and to solve the problem of filter strategy mismatch caused by inconsistent error sensitivity in different sections in existing methods. This invention provides a fiber shape sensing error correction method, program, device, and storage medium based on adaptive filtering. This invention enables the filter intensity to be adaptively adjusted according to the curve geometry, achieving the optimal comprehensive filtering effect for key data such as curvature and torsion.

[0006] A method for correcting optical fiber shape sensing errors based on adaptive filtering includes the following steps:

[0007] Historical data from the sensor is acquired, including strain data at each sampling point along the fiber arc length; the noise intensity of the sensor is estimated, noise is added to the strain data at each sampling point, and a fixed window length filter is applied to the strain data after adding noise to obtain the filtered strain data, thereby determining the curvature and bending direction angle at the sampling point;

[0008] By adjusting the window length, the optimal window length suitable for filtering strain data after adding noise at each sampling point can be determined.

[0009] Indicators reflecting the risk of error amplification are constructed, including the risk of curvature error amplification and the risk of bending direction angle error amplification. The positional features and boundary distance features of each sampling point are introduced, and combined with the risk of curvature error amplification and the risk of bending direction angle error amplification, a feature vector of each sampling point is constructed.

[0010] A training set is constructed based on the feature vectors of each sampling point and the optimal window length. The prediction model is trained using the training set so that the trained prediction model can output the corresponding optimal window length based on the input feature vector.

[0011] When the sensor is used for measurement, it acquires the real-time strain sequence along the arc length of the optical fiber, and then acquires the curvature and bending direction angle at each sampling point. It calculates the curvature error amplification risk, bending direction angle error amplification risk, position features and boundary distance to construct a feature vector. The feature vector is then input into the trained prediction model to obtain the optimal window length. Based on the optimal window length, the real-time strain sequence is filtered to achieve error correction.

[0012] Furthermore, the acquisition of historical data from the sensor includes strain data at each sampling point along the fiber arc length direction. , For the first along the fiber arc length direction The distance from the sampling point to the starting end of the optical fiber; For the first Strain data at the sampling point; This represents the total number of sampling points.

[0013] Furthermore, the noise intensity of the estimated sensor strain data at each sampling point The added noise amplitude is To remove noise, a fixed-window-length filter is applied to the strain data after adding noise, resulting in filtered strain data. This allows for the determination of the curvature at the sampling points. and bending direction angle .

[0014] Furthermore, the curvature error amplification risk amount for:

[0015]

[0016] in, These are the weighting coefficients; Used to prevent the denominator from being 0.

[0017] Furthermore, the risk of amplified bending direction angle error. for:

[0018]

[0019] in, These are the weighting coefficients.

[0020] Furthermore, the location features of the sampling points for:

[0021]

[0022] Furthermore, the boundary distance feature for:

[0023]

[0024] Constructing feature vectors .

[0025] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of an adaptive filtering-based fiber shape sensing error correction method.

[0026] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of an adaptive filtering-based fiber shape sensing error correction method.

[0027] A computer program product includes computer instructions that, when executed by a processor, implement steps of an adaptive filtering-based fiber shape sensing error correction method.

[0028] The beneficial effects of this invention are as follows:

[0029] This invention employs a reconstruction error propagation model to drive adaptive filtering. Addressing the issue of errors propagating from strain measurement to curvature and torsion in fiber shape reconstruction and accumulating and amplifying during integration, it introduces an error amplification risk index based on the error propagation relationship. The length of the spatial smoothing window is determined point-by-point, and range and stability constraints are applied to the window. This allows the filtering process to be based on noise propagation and the characteristics of the curve itself. It can be used for spatial domain smoothing of sampling sequences along arc length in fiber optic measurement systems and can be applied to improve the accuracy of key information such as curvature and torsion, ultimately enhancing the accuracy of the three-dimensional shape reconstruction results. Attached Figure Description

[0030] Figure 1 This is a diagram of the overall architecture of the present invention.

[0031] Figure 2 This is a graph showing adaptive window changes.

[0032] Figure 3 This is a comparison chart of the effects of fixed window traversal and adaptive filtering.

[0033] Figure 4 This is a graph showing the actual curvature and bending angle trends.

[0034] Figure 5 This is a comparison chart of the curvature filtering effects of the present invention with other methods.

[0035] Figure 6 This is a comparison chart of the bending angle filtering effect of the present invention with other methods.

[0036] Figure 7 This is a data table showing the improvement in curvature, bending angle, and position compared to other methods. Detailed Implementation

[0037] The present invention will now be further described with reference to the accompanying drawings.

[0038] This invention designs an optical fiber shape sensing error correction method based on adaptive filtering. By constructing an index that reflects the risk of error amplification, and determining the smoothing window length point by point, the method introduces window range constraints and stability constraints to make the filtering intensity adaptively change with the arc length: increasing the filtering intensity or implementing gating in the low curvature high-risk region to reduce the amplification of torsion-related errors, and reducing the filtering intensity in the high curvature detail region to preserve geometric details, thereby reducing the integral accumulation error and improving the stability and accuracy of shape reconstruction.

[0039] for For a fiber optic system, its apparent curvature vector can be expressed as:

[0040]

[0041] The curvature scalar is then:

[0042]

[0043] in, Represented as the first Strain data obtained from fiber optic measurements Indicates the fiber core spacing of each fiber. This represents the fixed angular offset between each fiber core and the optical fiber material coordinate system. and This is a unit vector along the y-axis and z-axis of the material coordinate system. The magnitude of the apparent curvature vector is determined by the strain value and the radial distance, while its direction is controlled by the angular offset determined by calibration.

[0044] Further decomposition of curvature into components in the material coordinate system:

[0045] ,

[0046] The bending angle and deflection are:

[0047]

[0048]

[0049] So in With a core fiber, we can obtain all the necessary data through strain measurement.

[0050] First, obtain the sampling sequence along the arc length direction. , For the first along the fiber arc length direction The distance from the sampling point to the starting end of the optical fiber; For the first Strain data at the sampling points This represents the total number of sampling points.

[0051] The measurement sequence can be a sequence of core strains, a sequence of curvature components, a sequence of curvature deflection, or other equivalent quantities related to curvature deflection. The acquired data is calculated from strain data obtained at different sensor placement points at the same measuring point under arbitrary bending curvature and angular directions.

[0052] Preliminary smoothing of the measurement sequence is performed using a preset initial window example. A common method is to take measurement points and sort them out. By using a fixed proportion that satisfies the constraints of odd number and minimum window, preliminary estimates are obtained, such as curvature component estimation. and the curvature amplitude calculated from it. Direction angle And calculate the spatial derivative characteristics, such as , wait.

[0053] For noise intensity estimation, robust statistical estimation of the residuals is preferred to obtain the measured noise amplitude or its proxy. Our default method is curvature component noise, but strain noise or equivalent noise intensities are also acceptable. This noise estimation does not require local point-by-point estimation; instead, a global robust estimation is first employed to ensure stability and simplicity of engineering implementation.

[0054] An error amplification risk index is constructed based on the error propagation mechanism. First-order linearization of the relationships between curvature and bending angle, and deflection and strain, yields: the curvature modulus and the bending direction angle are respectively... , Applying a first-order linearization to the above relationship yields:

[0055] ,

[0056] Therefore, when the curvature component error is on the order of constant, It remains constant, while contain The related terms significantly amplify the bending direction angle error in the small curvature range. Furthermore, if the deflection or torsional correlation quantity is... If the spatial derivative or equivalent form is obtained, the differentiation operation will further amplify the high-frequency noise, thus making the torsion error more significant than the curvature error.

[0057] Therefore, the aforementioned risk indicators include at least the following two categories:

[0058] Bending angle / deflection related error risk: Construct an amplification term positively correlated with deflection estimation and negatively correlated with curvature estimation, and combine it with noise amplitude to form a comprehensive risk. This design is used to characterize the non-uniformity of deflection error amplification in low curvature regions, thus providing a basis for subsequent window enlargement or gating. Curvature error risk: Construct a risk quantity mainly related to noise amplitude and not explicitly coupled with deflection, used to characterize the basic noise level of curvature estimation. Thus, the following equation applies:

[0059]

[0060]

[0061] in, Used to prevent the denominator from being 0; and The weighting coefficients represent the rate of change of the direction angle along the arc length, the severity of local torsion or turning, and the relative change of curvature in the local area, respectively. This term serves as an independent feature input window prediction model, where the learner implicitly weights and nonlinearly combines the contributions of the two during training, thereby adaptively learning the trade-off between noise level and detail intensity.

[0062] Location features and boundary distances are introduced to characterize endpoint or boundary effects, thus forming a comprehensive risk feature vector for window decision-making. The location and boundary distance can be defined as follows:

[0063] ,

[0064] Constructing feature vectors .

[0065] B. Window Decision

[0066] B1: Constructing a training sample and label generation mechanism. Ideally, diverse training data should be generated through simulation curves and noise injection, covering combinations of different radii of curvature, torsion levels, lengths and sampling densities, and noise amplitudes to enhance the model's generalization ability. For each sample curve, within the candidate window set... Smoothing is performed separately, and the error assessment of each sampling point under different windows is calculated based on the local error; the local error can be achieved by using a sliding window MSE / RMSE centered on the sampling point to improve tag stability and reduce the influence of outliers.

[0067] B2: Introduce window stability constraints to suppress window jitter. For each sampling point, when the local errors of multiple candidate windows are close to the optimal error, a smaller window is selected or a hysteresis mechanism is used to avoid frequent switching of windows between adjacent points, thereby improving the continuity and repeatability of filtering and reconstruction results.

[0068] B3: Introducing a low-curvature observability gating strategy. In low-curvature regions, torsion-related quantities may be unobservable or at high amplification risk. Therefore, the torsion window labels corresponding to this region can be invalidated. During the training and inference phases, these samples can be downweighted, removed, or their outputs frozen. Furthermore, during the inference phase, a stronger noise suppression window or gating strategy can be directly applied to low-curvature points. This mechanism prevents the learning model from being misled by noise in unobservable regions.

[0069] B4: Feature Normalization Weighting. To address the inconsistency problem of fixed optimal windows, for joint objectives, the curvature component, curvature amplitude, and direction error are integrated and normalized and weighted, enabling the model to learn the window selection rules for multi-objective trade-offs. Let the filtering window be... The estimate obtained at that time was The true value is The normalized weights are given as follows:

[0070]

[0071] B5: Training the Window Prediction Model. Using the feature vector obtained from risk modeling as input and the optimal window label as output, a window predictor is trained using a regression or ensemble regression model. During training, the model implicitly weights and nonlinearly combines the contributions of each feature to learn the trade-off between noise suppression and detail preservation in window selection.

[0072] The predictor can be a tree regression ensemble, gradient boosting, linear or nonlinear regression, or other models that can map features to window lengths. After training, during the inference phase, the prediction window length is output point-by-point for new data, and post-processing is performed: cropping to... Force odd numbers and smooth them according to stability rules.

[0073] C. Window smoothing

[0074] C1: Length of the point-by-point window based on the output of step B The measurement sequence is locally fitted and smoothed in the neighborhood of each sampling point. Savitzky-Golay local polynomial smoothing is preferred as the implementation method, and the polynomial order is set to match the window length; equivalent local polynomial regression, weighted least squares smoothing, or other local fitting and smoothing methods can also be used.

[0075] C2: For endpoints or boundary segments, adjust the window or fitting neighborhood based on boundary distance or location features to reduce the deviation caused by boundary effects.

[0076] C3: Outputs the smoothed curvature components, curvature amplitude, direction change, or torsion estimate; when needed, the smoothing result is input into the subsequent curve frame integration module for 3D shape reconstruction. Because the window automatically enhances noise suppression and has stability constraints in low-curvature, high-risk areas, it can suppress the amplification of torsion-related errors and reduce cumulative integration errors; in high-curvature, detailed areas, it automatically weakens smoothing to preserve geometric details, thereby improving the overall stability and accuracy of shape reconstruction and mitigating the inconsistency problem where a fixed window is optimal for one target but deteriorates for another.

[0077] Example 1:

[0078] A method for correcting optical fiber shape sensing errors based on adaptive filtering includes the following steps:

[0079] Step 1: Acquire historical data from the sensor, including strain sequences along the fiber arc length direction. ;

[0080] in, For the first along the fiber arc length direction The distance from the sampling point to the starting end of the optical fiber; For the first Strain data at the sampling point; This represents the total number of sampling points;

[0081] Step 2: Estimate the noise level of the sensor. strain data at each sampling point The added noise amplitude is To remove noise, a fixed-window-length filter is applied to the strain data after adding noise, resulting in filtered strain data. This allows for the determination of the curvature at the sampling points. and bending direction angle ;

[0082] By adjusting the window length, the optimal window length suitable for filtering strain data after adding noise at each sampling point can be determined. ;

[0083] Step 3: Construct indicators reflecting the risk of error amplification, including the curvature error amplification risk level. Risk of amplification with bending direction angle error ;

[0084]

[0085]

[0086] in, and These are the weighting coefficients; Used to prevent the denominator from being 0;

[0087] Introducing location features Distance from the boundary ;

[0088]

[0089] Constructing feature vectors ;

[0090] Step 4: Based on Feature vector of each sampling point and optimal window length Construct a training set;

[0091] By introducing window range constraints and stability constraints, and using a training set to train the prediction model, the trained prediction model can output the corresponding optimal window length based on the input feature vector.

[0092] Step 5: When the sensor is used for measurement, it acquires the real-time strain sequence along the arc length of the optical fiber, and then acquires the curvature and bending direction angle at each sampling point. It calculates the curvature error amplification risk, bending direction angle error amplification risk, position features and boundary distance, and constructs a feature vector. The feature vector is input into the trained prediction model to obtain the optimal window length. The real-time strain sequence is filtered based on the optimal window length to achieve error correction.

[0093] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for correcting optical fiber shape sensing errors based on adaptive filtering, characterized in that: Historical data from the sensor is acquired, including strain data at each sampling point along the fiber arc length; the noise intensity of the sensor is estimated, noise is added to the strain data at each sampling point, and a fixed window length filter is applied to the strain data after adding noise to obtain the filtered strain data, thereby determining the curvature and bending direction angle at the sampling point; By adjusting the window length, the optimal window length suitable for filtering strain data after adding noise at each sampling point can be determined. Indicators reflecting the risk of error amplification are constructed, including the risk of curvature error amplification and the risk of bending direction angle error amplification. The positional features and boundary distance features of each sampling point are introduced, and combined with the risk of curvature error amplification and the risk of bending direction angle error amplification, a feature vector of each sampling point is constructed. A training set is constructed based on the feature vectors of each sampling point and the optimal window length. The prediction model is trained using the training set so that the trained prediction model can output the corresponding optimal window length based on the input feature vector. When the sensor is used for measurement, it acquires the real-time strain sequence along the arc length of the optical fiber, and then acquires the curvature and bending direction angle at each sampling point. It calculates the curvature error amplification risk, bending direction angle error amplification risk, position features and boundary distance, and constructs a feature vector. The feature vector is input into the trained prediction model to obtain the optimal window length; the real-time strain sequence is then filtered based on the optimal window length to achieve error correction.

2. The fiber optic shape sensing error correction method based on adaptive filtering according to claim 1, characterized in that: The acquisition of historical data from the sensor includes strain data at each sampling point along the fiber arc length direction. , For the first along the fiber arc length direction The distance from the sampling point to the starting end of the optical fiber; For the first Strain data at the sampling point; This represents the total number of sampling points.

3. The fiber optic shape sensing error correction method based on adaptive filtering according to claim 2, characterized in that: The noise intensity of the estimated sensor strain data at each sampling point The added noise amplitude is To remove noise, a fixed-window-length filter is applied to the strain data after adding noise, resulting in filtered strain data. This allows for the determination of the curvature at the sampling points. and bending direction angle .

4. The fiber optic shape sensing error correction method based on adaptive filtering according to claim 3, characterized in that: The curvature error amplification risk for: in, These are the weighting coefficients; Used to prevent the denominator from being 0.

5. The fiber optic shape sensing error correction method based on adaptive filtering according to claim 4, characterized in that: The risk of amplified bending direction angle error for: in, These are the weighting coefficients.

6. The fiber optic shape sensing error correction method based on adaptive filtering according to claim 5, characterized in that: The location features of the sampling points for: 。 7. The fiber optic shape sensing error correction method based on adaptive filtering according to claim 6, characterized in that: The boundary distance feature for: Constructing feature vectors .

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1 to 7.

Images