Automatic calibration method and system for multi-channel cpo light engine of machine vision

Abstract

The application relates to the field of optical technology, in particular to a multi-channel CPO light engine automatic calibration method and system for machine vision, which comprises the following steps: based on vector diffraction theory and a fourth-order ABCD optical path transfer matrix, an opto-mechanical-electric coupling state space model is established; a multi-scale grid point array pattern is projected through a tunable structure grating and a time sequence control light source, and temperature and attitude data are synchronously collected to obtain an observation frame; a sparse phase retrieval and block Fourier frequency domain splicing method is used to realize local complex amplitude reconstruction and splicing, and a space-wavelength-temperature three-dimensional error tensor is generated; an adaptive extended Kalman filter and a Bayesian forgetting coefficient are applied to calculate an optimal compensation vector; and the compensation vector is used to dynamically rewrite a digital-analog conversion lookup table, so that electric, thermal and optical closed loop adjustment is realized. The method provided by the application can realize stable and high-precision automatic calibration of a CPO light engine.

Description

Technical Field

[0001] This application relates to the field of optical technology, specifically to a method and system for automatic calibration of a multi-channel CPO optical engine in machine vision. Background Technology

[0002] With the rapid development of machine vision and intelligent photonics, high-precision, multi-channel optical modulation technology has been widely used in industrial inspection, precision measurement, and automated manufacturing. CPO (Programmable Coherent Optical Engine) optical engines, as a novel integrated optical module, can achieve multi-channel independent programmable modulation of phase, amplitude, polarization, and other parameters. However, current CPO optical engines are susceptible to various factors during long-term operation, such as changes in ambient temperature, device aging, and mechanical disturbances, leading to multi-dimensional and complex dynamic drifts in optical output parameters, which in turn affect the overall measurement accuracy and stability of the system.

[0003] Existing calibration methods are mostly static compensation for a single physical quantity or rely on manual periodic calibration, which makes it difficult to achieve fully automatic, real-time, and high-precision calibration of multi-channel and multi-physical quantity collaborative dynamic errors. They cannot meet the requirements of long-term stability and high-precision output of CPO optical engines in complex application environments.

[0004] In view of this, this application proposes a method and system for automatic calibration of a multi-channel CPO optical engine for machine vision. Summary of the Invention

[0005] To achieve the above objectives, this invention provides a method and system for automatic calibration of a multi-channel CPO optical engine for machine vision, the specific technical solution of which is as follows:

[0006] Automatic calibration methods for multi-channel CPO optical engines in machine vision include:

[0007] Taking the CPO optical engine and electric drive circuit as the object, based on vector diffraction theory and the fourth-order ABCD optical path transfer matrix, the wavefront distribution, output power, attitude angle and temperature of each optical channel are used as state variables to establish a state-space model of opto-mechanical coupling.

[0008] By using a tunable structured grating and a time-controlled light source, a multi-scale lattice pattern is projected across the entire field of view of the CPO light engine, and the temperature and inertial attitude data of the CPO light engine are collected simultaneously to obtain the observation frame corresponding to the state space.

[0009] The sparse phase retrieval algorithm is used to iteratively reconstruct the local complex amplitude of the observation frame, and the block Fourier frequency domain stitching method is used to stitch together each local complex amplitude to obtain a single local complex amplitude image covering the entire field of view, and the space-wavelength-temperature three-dimensional error tensor is calculated and generated.

[0010] The three-dimensional error tensor and the state vector of the previous cycle are input into the adaptive extended Kalman filter. The Bayesian forgetting coefficient decay history covariance is introduced to calculate the compensation vectors of phase, current, polarization and temperature for the next calibration cycle of the CPO optical engine.

[0011] The digital-to-analog conversion lookup table of the CPO optical engine is rewritten based on the compensation vector for electro-thermal-optical closed-loop regulation, and the regulation result is fed back to the opto-mechatronic coupling state-space model.

[0012] Preferably, for the CPO optical engine and its electric drive circuit, an opto-mechanical-electric coupling state-space model is established. The state-space model is based on vector diffraction theory, uses three orthogonal components of the electric field vector to describe the polarization state and propagation characteristics of the light field, and uses a fourth-order ABCD optical path transfer matrix to describe the propagation evolution of the beam inside the CPO optical engine. The state variables include the wavefront distribution, output power, pitch angle, yaw angle, roll angle and temperature of each optical channel.

[0013] Preferably, within the full field of view of the CPO optical engine, a tunable structured grating is used to project a multi-scale grid dot pattern. The tunable structured grating dynamically adjusts the grating period by applying different voltage distributions, and a multi-scale design is used to achieve measurements at multiple scales.

[0014] Preferably, a timing-following encoded sequence is generated by controlling the light source according to the timing sequence, and the sequence is generated by Gray code encoding.

[0015] After the multi-scale grid dot pattern is transmitted through each channel, the light intensity image is acquired and output using a high-speed CMOS camera array. At the same time, temperature and inertial attitude data are acquired synchronously through a temperature sensor array and a six-axis inertial measurement unit, respectively.

[0016] Preferably, the acquired light intensity image is segmented and processed by using an improved Gerchberg-Saxton sparse phase retrieval algorithm with sparse constraints in the frequency domain to iteratively reconstruct the local complex amplitude distribution.

[0017] Preferably, a block-based Fourier frequency domain stitching method is used to globally stitch together each local complex amplitude, and the phase ambiguity of adjacent regions is solved by the least squares method combined with the conjugate gradient method, so as to seamlessly stitch together the local complex amplitudes of the entire field of view.

[0018] A spatial-wavelength-temperature three-dimensional error tensor is constructed based on the stitched full-field local complex amplitude image. The three-dimensional error tensor is obtained through multi-wavelength decomposition and temperature field interpolation.

[0019] Preferably, the three-dimensional error tensor is reduced in dimensionality by Tucker tensor decomposition to extract the error modes;

[0020] The dimensionality-reduced features are input together with the state vector of the previous period into an adaptive extended Kalman filter. The state equation of the filter defines an augmented state vector, which includes compensation vectors for the phase, current, polarization, and temperature of each optical channel.

[0021] Preferably, a Bayesian forgetting coefficient based on the observation information covariance is introduced into the adaptive extended Kalman filter to dynamically adjust the historical covariance weights, and physical constraints are applied to the compensation vector to limit the compensation value to a safe range.

[0022] Preferably, the digital-to-analog conversion lookup table of the CPO optical engine is dynamically rewritten according to the compensation vector;

[0023] The digital-to-analog conversion lookup table adopts a four-dimensional parameter quantization design of phase, current, polarization and temperature. The compensation vector is extended into a continuous compensation surface through cubic spline interpolation. Incremental and progressive lookup table update strategies are adopted. Combined with PID closed-loop control, dynamic adaptive adjustment of the electro-thermal-optical multi-loop is realized. The adjustment results are fed back to the state space model in real time for online parameter updates.

[0024] A machine vision multi-channel CPO optical engine automatic calibration system, which is used to implement the machine vision multi-channel CPO optical engine automatic calibration method, includes: an opto-electro-mechanical coupling module, an observation frame acquisition module, an error calculation module, a compensation calibration module, and a calibration implementation module.

[0025] The opto-mechatronics coupling module takes the CPO optical engine and the electric drive circuit as objects, and establishes a state space model of opto-mechatronics coupling based on vector diffraction theory and the fourth-order ABCD optical path transfer matrix, taking the wavefront distribution, output power, attitude angle and temperature of each optical channel as state variables.

[0026] The observation frame acquisition module projects a multi-scale grid dot pattern across the entire field of view of the CPO optical engine using a tunable structured grating and a time-controlled light source, and simultaneously acquires the temperature and inertial attitude data of the CPO optical engine to obtain the observation frame corresponding to the state space.

[0027] The error calculation module uses a sparse phase retrieval algorithm to iteratively reconstruct the local complex amplitude of the observation frame, and uses a block Fourier frequency domain stitching method to stitch together each local complex amplitude to obtain a single local complex amplitude image covering the entire field of view, and calculates and generates a three-dimensional error tensor of space-wavelength-temperature.

[0028] The compensation calibration module inputs the three-dimensional error tensor and the state vector of the previous cycle into the adaptive extended Kalman filter, introduces the Bayesian forgetting coefficient decay history covariance, and calculates the compensation vectors of phase, current, polarization and temperature of the CPO optical engine in the next calibration cycle.

[0029] The calibration implementation module rewrites the analog-to-digital conversion lookup table of the CPO optical engine based on the compensation vector for electro-thermal-optical closed-loop adjustment, and feeds the adjustment result back to the opto-mechanical-electrical coupling state-space model.

[0030] The beneficial effects of this invention are as follows: This application establishes a state-space model of opto-electro-mechanical coupling based on vector diffraction theory and a fourth-order ABCD optical path transfer matrix, which can comprehensively reflect the dynamic characteristics of the optical channel.

[0031] This application utilizes a tunable structured grating and a time-controlled light source to achieve multi-scale, high-resolution spatial sampling, simultaneously acquiring temperature and inertial attitude data, providing rich and complete observational information for error tracing and multi-dimensional calibration.

[0032] This application employs sparse phase retrieval and block Fourier frequency domain stitching methods, which can efficiently reconstruct complex amplitude images of the entire field of view and quantitatively characterize the three-dimensional error distribution of space, wavelength, and temperature, providing an accurate and low-noise data foundation for subsequent compensation algorithms.

[0033] This application effectively reduces computational complexity and improves real-time performance by using tensor decomposition for dimensionality reduction and an adaptive extended Kalman filter. It also utilizes the Bayesian forgetting coefficient to dynamically adapt to system changes, ensuring the accuracy and robustness of the compensation vector estimation.

[0034] This application dynamically rewrites the digital-to-analog conversion lookup table and introduces a closed-loop adjustment and feedback mechanism to achieve seamless updating of compensation parameters and adaptive correction of the system, ensuring long-term stable operation and high-precision intelligent calibration of the CPO optical engine.

[0035] This application realizes fully automated high-precision calibration of CPO optical engines with multiple channels and multiple physical quantities, reduces manual intervention and can operate reliably for a long time, and is widely applicable to high-end machine vision and precision optics applications. Attached Figure Description

[0036] Figure 1 Flowchart of the automatic calibration method for a multi-channel CPO optical engine in machine vision provided by the present invention;

[0037] Figure 2 A flowchart illustrating the state-space modeling process for the opto-mechatronic coupling provided by this invention;

[0038] Figure 3 This is a flowchart of the multimodal observation frame acquisition process provided by the present invention;

[0039] Figure 4 The flowchart for complex amplitude reconstruction and three-dimensional error tensor generation provided by this invention;

[0040] Figure 5 The flowchart for calculating calibration compensation parameters provided by this invention;

[0041] Figure 6 The calibration closed-loop execution and feedback update flowchart provided by this invention;

[0042] Figure 7 The structural diagram of the machine vision multi-channel CPO optical engine automatic calibration system provided by the present invention. Detailed Implementation

[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0044] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention can also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0046] Example 1:

[0047] Reference Figures 1 to 6 This is the first embodiment of the present invention, which provides an automatic calibration method for a multi-channel CPO optical engine in machine vision.

[0048] CPO (Cyclic Optical Probe) engines are integrated optical modules that programmably control the characteristics of optical waves. These engines typically consist of a coherent light source, an optical phase modulator, a drive circuit, and related control units, enabling precise programming and dynamic adjustment of parameters such as the phase, amplitude, and polarization of light. CPO engines are widely used in optical communication, wavefront shaping, spatial light modulation, laser processing, and adaptive optics, achieving multi-channel, high-resolution optical signal control. Their main characteristics are high programmability, fast response, and support for real-time modulation of complex optical fields, making them core components in modern intelligent photonics systems and integrated optical platforms.

[0049] Step 1: Taking the CPO optical engine and electric drive circuit as the object, based on vector diffraction theory and the fourth-order ABCD optical path transfer matrix, the wavefront distribution, output power, attitude angle, and temperature of each optical channel are used as state variables to establish a state-space model of opto-mechanical coupling. (See [link to relevant documentation]). Figure 2 The flowchart for modeling the state-space model of opto-electro-mechanical coupling in this step is as follows.

[0050] For the CPO optical engine and its electric drive circuit, an accurate opto-mechanical-electric coupling state-space model is established. The modeling process of the opto-mechanical-electric coupling state-space model is based on vector diffraction theory, fully considering the vector characteristics of light waves. Three orthogonal components of the electric field vector in space are introduced to describe the polarization state and propagation characteristics of the light field. The optical transmission matrix is ​​constructed using the fourth-order ABCD optical path transfer matrix method. :

[0051]

[0052] in, , , , The transmission parameters of the optical transmission matrix represent the lateral magnification, propagation distance effect, optical element curvature effect, and angular magnification, respectively. The ABCD optical path transmission matrix can accurately describe the propagation evolution of the beam within the CPO optical engine. Compared to traditional second-order matrices, the ABCD optical path transmission matrix can simultaneously handle beam propagation in both the meridional and sagittal planes, improving the accuracy of the opto-mechanical-electric coupling state-space model in describing non-axisymmetric optics.

[0053] Selecting the state variables of the opto-mechatronics coupling state-space model, the wavefront distribution of each optical channel is... Output power Posture angle , These represent pitch angle, yaw angle, roll angle, and temperature, respectively. As state variables in the state-space model of opto-mechatronic coupling, a vector of state variables is constructed. Represented as:

[0054]

[0055] in This represents the total number of optical channels in the CPO optical engine. The constructed state variable vector can comprehensively reflect the working state of the light engine, providing a complete state information foundation for subsequent adaptive calibration.

[0056] Based on the established state variables, a discrete-time state-space model of opto-mechatronic coupling is established:

[0057]

[0058]

[0059] in, , They are respectively time, The vector of state variables at each moment. To control the input vector, including drive current and temperature control parameters, For the observation vector, It is a nonlinear state transition function. For the observation function, and The process noise and observation noise are respectively assumed to be zero-mean Gaussian white noise.

[0060] Determining the nonlinear state transition function using opto-mechatronic coupling equations Specific form:

[0061]

[0062] in, This is wavefront distortion caused by temperature. It is the first Wavefront distribution at time, It is the first Output power at any moment Let be the electro-optical conversion efficiency function. For driving current, It is the first The posture and angle at any moment It is the angular velocity vector. It is the first Temperature at any moment The time interval between adjacent sampling times. For heat capacity, and These are the input and output thermal powers, respectively.

[0063] The opto-mechanical-electric coupling state-space model constructed in this step can accurately describe the impact of temperature changes on optical performance and realize thermal-optical coupling analysis. On the other hand, by introducing the relationship between electric drive parameters and optical output, it realizes electro-optical coupling modeling, which can be used to realize closed-loop control and adjustment of the CPO optical engine.

[0064] Step 2: Using a tunable structured grating and a time-controlled light source, project a multi-scale lattice pattern across the entire field of view of the CPO light engine, and simultaneously acquire temperature and inertial attitude data of the CPO light engine to obtain the corresponding observation frames in the state space. (See attached image) Figure 3 This is a flowchart of the multimodal observation frame acquisition process for this step.

[0065] Structured light projection and multimodal data acquisition are performed across the entire field of view of the CPO optical engine. A tunable structured grating based on a liquid crystal spatial light modulator (LC-SLM) is employed. The tunable structured grating achieves dynamic adjustment of its period by applying different voltage distributions, thereby modulating the phase function of the grating. Set as:

[0066]

[0067] in, For example, the grating scale level is set to 5 levels to cover the spatial frequency range from large to small. For the first The modulation depth of the first-order grating has a range of values. ; For the first The grating period is according to The regular increase, where the basic cycle ; and They are respectively and The initial phase of the direction is used to achieve spatial shifting of the grating pattern. Multi-scale grating design can simultaneously achieve large-scale coarse positioning and local fine measurement, improving the efficiency and accuracy of spatial sampling.

[0068] The timing-controlled light source is implemented using a high-speed LED array, and its emission timing follows an encoded sequence: ;in, Represents the light emission timing coding sequence, Indicates time, For example, the reference light intensity is set to [value]. ; For the first Binary encoding of a time slice, ; This represents the total number of time slices. The duration of a single time slice; This is a rectangular window function. It is generated using the Gray code encoding scheme. The sequence ensures that adjacent codewords differ by only one bit, avoiding measurement errors caused by timing switching. The employed timing-coded illumination maintains high temporal resolution while also improving the signal-to-noise ratio through coding gain, enabling information multiplexing in the temporal dimension.

[0069] After the structured light is transmitted through the various optical channels of the CPO optical engine, it forms a multi-scale lattice pattern on the detection plane. A high-speed CMOS camera array is used to capture the projected pattern. The camera is positioned at the output end face of the CPO optical engine, and the captured light intensity image is obtained. Represented as:

[0070]

[0071] in, for The first moment Electric field distribution of each optical channel output; For the first Point spread function of each optical channel; The spatial response function of the camera; Represents a two-dimensional convolution operation; This represents the total number of optical channels in the CPO optical engine.

[0072] During the projection process, a high-precision temperature sensor array and a six-axis inertial measurement unit (IMU) are used to collect temperature and inertial attitude data; the temperature sensors use... Temperature data is collected in an array distributed within the CPO optical engine.

[0073] Inertial attitude data is acquired through an integrated MEMS-IMU, which includes a three-axis accelerometer and a three-axis gyroscope. The attitude angles of the CPO optical engine are calculated using a complementary filtering algorithm.

[0074]

[0075] in, Let them be Euler angle vectors. These represent roll angle, pitch angle, and yaw angle, respectively. The angular velocity measured by the gyroscope; Attitude angles calculated for the accelerometer; These are complementary filter coefficients, preferably... ; The sampling interval is preferably defined as follows: The complementary filtering algorithm for calculating the attitude angle of the CPO optical engine combines the short-term accuracy of the gyroscope with the long-term stability of the accelerometer, effectively suppressing attitude drift during attitude angle calculation.

[0076] The collected data is used to assemble the data into the observation frame: ;in, For the first The light intensity image of the frame; This is the reading vector of the temperature sensor; Six-degree-of-freedom attitude data measured by the IMU; For timestamps.

[0077] This step not only obtained complete spatial distribution information of the output light field of the CPO optical engine, but also simultaneously recorded key physical parameters affecting optical performance, including temperature and attitude data, providing complete observation data for the established state-space model, enabling subsequent accurate separation and compensation of the impact of various error sources on the CPO optical engine.

[0078] Step 3: The sparse phase retrieval algorithm is used to iteratively reconstruct the local complex amplitude of the observed frame. A block-based Fourier frequency domain stitching method is then used to stitch together each local complex amplitude to obtain a single local complex amplitude image covering the entire field of view. Based on this local complex amplitude image, a three-dimensional error tensor of space-wavelength-temperature is calculated. (See [link / reference]). Figure 4 This is a flowchart of the complex amplitude reconstruction and three-dimensional error tensor generation process in this step.

[0079] The obtained observation frames are subjected to complex amplitude reconstruction processing based on sparse phase retrieval. First, the light intensity image in the observation frames is... Divided into A local region, for example, in which The size of each local region is Pixels, settings between adjacent areas The overlap ratio is adjusted to ensure the continuity of subsequent stitching. Block processing can effectively reduce the amount of data processed in a single operation and improve computational efficiency.

[0080] For the A local area, An improved Gerchberg-Saxton sparse phase retrieval algorithm is used for iterative reconstruction of complex amplitudes.

[0081]

[0082] in, For the first The local region in the first The complex amplitude distribution of the next iteration; The amplitude information is extracted from the measured light intensity. , Indicates the first Measured light intensity distribution in a local area; For the first Phase distribution of the next iteration; and These represent the two-dimensional Fourier transform and inverse Fourier transform operations, respectively; It is the imaginary unit; It is an exponential function; For the sparse constraint mask function, the sparse constraint mask function is defined as follows:

[0083]

[0084] in, Frequency domain spatial coordinates; It is the spectral distribution function; For sparse threshold coefficients, The value is selected based on the optimized spectral characteristics of the structured grating, preferably... .

[0085] In the iterative process of the improved Gerchberg-Saxton sparse phase retrieval algorithm, sparse constraints are applied in the frequency domain to retain the main spectral components while filtering out noise, thereby improving the stability of phase reconstruction. The iterative convergence condition is set as the root mean square error between two adjacent iterations being less than [value missing]. The sparse constraint utilizes the sparsity of the structured grating projection pattern in the frequency domain to accelerate algorithm convergence and improve phase reconstruction accuracy, showing a significant improvement in convergence speed compared to traditional phase retrieval algorithms.

[0086] After reconstructing the local complex amplitude, the block Fourier frequency domain stitching method is used to stitch the local complex amplitudes together globally. During the stitching process, it is necessary to solve the phase ambiguity problem between adjacent regions.

[0087] Define adjacent regions and In overlapping areas phase difference The calculation formula is:

[0088]

[0089] in, Let be the phase offset to be solved. For local areas Phase distribution, For local areas The phase distribution.

[0090] The global phase offset set is solved using the least squares method. Construct the splicing optimization objective function :

[0091]

[0092] in, Indicates the area The set of four neighboring regions; The splicing weight coefficient is dynamically determined based on the cross-correlation coefficient of the overlapping areas. The higher the cross-correlation coefficient, the greater the weight, thus ensuring the splicing quality.

[0093] The splicing optimization problem is solved using the conjugate gradient method to obtain a globally consistent local complex amplitude distribution. ,in and These represent the amplitude and phase distributions across the entire field of view, respectively. The frequency domain stitching method avoids the boundary discontinuities caused by direct spatial domain stitching, ensuring the phase continuity and high accuracy of the local complex amplitudes across the entire field of view.

[0094] Based on the stitched full-field local complex amplitude image, a three-dimensional error tensor of space-wavelength-temperature is constructed. The three-dimensional error tensor of space-wavelength-temperature can comprehensively characterize the error characteristics of the CPO optical engine.

[0095] Multi-wavelength decomposition of local complex amplitudes is performed, considering the operation of the CPO optical engine. band ( ),by For interval division Each wavelength channel. Calculate the corresponding wavefront distortion distribution :

[0096]

[0097] in, This represents the change in refractive index caused by temperature. for; The reference refractive index for optical waveguide materials; The thermo-optic coefficient of the material; This is the measured temperature value; For reference temperature; The effective optical path length; the wavefront distortion distribution reflects the physical mechanism by which temperature changes affect the optical phase by altering the material's refractive index.

[0098] Constructing a three-dimensional error tensor The three-dimensional error tensor comprehensively considers the coupled effects of spatial location, operating wavelength, and temperature.

[0099]

[0100] in, This represents the ideal wavefront distribution under calibration conditions. Phase modulation transfer function caused by temperature field ; This is the temperature-phase conversion coefficient. ; The temperature field distribution is two-dimensional, obtained by bicubic spline interpolation of discrete temperature sensor data, ensuring the spatial continuity of the temperature field.

[0101] Obtain the three-dimensional error tensor ,in The spatial dimension represents the error distribution across the entire field of view; In terms of wavelength, it covers the operating band of the CPO optical engine; This represents the temperature sampling dimension, corresponding to the number of measurement points in the temperature sensor array.

[0102] This step comprehensively quantifies the wavefront distortion characteristics of the CPO optical engine under different spatial locations, operating wavelengths, and temperature conditions, realizing a unified characterization of multi-dimensional error information. It provides complete error distribution information for subsequent adaptive compensation and can support precise decoupling and quantitative compensation for various error sources.

[0103] Step 4: Input the three-dimensional error tensor and the state vector of the previous cycle into the adaptive extended Kalman filter, introduce the Bayesian forgetting coefficient to attenuate the history covariance, and calculate the compensation vectors for phase, current, polarization, and temperature of the CPO optical engine in the next calibration cycle. See [link / reference] Figure 5 This is a flowchart of the compensation parameter calculation process for this step.

[0104] The generated three-dimensional error tensor The state vector from the previous calibration cycle is input to an adaptive extended Kalman filter (AEKF) for optimal state estimation and compensation vector calculation.

[0105] The dimensionality of the three-dimensional error tensor is reduced, and the error modes are extracted using tensor decomposition. The three-dimensional error tensor is then represented using Tucker decomposition as follows: ;in, This represents the product of the first modulus. The core tensor contains the main characteristics of the error; , and These are orthogonal mode matrices representing the spatial, wavelength, and temperature dimensions, respectively. , , The number of principal components retained for each dimension. By dimensionality reduction, the high-dimensional error tensor is compressed into a low-dimensional feature representation, which can significantly reduce the computational complexity of subsequent filters; tensor decomposition, while retaining the main features of the error, can also reduce the computational load by about two orders of magnitude, which can be used to realize real-time calibration of the CPO optical engine.

[0106] Construct the state equation and observation equation of the adaptive extended Kalman filter, and define the augmented state vector of the state equation and observation equation as follows: ;in, Here is the phase compensation vector for each channel. Number of channels; This is the drive current compensation vector; This is the polarization compensation vector, which includes TE mode compensation and TM mode compensation for each channel. TE mode compensation refers to compensation for transverse electric polarization, while TM mode compensation is compensation for transverse magnetic polarization. This is the temperature compensation vector. This represents the number of temperature control points.

[0107] The nonlinear state transition equation of the adaptive extended Kalman filter is:

[0108]

[0109] in, This is the state transition function. It is a linear state transition matrix. This represents the state vector at time k; This is a nonlinear term that describes the nonlinear effects of thermo-optical coupling and electro-optical modulation. This is process noise.

[0110] The observation equation for the adaptive extended Kalman filter is:

[0111]

[0112] in, For the observation equation, The observation vector is obtained by vectorizing the core tensor of the error tensor; The observation matrix; This represents the tensor vectorization operation; This is to address observation noise. The beneficial effect of this state-space modeling is that it establishes a clear mathematical mapping relationship between multidimensional error information and compensation parameters, laying the foundation for optimal estimation.

[0113] Based on the extended Kalman filter, a Bayesian forgetting coefficient is introduced to adaptively adjust the weights of the historical covariance. The Bayesian forgetting coefficient is defined as:

[0114]

[0115] in, and These are the lower and upper bounds of the Bayesian forgetting coefficient, respectively. This is the decay rate parameter; For the information covariance matrix, ; Represents the trace of the matrix; when the Bayesian forgetting coefficient changes drastically ( (relatively large), the forgetting coefficient approaches This increases the dependence on new measurements; when the Bayesian forgetting coefficient stabilizes, the forgetting coefficient approaches [the value]. It will retain more historical information; the updated prediction covariance matrix The calculation is as follows:

[0116]

[0117] in, Let be the posterior covariance matrix of the previous time step.

[0118] The optimal compensation vector is calculated through Kalman gain update and state estimation:

[0119]

[0120]

[0121]

[0122] in, The Kalman gain at time k; Let k be the posterior state estimate at time k. The prior state estimate at time k is the required compensation vector; It is an identity matrix.

[0123] Extracting components from the compensation vector: Phase compensation , This refers to the phase compensation vector required for all optical channels (such as each channel in an array-type optical engine) after Kalman filtering optimization; current compensation. Polarization compensation Temperature compensation .

[0124] To ensure the stability and feasibility of the compensation, constraints are applied to the compensation vector:

[0125]

[0126] in, It is a saturation function; and These are the physical constraint boundaries for each compensation vector; for example, phase compensation is constrained to... Within the specified range, current compensation is limited to the safe operating range of the driver. This constraint process prevents damage to the CPO optical engine from abnormal compensation values, ensuring the safety and robustness of the calibration process.

[0127] This step, by introducing tensor decomposition of the three-dimensional error tensor for dimensionality reduction and an adaptive extended Kalman filter algorithm, achieves efficient extraction of high-dimensional error information and optimal estimation of multi-channel compensation parameters, significantly improving the real-time performance and automatic calibration accuracy of the CPO optical engine. Tensor dimensionality reduction greatly reduces computational complexity, while the adaptive Kalman filter and Bayesian forgetting coefficients intelligently adjust the weights of historical data based on dynamic data changes, improving the CPO optical engine's responsiveness to environmental changes. Furthermore, by setting physical constraints on the compensation vector, the safety and robustness of the CPO optical engine calibration process are ensured.

[0128] Step 5: Rewrite the digital-to-analog conversion lookup table of the CPO optical engine based on the compensation vector for electro-thermal-optical closed-loop regulation, and feed the regulation results back to the opto-mechatronic coupling state-space model. (See [link]) Figure 6 This is a flowchart of the calibration closed-loop execution and feedback update process for this step.

[0129] Based on the calculated compensation vector, the digital-to-analog converter (DAC) lookup table of the CPO optical engine is dynamically updated and rewritten. The DAC lookup table employs a four-dimensional data structure, corresponding to four control dimensions: phase, current, polarization, and temperature. For example, the phase dimension is quantized to 256 levels, covering the complete phase range from 0 to 2π; the current dimension is quantized to 1024 levels, precisely controlling the drive current from the threshold to saturation; the polarization dimension is quantized to 64 levels, enabling fine-tuning of TE and TM modes; and the temperature dimension is quantized to 128 levels, covering the operating temperature range. The DAC lookup table is updated incrementally, by adding compensation increments to the original DAC lookup table, avoiding functional interruptions caused by a full table rewrite.

[0130] During the update process of the analog-to-digital conversion lookup table, polynomial interpolation is used to smoothly extend the discrete compensation vector values ​​to the entire parameter space. Using the sampling points in the compensation vector as control points, a continuous compensation surface is generated through cubic spline interpolation. Interpolation ensures a smooth transition between adjacent lookup table entries, preventing instability caused by abrupt changes in control parameters. A progressive update strategy is introduced; for example, the new analog-to-digital conversion lookup table and the old analog-to-digital conversion lookup table are weighted and merged at a ratio of 3:7, allowing the control parameters to gradually transition to the new operating point. This progressive update strategy prevents oscillations caused by abrupt compensation changes, ensuring the robustness of the calibration process.

[0131] When implementing closed-loop regulation and control of the electro-thermal-optical system, three mutually coupled but independently operating control loops are constructed. Regulation and control of the electrical, thermal, and optical components are achieved by controlling the drive current of each loop. A proportional-integral-derivative (PID) controller is employed, which adjusts the driver output in real time based on the current compensation value provided by the digital-to-analog converter lookup table.

[0132] A feedback mechanism for the adjustment results is established, feeding the control effect back to the established opto-electro-mechanical coupling state-space model in real time. The state-space model updates its parameters online based on the feedback information, including updating the coupling coefficients and process noise covariance in the state transition matrix. The updated opto-electro-mechanical coupling state-space model is used for state prediction in the next calibration cycle, forming a complete adaptive closed loop. By calibrating the CPO optical engine, the calibrated CPO optical engine can automatically adapt to long-term drift factors such as device aging and environmental changes, maintaining long-term stable performance requirements.

[0133] This step achieves fully automatic adaptive calibration of the CPO optical engine through the combined effects of digital-to-analog conversion lookup table rewriting, electro-thermal-optical closed-loop adjustment, and state feedback updates. The closed-loop calibration realizes a fully automated process from error detection and compensation calculation to execution control, maintaining the optimal working state of the CPO optical engine without manual intervention, which significantly improves the measurement accuracy and reliability of the machine vision system.

[0134] Example 2:

[0135] Reference Figure 2 The second embodiment of the present invention provides a machine vision-based multi-channel CPO optical engine automatic calibration system.

[0136] The system includes: an opto-electro-mechanical coupling module, an observation frame acquisition module, an error calculation module, a compensation and calibration module, and a calibration implementation module.

[0137] The opto-mechatronics coupling module takes the CPO optical engine and the electric drive circuit as objects, and establishes a state-space model of opto-mechatronics coupling based on vector diffraction theory and the fourth-order ABCD optical path transfer matrix, taking the wavefront distribution, output power, attitude angle and temperature of each optical channel as state variables.

[0138] The observation frame acquisition module projects a multi-scale grating dot pattern across the entire field of view of the CPO optical engine using a tunable structured grating and a time-controlled light source, and simultaneously acquires the temperature and inertial attitude data of the CPO optical engine to obtain the observation frame corresponding to the state space.

[0139] The error calculation module uses a sparse phase retrieval algorithm to iteratively reconstruct the local complex amplitude of the observation frame, and uses a block Fourier frequency domain stitching method to stitch together each local complex amplitude to obtain a single local complex amplitude image covering the entire field of view, and calculates and generates a three-dimensional error tensor of space-wavelength-temperature.

[0140] The compensation calibration module inputs the three-dimensional error tensor and the state vector of the previous cycle into the adaptive extended Kalman filter, introduces the Bayesian forgetting coefficient decay history covariance, and calculates the compensation vectors for phase, current, polarization and temperature of the CPO optical engine in the next calibration cycle.

[0141] The calibration implementation module rewrites the analog-to-digital conversion lookup table of the CPO optical engine based on the compensation vector for electro-thermal-optical closed-loop adjustment, and feeds the adjustment result back to the opto-mechanical-electrical coupling state-space model.

[0142] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0143] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. An automatic calibration method for a multi-channel CPO optical engine in machine vision, characterized in that, include: Step 1: Taking the CPO optical engine and electric drive circuit as the object, based on vector diffraction theory and the fourth-order ABCD optical path transfer matrix, the wavefront distribution, output power, attitude angle and temperature of each optical channel are used as state variables to establish a state-space model of opto-mechanical coupling. Step 2: Using a tunable structured grating and a time-controlled light source, project a multi-scale grid pattern across the entire field of view of the CPO light engine, and simultaneously collect temperature and inertial attitude data of the CPO light engine to obtain the observation frame corresponding to the state space. Step 3: The sparse phase retrieval algorithm is used to iteratively reconstruct the local complex amplitude of the observation frame. The block Fourier frequency domain stitching method is used to stitch together each local complex amplitude to obtain a single local complex amplitude image covering the entire field of view. The three-dimensional error tensor of space-wavelength-temperature is calculated and generated. Step 4: Input the three-dimensional error tensor and the state vector of the previous cycle into the adaptive extended Kalman filter, introduce the Bayesian forgetting coefficient decay history covariance, and calculate the compensation vectors of phase, current, polarization and temperature for the next calibration cycle of the CPO optical engine. Step 5: Rewrite the digital-to-analog conversion lookup table of the CPO optical engine based on the compensation vector for electro-thermal-optical closed-loop regulation, and feed the regulation results back to the opto-mechatronic coupling state space model.

2. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 1, characterized in that, For the CPO optical engine and its electric drive circuit, a state-space model of opto-mechanical-electric coupling is established. The state-space model is based on vector diffraction theory, uses three orthogonal components of the electric field vector to describe the polarization state and propagation characteristics of the light field, and uses a fourth-order ABCD optical path transfer matrix to describe the propagation evolution of the beam inside the CPO optical engine. The state variables include the wavefront distribution, output power, pitch angle, yaw angle, roll angle and temperature of each optical channel.

3. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 2, characterized in that, Within the full field of view of the CPO optical engine, a tunable structured grating is used to project a multi-scale grid dot pattern. The tunable structured grating dynamically adjusts the grating period by applying different voltage distributions, and a multi-scale design is used to achieve measurements at multiple scales.

4. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 3, characterized in that, A timing-following encoded sequence is generated by controlling the light source in a timing sequence, and the sequence is generated by Gray code encoding. After the multi-scale grid dot pattern is transmitted through each optical channel, the light intensity image is acquired and output using a high-speed CMOS camera array. At the same time, temperature and inertial attitude data are acquired synchronously through a temperature sensor array and a six-axis inertial measurement unit, respectively.

5. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 4, characterized in that, The acquired light intensity images are segmented and processed. An improved Gerchberg-Saxton sparse phase retrieval algorithm is used to apply sparse constraints in the frequency domain to iteratively reconstruct the local complex amplitude distribution.

6. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 5, characterized in that, A block-based Fourier frequency domain stitching method is used to globally stitch together each local complex amplitude. The phase ambiguity of adjacent regions is solved by the least squares method combined with the conjugate gradient method, and the local complex amplitudes of the entire field of view are seamlessly stitched together. A spatial-wavelength-temperature three-dimensional error tensor is constructed based on the stitched full-field local complex amplitude image. The three-dimensional error tensor is obtained through multi-wavelength decomposition and temperature field interpolation.

7. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 6, characterized in that, Tucker tensor decomposition is used to reduce the dimensionality of the 3D error tensor and extract the error modes. The dimensionality-reduced features are input together with the state vector of the previous cycle into an adaptive extended Kalman filter. The state equation of the filter defines an augmented state vector, which includes compensation vectors for the phase, current, polarization, and temperature of each channel.

8. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 7, characterized in that, In the adaptive extended Kalman filter, a Bayesian forgetting coefficient based on the covariance of observation information is introduced to dynamically adjust the weight of historical covariance, and physical constraints are applied to the compensation vector to limit the compensation value to a safe range.

9. The automatic calibration method for a multi-channel CPO optical engine in machine vision according to claim 8, characterized in that, The digital-to-analog conversion lookup table of the CPO optical engine is dynamically rewritten based on the compensation vector; The digital-to-analog conversion lookup table adopts a four-dimensional parameter quantization design of phase, current, polarization and temperature. The compensation vector is extended into a continuous compensation surface through cubic spline interpolation. Incremental and progressive lookup table update strategies are adopted. Combined with PID closed-loop control, dynamic adaptive adjustment of the electro-thermal-optical multi-loop is realized. The adjustment results are fed back to the state space model in real time for online parameter updates.

10. A machine vision multi-channel CPO optical engine automatic calibration system, used to implement the machine vision multi-channel CPO optical engine automatic calibration method according to any one of claims 1 to 9, characterized in that, include: Opto-electro-mechanical coupling module, observation frame acquisition module, error calculation module, compensation and calibration module, and calibration implementation module; The opto-mechatronics coupling module takes the CPO optical engine and the electric drive circuit as objects, and establishes a state space model of opto-mechatronics coupling based on vector diffraction theory and the fourth-order ABCD optical path transfer matrix, taking the wavefront distribution, output power, attitude angle and temperature of each optical channel as state variables. The observation frame acquisition module projects a multi-scale grid dot pattern across the entire field of view of the CPO optical engine using a tunable structured grating and a time-controlled light source, and simultaneously acquires the temperature and inertial attitude data of the CPO optical engine to obtain the observation frame corresponding to the state space. The error calculation module uses a sparse phase retrieval algorithm to iteratively reconstruct the local complex amplitude of the observation frame, and uses a block Fourier frequency domain stitching method to stitch together each local complex amplitude to obtain a single local complex amplitude image covering the entire field of view, and calculates and generates a three-dimensional error tensor of space-wavelength-temperature. The compensation calibration module inputs the three-dimensional error tensor and the state vector of the previous cycle into the adaptive extended Kalman filter, introduces the Bayesian forgetting coefficient decay history covariance, and calculates the compensation vectors of phase, current, polarization and temperature of the CPO optical engine in the next calibration cycle. The calibration implementation module rewrites the analog-to-digital conversion lookup table of the CPO optical engine based on the compensation vector for electro-thermal-optical closed-loop adjustment, and feeds the adjustment result back to the opto-mechanical-electrical coupling state-space model.

Images