A method and system for detecting the 3D topography of a fiber connector end face

Abstract

The present application relates to the technical field of optical fiber communication device detection, and proposes an optical fiber connector end face 3D topography detection method and system, which comprises: acquiring a multi-step phase shift interference image sequence of the optical fiber connector end face; using a fiber core boundary detection network based on radial gradient sensing to perform interface segmentation on the multi-step phase shift interference image sequence to obtain interface position data; based on the interface position data and the multi-step phase shift interference image sequence, using a double-coordinate system fusion phase unwrapping algorithm to perform phase unwrapping to obtain initial phase data; using a region confidence weighted reconstruction algorithm to perform global reconstruction on the initial phase data to obtain global phase field data; using an IEC standard constraint layered particle swarm optimization algorithm to perform spherical parameter fitting on the global phase field data to obtain spherical geometric parameters; and generating an optical fiber connector end face 3D topography detection result according to the spherical geometric parameters. The present application improves the precision and reliability of optical fiber connector end face 3D topography detection.

Description

Technical Field

[0001] This invention relates to the field of optical fiber communication device testing technology, and in particular to a method and system for detecting the 3D morphology of the end face of an optical fiber connector. Background Technology

[0002] Fiber optic connectors are core components in fiber optic communication systems, used to achieve detachable connections between optical fibers and efficient transmission of optical signals. The geometric quality of the connector endface directly affects key performance indicators such as insertion loss, return loss, and connection stability. During processing and use, the connector endface is subject to various defects such as surface scratches, bubbles, contamination, and geometric deviations due to factors like polishing process parameters, environmental temperature and humidity, and mechanical stress. 3D morphology inspection technology, as a primary means of evaluating fiber optic connector quality, identifies and assesses the endface quality status by detecting its three-dimensional geometric features. However, the unique core-cladding layered structure, circular geometry, and sub-micron precision requirements of fiber optic connectors present numerous challenges to 3D morphology signal analysis.

[0003] In existing technologies, 3D morphology inspection of fiber optic connector end faces mainly employs traditional white light interferometry and basic image processing methods to achieve basic end-face morphology inspection functions. However, existing methods do not adequately consider the inherent physical correlation mechanism between the complex layered structure of fiber optic connectors and the characteristics of interferometric measurement signals. This makes it difficult to organically integrate the objective physical constraints of optical interferometry with the actual geometric features of the fiber optic end face, resulting in the inability to achieve high-precision automatic 3D morphology inspection of fiber optic connector end faces. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for detecting the 3D morphology of fiber optic connector end faces, which solves the problem that existing methods do not adequately consider the inherent physical correlation mechanism between the complex layered structure of fiber optic connectors and the characteristics of interferometric measurement signals, making it difficult to organically integrate objective optical interference physical constraints with the actual geometric structural features of the fiber optic end faces, thus making it impossible to achieve high-precision automatic detection of the 3D morphology of fiber optic connector end faces.

[0005] The technical solution of this invention is implemented as follows: On one hand, this invention provides a method for 3D morphology detection of the end face of an optical fiber connector, comprising the following steps: Acquire a multi-step phase-shifting interferometry image sequence from the end face of an optical fiber connector; A fiber core boundary detection network based on radial gradient sensing is used to segment the interface of the multi-step phase-shifting interferometric image sequence to obtain interface position data; Based on the interface position data and the multi-step phase-shifting interferometric image sequence, a dual-coordinate system fusion phase unpacking algorithm is used to perform phase unpacking to obtain initial phase data; The initial phase data is globally reconstructed using a region confidence-weighted reconstruction algorithm to obtain global phase field data. The global phase field data were fitted with spherical parameters using the IEC standard constrained hierarchical particle swarm optimization algorithm to obtain the spherical geometric parameters. The 3D morphology detection results of the fiber optic connector end face are generated based on the spherical geometric parameters.

[0006] Based on the above technical solutions, preferably, the fiber core boundary detection network includes a polar coordinate convolutional layer, a radial gradient enhancement model, and a multi-scale circular attention mechanism; Polar coordinate convolutional layers are used to perform coordinate system transformation and feature extraction on the multi-step phase-shifting interferometric image sequence to obtain polar coordinate feature data; A radial gradient enhancement model is used to perform radial gradient calculation and enhancement processing on the polar coordinate feature data to obtain radially enhanced feature data. The radial gradient enhancement model includes a radial difference operator, a gradient magnitude calculation layer, and an adaptive enhancement layer. The radial enhancement feature data is weighted by attention and fused at multiple scales using a multi-scale circular attention mechanism to obtain the interface position data.

[0007] Based on the above technical solutions, preferably, the polar coordinate convolutional layer includes: The multi-step phase-shifting interferometric image sequence is transformed from Cartesian coordinates to polar coordinates using a coordinate transformation matrix to obtain polar coordinate image data; Radial and angular convolution kernels are used to perform radial and angular convolution operations on the polar coordinate image data to obtain bidirectional convolution features; a feature fusion network is used to fuse the bidirectional convolution features to obtain the polar coordinate feature data. The radial gradient enhancement model includes: The radial gradient data is obtained by performing first-order and second-order difference operations on the polar coordinate feature data in the radial direction using a radial difference operator. A gradient magnitude calculation layer is used to calculate the magnitude and direction of the radial gradient data to obtain gradient magnitude data; An adaptive enhancement layer is used to dynamically adjust the weights and perform nonlinear enhancement on the gradient magnitude data based on the local gradient intensity to obtain the radially enhanced feature data.

[0008] Based on the above technical solutions, preferably, the dual coordinate system fusion phase unpacking algorithm includes coordinate transformation error compensation and phase gradient constraint at the interface; Based on the interface position data, the multi-step phase-shifting interferometric image sequence is segmented to obtain core region image data and cladding region image data. Phase unpacking is performed on the core region image data and the cladding region image data in polar and Cartesian coordinate systems respectively to obtain dual-coordinate system phase data; The phase data of the dual coordinate system is corrected by using coordinate transformation error compensation to obtain corrected phase data; The initial phase data is obtained by performing gradient continuity processing on the corrected phase data using phase gradient constraints at the interface.

[0009] Based on the above technical solutions, preferably, the coordinate transformation error compensation includes: Establish a transformation error model from polar coordinates to Cartesian coordinates and obtain the error propagation matrix; Interpolation error data is obtained by calculating the interpolation error of the dual-coordinate system phase data based on the error propagation matrix. The corrected phase data is obtained by performing pixel-by-pixel error compensation on the dual-coordinate system phase data based on the interpolation error data. The formula for calculating the error propagation matrix is: ; in, For the error propagation matrix, the first... Line 1 Column elements; For the first Elements of the radial coordinate transformation Jacobian matrix at each radial position; For the first Elements of the Jacobian matrix for angular coordinate transformation at each angular position; These are the second-order error correction coefficients; For phase data in Cartesian coordinates The second-order mixed partial derivative at the point; For the first One polar coordinate radial sampling point, For the first One polar coordinate angle sampling point; For the first Sampling interval at each radial position, For the first Sampling interval at each angular position; This refers to the sampling error weighting coefficient; This is the second numerical stabilization parameter; This is the third numerical stabilization parameter.

[0010] Based on the above technical solutions, preferably, the regional confidence weighted reconstruction algorithm includes a regional confidence model and spherical geometric prior constraints; Based on the core region noise characteristics and cladding region noise characteristics of the initial phase data, a regional confidence model is established to obtain confidence weight data; Based on the geometric features of the spherical polishing of the fiber end face, a priori geometric constraints on the spherical surface are established to obtain the geometric constraint parameters. The initial phase data is globally reconstructed using a weighted least squares optimization method, with the confidence weight data as the weight matrix and the geometric constraint parameters as the constraint conditions, to obtain the global phase field data.

[0011] Based on the above technical solutions, preferably, the regional confidence model includes: The local variances of the core region and cladding region are calculated based on the initial phase data to obtain the region variance data. The signal-to-noise ratio (SNR) parameters for each region are calculated based on the regional variance data to obtain the SNR data; The signal-to-noise ratio data is normalized using an inverse variance weighting method to obtain the confidence weight data.

[0012] Based on the above technical solutions, preferably, the IEC standard constrained hierarchical particle swarm optimization algorithm includes hierarchical sampling initialization, adaptive search of constraint boundaries, and intelligent solution selection based on connection loss prediction; The global phase field data is initialized by hierarchical sampling based on the constraints of radius of curvature, vertex offset, and fiber height in the IEC standard to obtain initial particle swarm data. The initial particle swarm data is iteratively optimized using constrained boundary adaptive search to obtain a set of candidate spherical parameters; The candidate spherical parameter set is evaluated and screened using a connection loss prediction intelligent solution selection method to obtain the spherical geometric parameters.

[0013] Based on the above technical solutions, preferably, the hierarchical sampling initialization includes: Based on the constraints of the IEC standard, the spherical geometric parameter space is divided into multiple hierarchical subspaces; Initial particle positions are generated within each hierarchical subspace using the Latin hypercube sampling method, resulting in hierarchical particle position data. Based on the global phase field data, the fitness assessment and density adjustment of the hierarchical particle position data are performed to obtain the initial particle swarm data. The formula for dividing the spherical geometric parameter space into multiple hierarchical subspaces based on the IEC standard constraint range is as follows: ; in, The subspace number to which the spherical geometric parameters belong; These are the geometric parameter values ​​of the sphere to be layered; This represents the minimum constraint range for spherical geometric parameters in the IEC standard. This represents the maximum constraint range for spherical geometric parameters in the IEC standard. This represents the total number of hierarchical subspaces. This is the center offset correction factor; This is the distance attenuation parameter; The center value of the IEC standard constraint range; The width of the constraint range as defined by the IEC standard; It is an exponential function; This is the floor function.

[0014] On the other hand, the present invention also provides a 3D topography inspection system for fiber optic connector end faces, the system comprising: A multi-step phase-shifting interferometric image acquisition module is used to acquire a multi-step phase-shifting interferometric image sequence from the end face of an optical fiber connector; The radial gradient sensing interface segmentation module is used to segment the interface of the multi-step phase-shifting interferometric image sequence using a fiber core boundary detection network based on radial gradient sensing, and to obtain interface position data. The dual-coordinate system fusion phase unpacking module is used to perform phase unpacking based on the interface position data and the multi-step phase-shifting interferometric image sequence, using a dual-coordinate system fusion phase unpacking algorithm to obtain initial phase data. The regional confidence-weighted reconstruction module is used to globally reconstruct the initial phase data using a regional confidence-weighted reconstruction algorithm to obtain global phase field data. The IEC standard constrained spherical fitting module is used to fit the global phase field data to spherical parameters using the IEC standard constrained hierarchical particle swarm optimization algorithm to obtain spherical geometric parameters. The 3D topography detection result generation module is used to generate 3D topography detection results of the fiber optic connector end face based on the spherical geometric parameters.

[0015] The 3D morphology detection method and system for fiber optic connector end faces of the present invention have the following advantages over the prior art: (1) By using a radial gradient sensing fiber core boundary detection network and a dual-coordinate system fusion phase unpacking algorithm, the interface is accurately segmented and phase information is extracted using a multi-step phase-shifting interferometric image sequence. The weight allocation of different regions is dynamically adjusted by combining the region confidence weighted reconstruction algorithm. The parameter space of the hierarchical particle swarm optimization algorithm is limited and intelligent solution selection is performed according to the IEC standard constraints, which improves the accuracy and reliability of 3D morphology detection of fiber connector end face. (2) By using polar coordinate convolutional layers and radial gradient enhancement models, the Cartesian coordinate image is converted into polar coordinate representation using coordinate transformation matrix and bidirectional convolution feature extraction is performed. First-order and second-order gradient calculation and adaptive enhancement processing are performed by combining radial difference operators. Multi-scale circular attention mechanism is used to perform attention weighting and multi-scale fusion of radial enhancement features, which improves the accuracy and robustness of fiber core boundary detection at the end face of fiber connector. At the same time, the radial gradient perception and circular attention mechanism optimization meet the requirements for accurate recognition of circular geometric features at the end face of fiber connector. (3) By using the dual-coordinate system phase unpacking algorithm and coordinate transformation error compensation, the multi-step phase shifting interferometric image sequence is divided into the core region and the cladding region using the interface position data. The phase unpacking is performed in the polar coordinate system and the Cartesian coordinate system respectively. The interpolation error is calculated and the pixel-by-pixel error is compensated by combining the error propagation matrix. The phase gradient constraint at the interface is used to perform gradient continuity processing on the corrected phase data, which improves the accuracy and stability of phase unpacking of the fiber optic connector end face. At the same time, the error propagation matrix and gradient constraint optimization meet the requirements for accurate reconstruction of the phase information of the layered structure of the fiber optic connector end face. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a 3D morphology detection method for the end face of an optical fiber connector according to the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides a method for 3D topography detection of fiber optic connector end face, comprising the following steps: Acquire a multi-step phase-shifting interferometry image sequence from the end face of an optical fiber connector; A fiber core boundary detection network based on radial gradient sensing is used to segment the interface of the multi-step phase-shifting interferometric image sequence to obtain interface position data; Based on the interface position data and the multi-step phase-shifting interferometric image sequence, a dual-coordinate system fusion phase unpacking algorithm is used to perform phase unpacking to obtain initial phase data; The initial phase data is globally reconstructed using a region confidence-weighted reconstruction algorithm to obtain global phase field data. The global phase field data were fitted with spherical parameters using the IEC standard constrained hierarchical particle swarm optimization algorithm to obtain the spherical geometric parameters. The 3D morphology detection results of the fiber optic connector end face are generated based on the spherical geometric parameters.

[0020] Specifically, this embodiment utilizes a radial gradient-sensing core boundary detection network and a dual-coordinate system fusion phase unpacking algorithm. It employs a multi-step phase-shifting interferometric image sequence for precise interface segmentation and phase information extraction. A region confidence-weighted reconstruction algorithm dynamically adjusts the weight allocation for different regions, and the hierarchical particle swarm optimization algorithm is constrained by IEC standards through parameter space limitations and intelligent solution selection. This multi-coordinate system collaborative processing mechanism addresses the issues of insufficient core-cladding interface segmentation accuracy, accumulated phase unpacking errors, and inaccurate fitting of spherical geometric parameters on the fiber optic connector end face, thereby improving the accuracy and reliability of 3D topography detection of the fiber optic connector end face.

[0021] The acquisition of the multi-step phase-shifting interferometry image sequence of the fiber optic connector end face includes: An automatic positioning algorithm based on end-face reflection characteristics is used to locate the end face of the fiber optic connector and obtain pre-processed end-face data. The automatic positioning algorithm includes end-face center identification, angle correction, and adaptive light intensity adjustment.

[0022] In one specific embodiment, the automatic positioning algorithm based on end-face reflection features includes: The circular Hough transform detection algorithm is used to locate the center of the fiber optic connector end face and obtain the center coordinates of the end face. Based on the center coordinates of the end face, the least squares fitting algorithm is used to correct the end face angle and obtain the correction angle parameters. Based on the correction angle parameters and the end-face reflection intensity distribution, a histogram equalization algorithm is used to adjust the light intensity to obtain the preprocessed end-face data.

[0023] Based on the preprocessed end face data, a multi-step phase shift interferometry measurement is performed using a constant step phase shift control algorithm to obtain the multi-step phase shift interferometry image sequence. The constant step phase shift control algorithm includes phase shift step calculation, optical path difference control, and image synchronous acquisition.

[0024] In one specific embodiment, the constant-step phase shift control algorithm includes: The optimal phase shift step size is calculated based on interference fringe contrast analysis to obtain the step size parameters; Based on the step size parameters, a piezoelectric ceramic actuator is used to control the movement of the reference mirror to adjust the optical path difference and obtain multiple phase shift positions. Interference images are acquired synchronously using a CCD camera at each phase shift position to obtain the multi-step phase shift interference image sequence.

[0025] Specifically, this embodiment integrates an automatic positioning algorithm based on end-face reflection features with a constant-step phase-shift control algorithm. It utilizes circular Hough transform detection and least-squares fitting for end-face center positioning and angle correction. A histogram equalization algorithm is combined for adaptive light intensity adjustment, and interference fringe contrast analysis is used to calculate the optimal phase-shift step size. A piezoelectric ceramic actuator controls the reference mirror movement, and a CCD camera synchronously acquires data to achieve precise optical path difference adjustment. This automatic positioning and precise phase-shift coordinated control mechanism solves the problems of insufficient manual positioning accuracy of fiber optic connector end faces, uneven phase-shift step sizes, and inconsistent light intensity distribution leading to decreased interference image quality. It improves the stability and consistency of multi-step phase-shift interference image sequence acquisition. Furthermore, end-face reflection feature recognition and constant-step control optimization meet the automated acquisition requirements for high-precision interferometric measurements of fiber optic connector end faces.

[0026] The fiber core boundary detection network includes polar coordinate convolutional layers, a radial gradient enhancement model, and a multi-scale circular attention mechanism. Polar coordinate convolutional layers are used to perform coordinate system transformation and feature extraction on the multi-step phase-shifting interferometric image sequence to obtain polar coordinate feature data; A radial gradient enhancement model is used to perform radial gradient calculation and enhancement processing on the polar coordinate feature data to obtain radially enhanced feature data. The radial gradient enhancement model includes a radial difference operator, a gradient magnitude calculation layer, and an adaptive enhancement layer. The radial enhancement feature data is weighted by attention and fused at multiple scales using a multi-scale circular attention mechanism to obtain the interface position data.

[0027] The polar coordinate convolutional layer includes: The multi-step phase-shifting interferometric image sequence is transformed from Cartesian coordinates to polar coordinates using a coordinate transformation matrix to obtain polar coordinate image data; Radial and angular convolution kernels are used to perform radial and angular convolution operations on the polar coordinate image data to obtain bidirectional convolution features; a feature fusion network is used to fuse the bidirectional convolution features to obtain the polar coordinate feature data. The radial gradient enhancement model includes: The radial gradient data is obtained by performing first-order and second-order difference operations on the polar coordinate feature data in the radial direction using a radial difference operator. A gradient magnitude calculation layer is used to calculate the magnitude and direction of the radial gradient data to obtain gradient magnitude data; An adaptive enhancement layer is used to dynamically adjust the weights and perform nonlinear enhancement on the gradient magnitude data based on the local gradient intensity to obtain the radially enhanced feature data.

[0028] In one specific embodiment, the formula for calculating the gradient magnitude data is: ; in, The gradient magnitude data serves as the output of the gradient magnitude calculation layer. The first-order radial difference result of the polar coordinate feature data is obtained by calculating the radial difference operator; The second-order radial difference result of the polar coordinate feature data is obtained by calculating the radial difference operator; These are the second-order difference weighting coefficients, used to adjust the contribution of the second-order gradient in magnitude calculation; These are the first-order and second-order cross-term coefficients, used to control the coupling strength between the first-order and second-order gradients; These are angle modulation parameters used to incorporate directional information into gradient magnitude calculations. This is the first numerical stabilization parameter, used to prevent division by zero operations; This is the difference enhancement coefficient, used to enhance the difference characteristics between the first and second order differences; It is a non-linear enhancement index used to achieve non-linear enhancement of difference features.

[0029] Specifically, the gradient magnitude data in this embodiment differs from the traditional gradient magnitude calculation which typically uses the simple Euclidean norm. By introducing adaptive weight coefficients The contribution weights are dynamically adjusted based on the second-order difference characteristics; coupled cross terms are used. The relationship between the first and second order gradients is captured, and the angle information is provided by the cosine function; a nonlinear enhancement term is used. This highlights the difference characteristics between the first and second order differences, and combines them with the nonlinear enhancement exponent. Achieve nonlinear amplification.

[0030] The formula for calculating the adaptive enhancement weights of the adaptive enhancement layer is: ; in, Adaptive weight enhancement, used for dynamic weight adjustment and nonlinear enhancement; This is a gradient sensitivity adjustment factor used to control the degree to which the weights respond to the gradient strength. This is gradient magnitude data; The standard deviation of the local gradient strength is used to reflect the degree of dispersion of the gradient distribution; This is a dynamic denominator adjustment parameter used to dynamically adjust the weight response based on the gradient deviation; This is the mean of the local gradient intensity, used to characterize the gradient level in a local region; This is the exponential enhancement coefficient, used to control the degree to which the exponential term enhances the weights; This is a local similarity adjustment parameter used to control the degree to which the weights depend on the local gradient mean; This is a numerical stabilization parameter used to ensure the numerical stability of the denominator; It is the hyperbolic tangent activation function; It is an exponential function.

[0031] Specifically, the adaptive enhancement weights in this embodiment differ from traditional linear weight adjustments, employing a dual adaptive mechanism. Through dynamic denominator adjustment, using... The weight response is dynamically adjusted based on the gradient deviation; based on the exponential enhancement factor. Provides secondary enhancement based on gradient similarity.

[0032] This embodiment significantly improves the accuracy and robustness of core boundary detection by fusing first-order and second-order gradient information and introducing angular features. The adaptive weighting mechanism can dynamically adjust the detection sensitivity according to local gradient characteristics, and can still accurately identify the core-cladding interface in noisy environments, greatly improving detection accuracy compared to traditional methods.

[0033] The dual-coordinate system fusion phase unpacking algorithm includes coordinate transformation error compensation and phase gradient constraint at the interface; Based on the interface position data, the multi-step phase-shifting interferometric image sequence is segmented to obtain core region image data and cladding region image data. Phase unpacking is performed on the core region image data and the cladding region image data in polar and Cartesian coordinate systems respectively to obtain dual-coordinate system phase data; The phase data of the dual coordinate system is corrected by using coordinate transformation error compensation to obtain corrected phase data; The initial phase data is obtained by performing gradient continuity processing on the corrected phase data using phase gradient constraints at the interface.

[0034] The coordinate transformation error compensation includes: Establish a transformation error model from polar coordinates to Cartesian coordinates and obtain the error propagation matrix; Interpolation error data is obtained by calculating the interpolation error of the dual-coordinate system phase data based on the error propagation matrix. The corrected phase data is obtained by performing pixel-by-pixel error compensation on the dual-coordinate system phase data based on the interpolation error data.

[0035] In one specific embodiment, the error propagation matrix is ​​calculated as follows: ; in, For the error propagation matrix, the first... Line 1 Column elements, used to describe the first The radial position and the first Error propagation relationship between angular positions; For the first Elements of the radial coordinate transformation Jacobian matrix at each radial position; For the first Elements of the Jacobian matrix for angular coordinate transformation at each angular position; This is a second-order error correction coefficient, used to account for nonlinear errors in coordinate transformation; For phase data in Cartesian coordinates The second-order mixed partial derivative at the point; For the first One polar coordinate radial sampling point, For the first One polar coordinate angle sampling point; For the first Sampling interval at each radial position, For the first Sampling interval at each angular position; This is the sampling error weighting coefficient, used to adjust the impact of the sampling interval on error propagation; This is the second numerical stabilization parameter, used to prevent numerical instability when the radial position coordinates approach zero; This is the third numerical stabilization parameter, used to prevent numerical instability when the angular position coordinates are close to zero.

[0036] Specifically, the error propagation matrix in this embodiment differs from traditional error propagation models that only consider the first-order Jacobian transform. It introduces a three-layer error correction mechanism, including a spatially indexed Jacobian matrix, a second-order nonlinear correction term, and adaptive compensation for the sampling interval. The spatially indexed Jacobian matrix is ​​obtained through… Clearly define position dependence to avoid global uniformization error; the second-order nonlinear correction term is achieved through... Considering the curvature effect of coordinate transformation; the adaptive compensation for the sampling interval is based on The error weights are adjusted based on the sampling density at different locations.

[0037] The formula for calculating the phase gradient constraint at the interface is: ; in, The initial phase data after gradient continuity processing is used as the final correction result; To correct the phase data, it is used as input for gradient constraints; This is the gradient constraint strength parameter, used to control the strength of the Laplace constraint. The Laplace operator is used to correct phase data and characterize the second-order spatial transformation of the phase. This is the interface distance attenuation coefficient, used to control the attenuation characteristics of the constraint effect with interface distance; The distance from the pixel to the core-cladding interface is used to determine the spatial range of the constraint. These are the normal gradient weight coefficients, used to adjust the contribution of the normal gradient constraint; To correct the gradient of the phase data, used to calculate the normal component; This is the interface normal vector, used to define the normal direction of the interface; This is the fourth numerical stabilization parameter, used to prevent division by zero errors when the gradient magnitude is zero; It is an exponential function.

[0038] Specifically, the phase gradient constraint at the interface in this embodiment differs from a simple Laplace constraint, employing a hierarchical gradient constraint strategy, including spatial localization constraints, normal gradient terms, and bidirectional gradient balancing. The spatial localization constraint is achieved through... Ensure that the constraint only applies near the interface; the normal gradient term is obtained through... It specifically handles the gradient continuity of the interface normal; the bidirectional gradient balancing comprehensively considers the synergistic effect of Laplace constraints and normal constraints.

[0039] This embodiment effectively solves the error accumulation problem in dual-coordinate system fusion by accurately modeling coordinate transformation errors and phase gradient constraints at the interface. It improves phase unpacking accuracy, particularly significantly enhancing phase continuity at the core-cladding interface.

[0040] The region confidence weighted reconstruction algorithm includes a region confidence model and spherical geometric prior constraints; Based on the core region noise characteristics and cladding region noise characteristics of the initial phase data, a regional confidence model is established to obtain confidence weight data; Based on the geometric features of the spherical polishing of the fiber end face, a priori geometric constraints on the spherical surface are established to obtain the geometric constraint parameters. The initial phase data is globally reconstructed using a weighted least squares optimization method, with the confidence weight data as the weight matrix and the geometric constraint parameters as the constraint conditions, to obtain the global phase field data.

[0041] The regional confidence model includes: The local variances of the core region and cladding region are calculated based on the initial phase data to obtain the region variance data. The signal-to-noise ratio (SNR) parameters for each region are calculated based on the regional variance data to obtain the SNR data; The signal-to-noise ratio data is normalized using an inverse variance weighting method to obtain the confidence weight data.

[0042] In one specific embodiment, the signal-to-noise ratio data is calculated as follows: ; in, The signal-to-noise ratio data is used to evaluate the phase quality of the core and cladding regions. The phase signal power within the region is used to reflect the strength of the effective phase information. This is regional variance data, used to characterize the degree of dispersion of phase data within a region; Cross-regional variance is used to characterize the effect of phase changes between adjacent regions; This is the cross-regional variance weighting coefficient, used to adjust the degree of influence of adjacent regions on the signal-to-noise ratio of the current region; This is the variance scaling factor, used to enhance the influence of the variance ratio between areas within and outside the region; This is the variance ratio adjustment parameter, used to control the sensitivity of the variance ratio to the signal-to-noise ratio; This is the fifth numerical stabilization parameter, used to prevent division by zero errors when the variance across regions is zero; This is a region area correction factor used to compensate for the impact of different region areas on the signal-to-noise ratio; This represents the current area, used to reflect the spatial scale of the region. The total detection area is used to normalize the proportion of the region area. It is an exponential function.

[0043] Specifically, the signal-to-noise ratio (SNR) data in this embodiment differs from the traditional SNR calculation which uses a fixed-weight power-variance ratio. It achieves multi-dimensional adaptive evaluation, including cross-regional coupling considerations, variance ratio adjustment, and region area normalization. The cross-regional coupling considerations are achieved through… The variance variation of adjacent regions is incorporated into the evaluation system; the variance ratio adjustment is achieved through... Dynamic adjustments are made based on the ratio of variance between the inside and outside of the region; the region area normalization is achieved through... Compensation for the impact of different area sizes on the signal-to-noise ratio.

[0044] The formula for calculating the confidence weight data is: ; in, The normalized confidence weight data serves as the weight matrix for weighted least squares optimization. For regional signal-to-noise ratio data; It is the inverse variance weighted index, used to control the sensitivity of the weights to the signal-to-noise ratio; This is the exponential correction factor, used to correct the weighting of the inverse variance by introducing the exponential term; This is the signal-to-noise ratio attenuation parameter, used to control the rate of exponential decay; For the first Regional signal-to-noise ratio data for each region are used for normalization calculation; This is the hyperbolic tangent enhancement coefficient, used to enhance the weights in high signal-to-noise ratio regions; This is a signal-to-noise ratio response adjustment parameter used to control the response characteristics of the hyperbolic tangent function; It is the hyperbolic tangent activation function; It is an exponential function.

[0045] Specifically, the confidence weight data in this embodiment differs from simple inverse variance weighting; it employs a hybrid weighting strategy, including a dual weighting mechanism, adaptive threshold response, and global normalization guarantee. The dual weighting mechanism utilizes… The adaptive threshold response term combines inverse variance weights and exponentially decaying weights. The threshold adjustment is achieved through the hyperbolic tangent function; based on the global normalization guarantee, the sum of weights in all regions is ensured to be 1, thus maintaining numerical stability.

[0046] This embodiment significantly improves the accuracy and stability of phase reconstruction through multi-dimensional signal-to-noise ratio evaluation and adaptive weight allocation. In the presence of local noise or defects, it can reduce the weight of problematic regions while enhancing the contribution of high-quality regions, thereby reducing overall reconstruction error. It is particularly suitable for complex scenarios where the fiber end face is contaminated or scratched.

[0047] The IEC standard constrained hierarchical particle swarm optimization algorithm includes hierarchical sampling initialization, adaptive search of constraint boundaries, and intelligent solution selection based on connection loss prediction. The global phase field data is initialized by hierarchical sampling based on the constraints of radius of curvature, vertex offset, and fiber height in the IEC standard to obtain initial particle swarm data. The initial particle swarm data is iteratively optimized using constrained boundary adaptive search to obtain a set of candidate spherical parameters; The candidate spherical parameter set is evaluated and screened using a connection loss prediction intelligent solution selection method to obtain the spherical geometric parameters.

[0048] The hierarchical sampling initialization includes: Based on the constraints of the IEC standard, the spherical geometric parameter space is divided into multiple hierarchical subspaces; Initial particle positions are generated within each hierarchical subspace using the Latin hypercube sampling method, resulting in hierarchical particle position data. The initial particle swarm data is obtained by evaluating the fitness and adjusting the density of the hierarchical particle position data based on the global phase field data.

[0049] In one specific embodiment, the calculation formula for dividing the spherical geometric parameter space into multiple hierarchical subspaces based on the IEC standard constraint range is as follows: ; in, The subspace number to which the spherical geometric parameter belongs is used to determine which subspace the spherical geometric parameter is in; These are the geometric parameter values ​​of the sphere to be layered, including radius of curvature, vertex offset, and fiber height; This is the minimum constraint range for the spherical geometric parameter in the IEC standard, used to define the lower bound of the parameter; This is the maximum constraint range for the spherical geometric parameter in the IEC standard, used to define the upper bound of the parameter; This represents the total number of hierarchical subspaces, used to control the precision of parameter space partitioning; This is the center offset correction coefficient, used to optimize the symmetry of the parameter distribution; This is a distance attenuation parameter used to control the intensity of layered adjustment based on center distance; This is the center value of the IEC standard constraint range, used to calculate the degree of parameter deviation; The width of the constraint range of the IEC standard is equal to ; It is an exponential function; This is a floor function used to round down a real number to the nearest integer. Specifically, the spherical geometric parameter space partitioning in this embodiment differs from traditional layering methods that use linear, equally spaced partitioning. It achieves nonlinear adaptive layering, including linear basic layering and Gaussian distribution correction, and ensures the integer characteristic of the layer numbers through two rounding operations. The linear basic layering is achieved through… Provides a basic uniform stratification; the Gaussian distribution correction is achieved through... Increase the stratification density near the standard center.

[0050] The formula for calculating the fitness evaluation and density adjustment of the layered particle position data based on the global phase field data is as follows: ; in, This is the particle fitness evaluation value, used to assess the quality of the spherical parameter fitting. The smaller the value, the better the fitting quality of the corresponding spherical geometric parameters; For the first The weighting coefficients for each phase field data point are used to reflect the importance of the data point. For global phase field data, the spatial coordinates of the phase field data points The actual phase value at that point; These are theoretical phase values ​​constructed based on spherical parameters, used for comparison with actual phase values; The radius of curvature parameter, represented by a particle, is one of the geometric parameters of a sphere. The vertex offset parameter, represented by the particle, is one of the geometric parameters of the sphere. The fiber height parameter is represented by a particle, and it is one of the geometric parameters of a sphere. The density is adjusted by weighting coefficients to balance fitting accuracy and particle distribution uniformity. This is a particle density adjustment term, used to prevent excessive particle aggregation from leading to local optima; To constrain the penalty coefficient, used to penalize particles that violate IEC standards; For the first The current value of each IEC standard parameter; For the first The maximum permissible value for each IEC standard parameter; For the Heaviside step function, when the parameter Returns 1 if the constraint is exceeded, otherwise returns 0.

[0051] Specifically, the particle fitness evaluation in this embodiment differs from traditional particle swarm fitness functions that only consider fitting error. It introduces a multi-objective balancing mechanism, including weighted fitting error, density adjustment mechanism, and hard constraint penalty. The weighted fitting error is achieved through… Taking into account the importance of data at different locations; the density adjustment mechanism is achieved through... To prevent excessive particle aggregation and maintain population diversity, the hard constraint penalty is implemented through... The project will severely punish solutions that violate IEC standards.

[0052] This embodiment achieves high-precision spherical parameter optimization while meeting IEC standard constraints through an intelligent hierarchical strategy and multi-objective fitness evaluation. The nonlinear hierarchical strategy enables higher search density of particles in key regions within the standard range, improving the optimization convergence speed while ensuring that all solutions strictly meet IEC standard requirements, providing a reliable parameter estimation basis for the standardized testing of fiber optic connectors.

[0053] The acquisition of the multi-step phase-shifting interferometry image sequence of the fiber optic connector end face includes: The spherical geometric parameters are used to perform 3D morphology reconstruction using a spherical fitting reconstruction algorithm to obtain end face 3D morphology data. The spherical fitting reconstruction algorithm includes spherical equation construction, spatial coordinate calculation, and morphology interpolation processing.

[0054] In one specific embodiment, the spherical fitting reconstruction algorithm includes: Based on the radius of curvature and center coordinates of the spherical geometric parameters, a spherical fitting equation is constructed to obtain the spherical equation parameters; The three-dimensional coordinates of the end face region are calculated based on the spherical equation parameters to obtain spatial coordinate data; The spatial coordinate data is interpolated using a bicubic interpolation method to obtain the 3D topographic data of the end face.

[0055] A multi-dimensional quality assessment algorithm is applied to the end face 3D topography data to generate a test result report and obtain the 3D topography test result of the fiber optic connector end face. The multi-dimensional quality assessment algorithm includes geometric parameter statistics, surface roughness calculation and connection performance prediction.

[0056] In one specific embodiment, the multi-dimensional quality assessment algorithm includes: Based on the 3D topographic data of the end face, the statistical characteristics of geometric parameters such as radius of curvature, vertex offset, and end face angle are calculated to obtain geometric statistical data. The root mean square roughness of the surface height undulation is calculated based on the 3D topography data of the end face to obtain the roughness data; Based on the geometric statistics and roughness data, optical connection loss prediction calculations are performed to obtain the 3D morphology detection results of the fiber optic connector end face.

[0057] Specifically, this embodiment integrates a spherical fitting reconstruction algorithm with a multi-dimensional quality assessment algorithm. It utilizes spherical geometric parameters to construct a spherical fitting equation and performs three-dimensional coordinate calculations and bicubic interpolation surface reconstruction. This is combined with geometric parameter statistical feature calculations and root-mean-square roughness analysis of surface height undulations. Based on geometric statistical data and roughness data, optical connection loss prediction is calculated. This integrated 3D reconstruction and quality assessment mechanism addresses the issues of insufficient accuracy in converting spherical geometric parameters to three-dimensional morphology, the single dimension of surface quality assessment, and the lack of quantitative analysis in connection performance prediction. This improves the completeness and practicality of the 3D morphology inspection results for fiber optic connector end faces.

[0058] In one specific embodiment, the complete process of 3D morphology inspection of the end face of a single-mode fiber optic connector is described: To better illustrate the technical solution of the present invention, a specific embodiment of 3D morphology detection of the end face of a single-mode fiber optic connector is provided below, wherein the single-mode fiber optic connector is of the FC / PC type.

[0059] Setting up the scenario for the implementation example: The test object is a single-mode FC / PC fiber optic connector with a core diameter of 9 micrometers, a cladding diameter of 125 micrometers, and an end-face polishing angle within the range of 0-8 degrees.

[0060] The testing requirements include: testing the spherical geometric parameters according to the IEC 61300-3-16 standard, including the radius of curvature, vertex offset, and fiber height.

[0061] Step 1: Acquisition of a multi-step phase-shifting interferometry image sequence: First, the single-mode FC / PC fiber optic connector is fixed in a precision fixture, and an image acquisition is performed using a white light interferometry system based on the Michelson interferometry principle.

[0062] The automatic positioning process includes: The circular outline of the fiber end face is identified by the circular Hough transform detection algorithm, and the center coordinates of the end face are obtained as (512, 512) pixels. The least squares fitting algorithm was used to detect a tilt angle of 2.3 degrees on the end face, and angle correction was performed. The image grayscale values ​​are adjusted to the optimal dynamic range of 0-255 using a histogram equalization algorithm.

[0063] The phase-shifting interferometric image acquisition process includes: Set a 5-step phase shift acquisition mode with a phase shift step size of π / 2 radians; The movement of the reference mirror is precisely controlled by a piezoelectric ceramic actuator, with each movement distance being 157 nanometers. At each phase shift position, an interference image with a resolution of 1024×1024 pixels is simultaneously acquired by a CCD camera; A multi-step phase-shifting interferometric image sequence containing 5 images was obtained.

[0064] Step 2, Radial gradient sensing core boundary detection: The polar coordinate transformation process includes: processing the acquired multi-step phase-shifting interferometric image sequence using a radial gradient-aware core boundary detection network. First, the image data in Cartesian coordinates is converted to polar coordinates using a polar coordinate convolutional layer. The converted image size is 512 pixels radially × 360 pixels angularly.

[0065] The radial gradient enhancement process includes: calculating the first and second order radial differences using the radial difference operator to obtain radial gradient data. At the core-cladding interface, the first order difference value reaches a maximum of approximately 0.85, and the second order difference value is -0.32, indicating a significant gradient jump characteristic at the interface.

[0066] The process of determining the interface location includes: weighted fusion of radial enhancement features using a multi-scale circular attention mechanism to ultimately determine the core boundary location. The detection results show that the core radius is 4.52 micrometers, with an error of only 0.02 micrometers compared to the standard value of 4.5 micrometers.

[0067] Step 3, dual-coordinate system fusion phase unpacking: The region segmentation process includes: based on determined interface location data, segmenting the multi-step phase-shifting interferometry image sequence into a core region and a cladding region. The core region contains approximately 628 pixels, and the cladding region contains approximately 120,000 pixels.

[0068] The dual-coordinate phase unpacking process includes: applying a polar coordinate phase unpacking algorithm to the core region and a Cartesian coordinate phase unpacking algorithm to the cladding region. The initial phase distribution is calculated using a five-step phase shift algorithm, with the phase values ​​in the core region ranging from [value missing]. arrive radian.

[0069] The error correction process includes: constructing a 128×128 error propagation matrix using a coordinate transformation error compensation algorithm to correct the phase data in the dual coordinate system. The corrected phase data exhibits a 65% reduction in phase jumps at the interface, achieving better phase continuity.

[0070] Step 4, Weighted Reconstruction of Regional Confidence: The confidence assessment process includes: establishing a regional confidence model based on the phase data quality. By calculating the signal-to-noise ratio (SNR) parameters for each region, the SNR of the fiber core region is found to be 18.3 dB, the SNR of the cladding center region is 22.7 dB, while the SNR of the cladding edge region is only 12.1 dB.

[0071] The weighting strategy includes: employing an adaptive weighting algorithm to assign higher weights to high-quality regions. The core region has a weight of 0.425, the cladding center region has a weight of 0.46, and the cladding edge region has a weight of 0.115. Global phase field reconstruction is performed using a weighted least squares optimization method.

[0072] The process of verifying the reconstruction results includes: covering the entire end face region with the reconstructed global phase field data to make the phase distribution smoother and more continuous, thereby reducing the noise level by about 45%.

[0073] Step 5, IEC standard-constrained hierarchical particle swarm optimization: The parameter space layering process includes: according to the IEC 61300-3-16 standard, setting the curvature radius constraint range to 10-25 mm, the vertex offset constraint range to ±1 μm, and the fiber height constraint range to ±0.5 μm. The three-dimensional parameter space is then divided into 8×8×6=384 layered subspaces.

[0074] The particle swarm initialization process includes: generating initial particle positions in each subspace using Latin hypercube sampling, initializing a total of 1200 particles. 800 high-quality particles are then selected as the initial particle swarm through fitness evaluation.

[0075] The optimization iterative process included: after 120 iterations, the particle swarm converged to the optimal solution. The final obtained spherical geometric parameters included: a radius of curvature of 18.7 mm; a vertex offset of +0.23 μm; and an fiber height of -0.08 μm. It should be noted that the fiber height here refers to the relative height difference between the fiber core and cladding and the end face of the connector ceramic sleeve.

[0076] Step 6, 3D topography reconstruction and result generation: The 3D topography reconstruction process includes: generating complete end-face 3D topography data using a spherical fitting reconstruction algorithm based on optimized spherical geometric parameters. The reconstructed 3D topography has a height resolution of 2 nanometers and covers a detection area of ​​140 × 140 micrometers.

[0077] The quality assessment process includes: analysis using a multi-dimensional quality assessment algorithm to obtain the following test results: surface roughness Ra value is 1.8 nanometers; end face angle is 6.2 degrees; geometric parameters meet IEC standard requirements; predicted insertion loss is 0.12 dB; predicted return loss is -58 dB.

[0078] Generate Inspection Report: Automatically generates a complete inspection report including 3D topography images, geometric parameter data tables, quality ratings, and connectivity performance predictions. The entire inspection process takes approximately 45 seconds, with inspection accuracy reaching the nanometer level.

[0079] This embodiment improves the accuracy of phase unpacking and spherical parameter fitting, enhances noise resistance, and increases the detection success rate. It realizes a fully automated detection process from image acquisition to result generation, and the detection results fully comply with IEC international standard requirements.

[0080] The present invention also provides a 3D topography inspection system for fiber optic connector end faces, the system comprising: A multi-step phase-shifting interferometric image acquisition module is used to acquire a multi-step phase-shifting interferometric image sequence from the end face of an optical fiber connector; The radial gradient sensing interface segmentation module is used to segment the interface of the multi-step phase-shifting interferometric image sequence using a fiber core boundary detection network based on radial gradient sensing, and to obtain interface position data. The dual-coordinate system fusion phase unpacking module is used to perform phase unpacking based on the interface position data and the multi-step phase-shifting interferometric image sequence, using a dual-coordinate system fusion phase unpacking algorithm to obtain initial phase data. The regional confidence-weighted reconstruction module is used to globally reconstruct the initial phase data using a regional confidence-weighted reconstruction algorithm to obtain global phase field data. The IEC standard constrained spherical fitting module is used to fit the global phase field data to spherical parameters using the IEC standard constrained hierarchical particle swarm optimization algorithm to obtain spherical geometric parameters. The 3D topography detection result generation module is used to generate 3D topography detection results of the fiber optic connector end face based on the spherical geometric parameters.

[0081] Specifically, this embodiment of a 3D morphology inspection system for fiber optic connector end faces constructs a multi-step phase-shifting interferometric image acquisition module, a radial gradient sensing interface segmentation module, a dual-coordinate system fusion phase unpacking module, a region confidence weighted reconstruction module, an IEC standard-constrained spherical fitting module, and a 3D morphology inspection result generation module. By utilizing the data flow collaborative processing and algorithm optimization complementarity between the modules, and combining the precise interface segmentation of the radial gradient sensing network, the high-precision phase unpacking of the dual-coordinate system fusion, and the optimized fitting of spherical parameters constrained by the IEC standard, and employing a region confidence weighted mechanism for global phase field reconstruction, the system achieves automated high-precision inspection of the 3D morphology of fiber optic connector end faces.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting the 3D morphology of an optical fiber connector end face, characterized in that, Includes the following steps: Acquire a multi-step phase-shifting interferometric image sequence from the end face of an optical fiber connector; A fiber core boundary detection network based on radial gradient sensing is used to segment the interface of the multi-step phase-shifting interferometric image sequence to obtain interface position data; The fiber core boundary detection network includes polar coordinate convolutional layers, a radial gradient enhancement model, and a multi-scale circular attention mechanism. Polar coordinate convolutional layers are used to perform coordinate system transformation and feature extraction on the multi-step phase-shifting interferometric image sequence to obtain polar coordinate feature data; A radial gradient enhancement model is used to perform radial gradient calculation and enhancement processing on the polar coordinate feature data to obtain radially enhanced feature data. The radial gradient enhancement model includes a radial difference operator, a gradient magnitude calculation layer, and an adaptive enhancement layer. A multi-scale circular attention mechanism is used to perform attention weighting and multi-scale fusion on the radial enhancement feature data to obtain the interface position data. Based on the interface position data and the multi-step phase-shifting interferometric image sequence, a dual-coordinate system fusion phase unpacking algorithm is used to perform phase unpacking to obtain initial phase data; The dual-coordinate system fusion phase unpacking algorithm includes coordinate transformation error compensation and phase gradient constraint at the interface; Based on the interface position data, the multi-step phase-shifting interferometric image sequence is segmented to obtain core region image data and cladding region image data. Phase unpacking is performed on the core region image data and the cladding region image data in polar and Cartesian coordinate systems respectively to obtain dual-coordinate system phase data; The phase data of the dual coordinate system is corrected by using coordinate transformation error compensation to obtain corrected phase data; The initial phase data is obtained by performing gradient continuity processing on the corrected phase data using phase gradient constraints at the interface. The initial phase data is globally reconstructed using a region confidence-weighted reconstruction algorithm to obtain global phase field data. The global phase field data were fitted with spherical parameters using the IEC standard constrained hierarchical particle swarm optimization algorithm to obtain the spherical geometric parameters. The 3D morphology detection results of the fiber optic connector end face are generated based on the spherical geometric parameters.

2. The method for 3D morphology detection of the end face of an optical fiber connector as described in claim 1, characterized in that, The polar coordinate convolutional layer includes: The multi-step phase-shifting interferometric image sequence is transformed from Cartesian coordinates to polar coordinates using a coordinate transformation matrix to obtain polar coordinate image data; Radial and angular convolution kernels are used to perform radial and angular convolution operations on the polar coordinate image data to obtain bidirectional convolution features; a feature fusion network is used to fuse the bidirectional convolution features to obtain the polar coordinate feature data. The radial gradient enhancement model includes: The radial gradient data is obtained by performing first-order and second-order difference operations on the polar coordinate feature data in the radial direction using a radial difference operator. A gradient magnitude calculation layer is used to calculate the magnitude and direction of the radial gradient data to obtain gradient magnitude data; An adaptive enhancement layer is used to dynamically adjust the weights and perform nonlinear enhancement on the gradient magnitude data based on the local gradient intensity to obtain the radially enhanced feature data.

3. The method for 3D morphology inspection of the end face of an optical fiber connector as described in claim 1, characterized in that, The coordinate transformation error compensation includes: Establish a transformation error model from polar coordinates to Cartesian coordinates and obtain the error propagation matrix; Interpolation error data is obtained by calculating the interpolation error of the dual-coordinate system phase data based on the error propagation matrix. The corrected phase data is obtained by performing pixel-by-pixel error compensation on the dual-coordinate system phase data based on the interpolation error data. The formula for calculating the error propagation matrix is: ； in, For the error propagation matrix, the first... Line 1 Column elements; For the first Elements of the radial coordinate transformation Jacobian matrix at each radial position; For the first Elements of the Jacobian matrix for angular coordinate transformation at each angular position; These are the second-order error correction coefficients; For phase data in Cartesian coordinates The second-order mixed partial derivative at the point; For the first One polar coordinate radial sampling point, For the first One polar coordinate angle sampling point; For the first Sampling interval at each radial position, For the first Sampling interval at each angular position; This refers to the sampling error weighting coefficient; This is the second numerical stabilization parameter; This is the third numerical stabilization parameter.

4. The method for 3D morphology inspection of the end face of an optical fiber connector as described in claim 1, characterized in that, The region confidence weighted reconstruction algorithm includes a region confidence model and spherical geometric prior constraints; Based on the core region noise characteristics and cladding region noise characteristics of the initial phase data, a regional confidence model is established to obtain confidence weight data; Based on the geometric features of the spherical polishing of the fiber end face, a priori geometric constraints on the spherical surface are established to obtain the geometric constraint parameters. The initial phase data is globally reconstructed using a weighted least squares optimization method, with the confidence weight data as the weight matrix and the geometric constraint parameters as the constraint conditions, to obtain the global phase field data.

5. The method for 3D morphology inspection of the end face of an optical fiber connector as described in claim 4, characterized in that, The regional confidence model includes: The local variances of the core region and cladding region are calculated based on the initial phase data to obtain the region variance data. The signal-to-noise ratio (SNR) parameters for each region are calculated based on the regional variance data to obtain the SNR data; The signal-to-noise ratio data is normalized using an inverse variance weighting method to obtain the confidence weight data.

6. The method for 3D morphology inspection of the end face of an optical fiber connector as described in claim 1, characterized in that, The IEC standard constrained hierarchical particle swarm optimization algorithm includes hierarchical sampling initialization, adaptive search of constraint boundaries, and intelligent solution selection based on connection loss prediction. The global phase field data is initialized by hierarchical sampling based on the constraints of radius of curvature, vertex offset, and fiber height in the IEC standard to obtain initial particle swarm data. The initial particle swarm data is iteratively optimized using constrained boundary adaptive search to obtain a set of candidate spherical parameters; The candidate spherical parameter set is evaluated and screened using a connection loss prediction intelligent solution selection method to obtain the spherical geometric parameters.

7. The method for 3D morphology inspection of the end face of an optical fiber connector as described in claim 6, characterized in that, The hierarchical sampling initialization includes: Based on the constraints of the IEC standard, the spherical geometric parameter space is divided into multiple hierarchical subspaces; Initial particle positions are generated within each hierarchical subspace using the Latin hypercube sampling method, resulting in hierarchical particle position data. Based on the global phase field data, the fitness assessment and density adjustment of the hierarchical particle position data are performed to obtain the initial particle swarm data. The formula for dividing the spherical geometric parameter space into multiple hierarchical subspaces based on the IEC standard constraint range is as follows: ； in, The subspace number to which the spherical geometric parameters belong; These are the geometric parameter values ​​of the sphere to be layered; This represents the minimum constraint range for spherical geometric parameters in the IEC standard. This represents the maximum constraint range for spherical geometric parameters in the IEC standard. This represents the total number of hierarchical subspaces. This is the center offset correction factor; This is the distance attenuation parameter; The center value of the IEC standard constraint range; The width of the constraint range as defined by the IEC standard; It is an exponential function; This is the floor function.

8. A 3D topography inspection system for an optical fiber connector end face, used to perform a 3D topography inspection method for an optical fiber connector end face as described in any one of claims 1-7, characterized in that, The system includes: A multi-step phase-shifting interferometric image acquisition module is used to acquire a multi-step phase-shifting interferometric image sequence from the end face of an optical fiber connector; The radial gradient sensing interface segmentation module is used to segment the interface of the multi-step phase-shifting interferometric image sequence using a fiber core boundary detection network based on radial gradient sensing, and to obtain interface position data. The dual-coordinate system fusion phase unpacking module is used to perform phase unpacking based on the interface position data and the multi-step phase-shifting interferometric image sequence, using a dual-coordinate system fusion phase unpacking algorithm to obtain initial phase data. The regional confidence-weighted reconstruction module is used to perform global reconstruction of the initial phase data using a regional confidence-weighted reconstruction algorithm to obtain global phase field data; The IEC standard constrained spherical fitting module is used to fit the global phase field data to spherical parameters using the IEC standard constrained hierarchical particle swarm optimization algorithm to obtain spherical geometric parameters. The 3D topography detection result generation module is used to generate 3D topography detection results of the fiber optic connector end face based on the spherical geometric parameters.

Images