A four-quadrant detector residual calibration method for laser deflection angle detection

Abstract

The application discloses a four-quadrant detector residual calibration method for laser deflection angle detection and relates to the technical field of laser detection. The output currents of four quadrants of a current four-quadrant detector target surface are input into a pre-calibrated front-stage solver to obtain an initial deflection angle of a laser beam relative to the four-quadrant detector; normalized differential features, diagonal differential features and power scale features are calculated to form a response feature vector; according to the normalized response radius and the response direction angle of a laser spot on the target surface of the four-quadrant detector, the partition number of the spot on the target surface is determined; according to the environmental interference parameters of the current four-quadrant detector and the physical parameters of the laser beam, the working condition index corresponding to the spot is determined; and the response feature vector is input into a local residual model corresponding to the partition number and the working condition index to obtain a residual estimate. The method can accurately estimate the local system error of the front-stage solver which has not been eliminated in the real system, so that the precision of the deflection angle is improved.

Description

Technical Field

[0001] This application relates to the field of laser detection technology, and in particular to a method for residual calibration of a four-quadrant detector for laser deflection angle detection. Background Technology

[0002] Space laser communication, laser guidance, and precision beam alignment systems all require high-precision detection of minute deflection angles of laser beams. Laser beams have small divergence angles and strong directionality, making them susceptible to factors such as platform vibration, atmospheric disturbances, optical path assembly errors, and background noise during propagation and acquisition, leading to minute shifts or morphological distortions of the beam spot on the detection surface.

[0003] Four-quadrant detectors are commonly used detection elements in the detection of laser beam position and deflection angle due to their fast response speed, high sensitivity, and simple signal processing structure. The basic idea is to acquire the current or voltage output from the four quadrants and calculate the deflection angle using a pre-stage solver such as the difference ratio method.

[0004] Under ideal conditions, the light spot approximates a centrally symmetric Gaussian distribution, with no dead zones between the four quadrants and consistent gain across all channels. In this case, the difference ratio output and the light spot displacement have a good correlation within a small range. However, real four-quadrant detectors often suffer from non-ideal factors such as detector dead zones, non-uniform quadrant response, channel gain mismatch, electronic noise, background light, defocus, and optical system-level errors, resulting in lower accuracy of the deflection angle calculated by the pre-stage solver. Summary of the Invention

[0005] Therefore, it is necessary to provide a residual calibration method for a four-quadrant detector used for laser deflection angle detection to address the aforementioned technical problems.

[0006] The following technical solution is adopted in this specification: This specification provides a method for residual calibration of a four-quadrant detector for laser deflection angle detection, including: Based on the output current of the four quadrants of the target surface of the four-quadrant detector, the normalized differential feature, diagonal differential feature, and power scale feature are calculated to form a response feature vector. Based on the normalized differential characteristics, the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector are determined, and the partition number of the spot on the target surface is determined based on the normalized response radius and response direction angle; based on the environmental interference parameters of the current four-quadrant detector and the physical parameters of the laser beam, the operating condition index corresponding to the spot is determined. Obtain the local residual model corresponding to the partition number and operating condition index, and input the response feature vector into the local residual model to obtain the target residual estimate; the target residual estimate reflects the local systematic error that the previous solver has not yet eliminated in the real system.

[0007] Optionally, the partition number of the light spot on the target surface is determined based on the normalized response radius and response direction angle, including: The radial region number of the light spot on the target surface is determined based on the normalized response radius; Based on the response direction angle, the sector number of the light spot on the target surface is determined; the circumference of the target surface is divided into multiple sectors according to the angle. The radial region number and sector number of the light spot on the target surface are encoded to obtain a unique partition number of the light spot on the target surface.

[0008] Optionally, the radial region includes a central region, an effective region, a transition region, and an edge region; the radial region numbering of the light spot on the target surface is determined based on the normalized response radius, including: If the normalized response radius is greater than or equal to 0 and less than the first preset radius threshold, then the light spot is determined to be in the central region of the target surface. If the normalized response radius is greater than or equal to the first preset radius threshold and less than the second preset radius threshold, then the effective area of ​​the light spot on the target surface is determined. If the normalized response radius is greater than or equal to the second preset radius threshold and less than the third preset radius threshold, then the transition zone of the light spot on the target surface is determined. If the normalized response radius is greater than or equal to the third preset threshold and less than or equal to 1, then the spot is determined to be in the edge region of the target surface. The radial region number of the light spot on the target surface is determined based on the radial region of the light spot on the target surface.

[0009] Optionally, based on the environmental interference parameters and laser beam physical parameters of the current four-quadrant detector, the operating condition index corresponding to the light spot is determined, including: The environmental interference parameters and laser beam physical parameters are quantified and classified separately to obtain the level code of each parameter; the environmental interference parameters include the status of ambient lights and background noise; the laser beam physical parameters include optical power, attenuator transmittance, signal-to-noise ratio, spot size and spot morphology characteristics. Based on the level code corresponding to each parameter, a unique working condition index for the light spot is generated.

[0010] Optionally, obtain the local residual model corresponding to the partition number and operating condition index, including: The partition number and the operating condition index are combined to form a two-dimensional joint index vector; Based on the two-dimensional joint index vector, a search is performed in the pre-built residual model library; the residual model library uses each combination of partition number and working condition index as a unique storage address key value, and each combination of partition number and working condition index uniquely corresponds to a set of pre-stored local residual model parameters; different combinations correspond to different local residual model parameters. The retrieved local residual model parameters that match the two-dimensional joint index vector are loaded into the residual estimator to obtain a local residual model that uniquely matches the current spot spatial location and the current working state.

[0011] Optionally, the method further includes: When determining the partition number of the light spot, if the normalized response radius or response direction corner falls into the preset boundary transition zone, then the partition numbers adjacent to each other on both sides of the boundary are used as candidate partition numbers. Based on the two candidate partition numbers and the working condition index corresponding to the spot, the first local residual model and the second local residual model are determined. The response feature vector is input into the first local residual model and the second local residual model respectively to obtain the first residual estimate and the second residual estimate; The interpolation weights are determined based on the position of the current normalized response radius or response direction angle within the boundary transition zone, and the first residual estimate and the second residual estimate are weighted and interpolated to obtain the target residual estimate.

[0012] Alternatively, the residual model library can be constructed in the following ways: The sample output current of the four quadrants of the four-quadrant detector target surface was collected under different deflection angles, different environmental interference parameters, and different laser beam physical parameters. The sample output currents of the four quadrants are input into a pre-calibrated front-end solver to obtain the initial deflection angle of the laser beam relative to the four-quadrant detector. The sample residual is defined as the difference between the actual deflection angle and the initial deflection angle of the sample. Based on the sample output current in the four quadrants, construct the sample response feature vector, as well as the partition number and operating condition index for each sample. For each combination of partition number and operating condition index, a local residual model is constructed using the sample residuals and sample response feature vectors of the corresponding samples. Store the model parameters of the local residual model under each combination of partition number and operating condition index into the residual model library.

[0013] Alternatively, the local residual model can be constructed using weighted ridge regression, local linear model, low-order polynomial model, or lightweight neural network.

[0014] Optionally, based on the output current in the four quadrants, normalized differential characteristics, diagonal differential characteristics, and power scale characteristics are calculated to form a response feature vector, including: Calculate the sum of the output currents in the four quadrants to obtain the total photocurrent; Based on the output current in each quadrant, calculate the offset of the light spot in the X and Y directions on the target surface; Based on the total photocurrent, the offset of the light spot in the X and Y directions of the target surface is normalized to obtain the normalized differential characteristics of the light spot on the target surface of the four-quadrant detector. The diagonal current difference is determined based on the difference in current between the diagonal quadrants among the output currents in the four quadrants. The diagonal difference characteristics are obtained by normalizing the diagonal current difference using the total photocurrent. Logarithmic calculation of the total photocurrent yields power-scale characteristics; The normalized difference features, diagonal difference features, and power scale features are combined to form the response feature vector.

[0015] Optionally, the normalized difference features include normalized difference features in the X direction and normalized difference features in the Y direction; based on the normalized difference features, the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector are determined, including: Calculate the square root of the sum of squares of the normalized difference features in the X direction and the normalized difference features in the Y direction, and divide by . The obtained value is determined as the normalized response radius; By inputting the normalized difference features in the vertical direction as the first variable and the normalized difference features in the horizontal direction as the second variable into the bivariate arctangent function, the response direction angle is obtained.

[0016] This specification provides a residual calibration device for a four-quadrant detector used for laser deflection angle detection, comprising: The calculation module is used to calculate the normalized differential features, diagonal differential features, and power scale features based on the output current of the four quadrants of the target surface of the four-quadrant detector, and form a response feature vector. The partitioning module is used to determine the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector based on the normalized differential characteristics, and to determine the partition number of the spot on the target surface based on the normalized response radius and response direction angle; and to determine the operating condition index corresponding to the spot based on the current environmental interference parameters of the four-quadrant detector and the physical parameters of the laser beam. The residual calibration module is used to obtain the local residual model corresponding to the partition number and operating condition index, and input the response feature vector into the local residual model to obtain the target residual estimate; the target residual estimate reflects the local systematic error that the previous solver has not yet eliminated in the real system.

[0017] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for residual calibration of a four-quadrant detector for laser deflection angle detection.

[0018] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-described method for residual calibration of a four-quadrant detector for laser deflection angle detection.

[0019] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: In the four-quadrant detector residual calibration method for laser deflection angle detection provided in this manual, by adaptively calling the corresponding local residual model based on the partition number of the light spot on the target surface and the working condition index determined by environmental interference and physical parameters, it can output a high-precision target residual estimate for different landing point areas and interference conditions. This estimate truly reflects the local system deviations that have not yet been eliminated by the front-end solver under the influence of non-ideal factors (such as detector dead zone, gain mismatch, noise, background light and optical errors), significantly improving the reliability of the deflection angle residual estimate. This provides an accurate and reliable correction basis for the subsequent error compensation stage, laying a solid foundation for improving the accuracy of deflection angle detection. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1 This document provides a schematic flowchart of a residual calibration method for a four-quadrant detector used for laser deflection angle detection. Figure 2 This is a schematic diagram of another method for residual calibration of a four-quadrant detector for laser deflection angle detection provided in this specification. Figure 3 This document provides a simulation result of the PRC-QD error. Figure 4 This specification provides a comparison chart of residuals before and after partitioning on a calibration set; Figure 5 This manual provides a comparison chart of two-dimensional vector RMSE under different test conditions. Figure 6 Heatmaps showing the improvement rate of X / Y axis RMSE under different test conditions provided in this manual; Figure 7 This specification provides a framework diagram of a laser micro-deflection angle detection system based on a four-quadrant detector; Figure 8 This specification provides a schematic diagram of a computer device for implementing a residual calibration method for a four-quadrant detector used for laser deflection angle detection. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0023] In existing technologies, the residual of the deflection angle can be calibrated or the deflection angle can be compensated or corrected, mainly including the following methods: 1) Global linear calibration scheme: This scheme assumes that the four-quadrant difference ratio output is approximately linear with the displacement or deflection angle within a certain range, and obtains the angle estimate through slope and zero-point correction. This scheme is simple to implement and has low computational cost, but it is usually only suitable for a small central area and low noise conditions. When the detector dead zone, edge truncation, or spot distortion causes nonlinear response, the unified linear model is difficult to take into account the accuracy of the central area, effective area, transition area, and edge area.

[0024] 2) Segmented fitting calibration scheme: The output response interval is divided into several sub-segments, and local linear or low-order polynomial fitting is performed in each segment. This scheme can partially alleviate the problem of inconsistent error patterns in different regions, but the segmentation is usually based only on single-axis output or displacement range, which does not make full use of the topological characteristics of the four-quadrant response, and rarely considers the drift of operating conditions such as optical power, spot shape and background noise.

[0025] 3) Analytical Compensation Scheme: Based on the ideal Gaussian spot model, dead zone model, or detector geometry model, the analytical relationship between the difference ratio output and the actual displacement is derived, and then corrected through parameter compensation. This scheme has clear physical meaning, but it makes strong assumptions about spot distribution, channel consistency, and response region shape. When the real system has multiple sources of error such as non-uniform response, channel gain mismatch, bias drift, and background light variation, a single analytical model is difficult to cover all error sources.

[0026] 4) System-level black-box correction scheme: Directly train a global mapping model, such as a high-order polynomial or neural network, using standard inputs and measurement outputs. This scheme can reduce errors on specific datasets, but model interpretability and transferability are limited; once operating conditions change or sample coverage is insufficient, local overfitting or decreased accuracy in edge regions is likely to occur.

[0027] In summary, the core shortcomings of existing solutions are: most methods treat the error as a globally uniform law, or segment it only according to a single dimension, making it difficult to simultaneously handle the residual differences caused by "response area differences" and "operating condition changes"; at the same time, directly reconstructing the complete solver will weaken the interpretability and reusability of the existing difference ratio solution framework.

[0028] The technical problems that this invention can solve include: 1) Without changing the original four-quadrant difference ratio solution framework, how to perform a secondary correction on the system residuals in the previous estimation results; 2) How to use the response characteristics such as normalized difference, diagonal difference, and total power in the four-quadrant output to separately process the error patterns in the central region, effective region, transition region, and edge region; 3) How to incorporate operating condition information such as optical power, spot size, background noise, and spot morphology into the same residual compensation framework to improve adaptability under complex operating conditions; 4) How to form a storable and interpretable local residual model during the offline calibration stage, and complete real-time invocation with a small amount of computation during the online detection stage; 5) How to reduce the angle estimation error caused by dead zone, non-uniform response, channel mismatch, noise and system-level optical error, and improve the detection accuracy of small deflection angle of laser beam.

[0029] Therefore, this invention provides a residual calibration method for a four-quadrant detector used for laser deflection angle detection. This method proposes a two-stage calibration framework of "pre-stage solution + partitioned residual correction." The first layer retains the original four-quadrant difference ratio solver, polynomial solver, or neural network solver, using their outputs to determine the initial deflection angle. The second layer constructs the residual using the standard angle truth value and establishes a local residual model based on the four-quadrant response characteristic partitioning and operating condition index, thereby performing a secondary correction on the pre-stage deflection angle estimation result. This method does not completely replace the original solver but rather learns the residuals that are difficult to describe using a unified analytical expression in real systems. This preserves the engineering interpretability of the pre-stage solver while enabling local compensation of residuals in different response regions (central region, transition region, edge region, etc.) and under different operating conditions, improving overall angle measurement accuracy and engineering adaptability.

[0030] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0031] Figure 1 This is a schematic diagram of a residual calibration method for a four-quadrant detector used for laser deflection angle detection, as described in this specification. The method includes the following steps: S101, based on the output current of the four quadrants of the target surface of the four-quadrant detector, calculate the normalized differential feature, diagonal differential feature, and power scale feature to form a response feature vector; the normalized differential feature is used to characterize the positional offset of the laser spot in the horizontal and vertical directions of the target surface.

[0032] To obtain the output current of the target surface of the current four-quadrant detector in each of the four quadrants, specifically, this includes obtaining the raw values ​​of the target surface of the current four-quadrant detector in each of the four quadrants. V A , V B , V C , V D The system can preprocess the raw values ​​in the four quadrants, including noise floor subtraction, saturation detection, and signal quality checks, to obtain the output current in each of the four quadrants of the current four-quadrant detector target surface. , , , .

[0033] The four quadrants of the four-quadrant detector are A, B, C and D. A is located in the upper left position of the four-quadrant detector target surface, B is located in the upper right position of the four-quadrant detector target surface, C is located in the lower right position of the four-quadrant detector target surface, and D is located in the lower left position of the four-quadrant detector target surface.

[0034] In this specification, the positive X-axis is defined as the left side of the target surface, and the positive Y-axis is defined as upwards; therefore, >0 indicates that the energy center of the light spot is biased to the left. >0 indicates that the energy center of the light spot is biased towards the upper side.

[0035] Raw values ​​for the four quadrants are acquired in real time via a front-end printed circuit board (PCB) hardware module. V A , V B , V C , V D .

[0036] Diagonal difference features are used to characterize the asymmetry of the laser spot along the two diagonals, while power scale features are used to reflect the total energy of the spot and its nonlinear scale information.

[0037] In one embodiment, based on the output currents in the four quadrants of the four-quadrant detector target surface, normalized differential characteristics, diagonal differential characteristics, and power scale characteristics are calculated to form a response feature vector, including the following steps: S201, calculate the sum of the output currents in the four quadrants to obtain the total photocurrent.

[0038] Calculate the total photocurrent: .

[0039] S202, calculate the offset of the light spot in the X and Y directions of the target surface based on the output current of each quadrant; normalize the offset of the light spot in the X and Y directions of the target surface based on the total photocurrent to obtain the normalized differential characteristics of the light spot on the target surface of the four-quadrant detector.

[0040] The offset of the light spot in the X direction on the target surface can be: The offset of the light spot in the Y direction on the target surface can be: .

[0041] Therefore, the normalized difference feature in the X (horizontal) direction The calculation formula is: in, To find a very small positive number that prevents division by zero in low light or no light conditions, preferably, we can take... , where median(S_cal) is the median of the total signal in the calibration set.

[0042] Normalized difference features in the Y (vertical) direction The calculation formula is: S203. Based on the difference in current between the diagonal quadrants among the output currents of the four quadrants, the diagonal current difference is determined; the diagonal current difference is normalized by the total photocurrent to obtain the diagonal difference characteristics.

[0043] Diagonal current differential includes: , .

[0044] Diagonal difference features include and : S204 performs a logarithmic operation on the total photocurrent to obtain the power scale characteristics.

[0045] Power scale characteristics are .

[0046] S205 combines the normalized difference features, diagonal difference features, and power scale features to form a response feature vector.

[0047] Response feature vector , , This reflects an imbalance between horizontal and vertical energy. , This reflects the inconsistency between the diagonal quadrants and the changes in the shape of the light spot; Used to characterize total optical power.

[0048] S102, based on the normalized differential characteristics, determine the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector, and determine the partition number of the spot on the target surface based on the normalized response radius and response direction angle; determine the operating condition index corresponding to the spot based on the current environmental interference parameters of the four-quadrant detector and the physical parameters of the laser beam.

[0049] In one embodiment, determining the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector based on the normalized difference feature includes: calculating the modulus of the normalized difference feature, determining the modulus of the normalized difference feature as the normalized response radius; and inputting the normalized difference feature as a variable into the arctangent function to obtain the response direction angle.

[0050] Specifically, normalized response radius The calculation formula is: The formula for calculating the response direction angle is: in, This represents the normalized response radius, with a value range of [0,1]. Indicates the response direction angle.

[0051] In one embodiment, determining the partition number of the light spot on the target surface based on the normalized response radius and response direction angle includes: determining the radial region number of the light spot on the target surface based on the normalized response radius; determining the sector number of the light spot on the target surface based on the response direction angle; dividing the circumference of the target surface into multiple sectors according to angles; and encoding the radial region number and sector number of the light spot on the target surface to obtain a unique partition number of the light spot on the target surface.

[0052] The radial region includes a central region, an effective region, a transition region, and an edge region; three radius thresholds are preset. , and The target surface is divided radially into four concentric zones: the central zone, the effective zone, the transition zone, and the edge zone, which are numbered accordingly. q =1,2,3,4; Based on the normalized response radius, determine the radial region number of the light spot on the target surface, including: if the normalized response radius is greater than or equal to 0 and less than the first preset radius threshold. If the light spot is located in the central region of the target surface, then the normalized response radius is greater than or equal to the first preset radius threshold. And less than the second preset radius threshold Then the effective area of ​​the light spot on the target surface is determined; if the normalized response radius is greater than or equal to the second preset radius threshold... And less than the third preset radius threshold Then, the transition region of the light spot on the target surface is determined; if the normalized response radius is greater than or equal to the third preset radius threshold... If the value is less than or equal to 1, then the spot is located in the edge region of the target surface; based on the radial region of the spot on the target surface, the radial region number of the spot on the target surface is determined.

[0053] Optionally, let the radial boundary 0 = r 0< r 1<…< Then the radial region number q satisfy .in, Indicates the total number of radial regions; r0 to This indicates the radial boundaries set in ascending order; q Indicates the radial region number. q ∈{1,2,…, }; r This represents the normalized response radius. When... When =4, r 1. r 2. r 3 respectively correspond to the first preset radius threshold R 1. Second preset radius threshold R 2 and the third preset radius threshold R 3.

[0054] The target surface circumference is divided into multiple sectors according to angles, including: dividing the target surface circumference into equal sections according to angles. Each sector.

[0055] According to the response direction angle Determine the sector number of the light spot on the target surface. .

[0056] The radial region and sector numbering of the light spot on the target surface are based on... Encoding is performed to obtain a unique partition number of the light spot on the target surface. .

[0057] In one embodiment, the operating condition index corresponding to the light spot is determined based on the environmental interference parameters and laser beam physical parameters of the current four-quadrant detector. This includes: quantifying and classifying the environmental interference parameters and laser beam physical parameters to obtain a level code for each parameter; the environmental interference parameters include ambient light status and background noise; the laser beam physical parameters include optical power, attenuator transmittance, signal-to-noise ratio, light spot size, and light spot morphology characteristics; and generating a unique operating condition index for the light spot based on the level code corresponding to each parameter.

[0058] The working condition vector can be written as .in, T To reduce the transmittance of the attenuator, ρ The total power is the metric. For signal-to-noise ratio, For the scale of light spots, B In ambient light mode. H Labels for light spot shapes such as round and irregular. N This is background noise.

[0059] The ambient light status indicates whether the lights are on or off.

[0060] The parameters of the working condition vector are discretized according to the calibration nodes, and the working condition index is denoted as... .

[0061] Specifically, the total power scale is compared with a preset optical power threshold to determine the optical power level code. The transmittance of the attenuator is compared with a preset transmittance threshold to determine the transmittance level code. The signal-to-noise ratio (SNR) is compared with a preset SNR threshold to determine the SNR level coding. The spot size is compared with a preset size threshold to determine the spot size level code. Use the code corresponding to the ambient light status as the code for the light on / off status level. The morphological characteristics of the light spot are matched with the preset light spot morphology categories to determine the light spot morphology level code. The background noise is compared with a preset noise threshold to determine the noise level code. .

[0062] The codes corresponding to the ambient light status include 0 and 1. If the ambient light status is "on", the code corresponding to the ambient light status is 1. If the ambient light status is "off", the code corresponding to the ambient light status is 0.

[0063] In addition, the threshold values ​​for parameters other than ambient light status can include at least one. For example, if a parameter has two threshold values, taking the transmittance threshold as an example, the transmittance threshold includes a first transmittance threshold and a second transmittance threshold, where the second transmittance threshold is greater than the first transmittance threshold. The transmittance of the attenuator is compared with the preset transmittance threshold to determine the transmittance level code. This includes: if the transmittance of the attenuator is less than the first transmittance threshold, then it is determined that the transmittance is severely attenuated, and a transmittance level coding is performed. The value is 0; if the transmittance of the attenuator is greater than or equal to the first transmittance threshold and less than or equal to the second transmittance threshold, then the transmittance is determined to be moderately attenuated, and the transmittance level is coded. The value is 1; if the transmittance of the attenuator is greater than the second transmittance threshold, then it is determined to be high transmittance, and the transmittance level is coded. The value is 2.

[0064] It should be noted that the specific threshold and corresponding level code values ​​can be set according to actual needs.

[0065] S103, obtain the local residual model corresponding to the partition number and operating condition index, and input the response feature vector into the local residual model to obtain the residual estimate; the residual estimate reflects the local system error that the previous solver has not yet eliminated in the real system.

[0066] In one embodiment, obtaining the local residual model corresponding to the partition number and the operating condition index includes: combining the partition number and the operating condition index to form a two-dimensional joint index vector; searching in a pre-built residual model library based on the two-dimensional joint index vector; the residual model library uses each combination of partition number and operating condition index as a unique storage addressing key value, and each combination of partition number and operating condition index uniquely corresponds to a set of pre-stored local residual model parameters; different combinations correspond to different local residual model parameters; loading the retrieved local residual model parameters that match the two-dimensional joint index vector into the residual estimator to obtain a local residual model that uniquely matches the current spot spatial position and the current working state.

[0067] In one embodiment, when determining the partition number of the light spot, if the normalized response radius or response direction angle falls into a preset boundary transition zone, then the partition numbers adjacent to each other on both sides of the boundary are used as candidate partition numbers; based on the two candidate partition numbers and the operating condition index corresponding to the light spot, a first local residual model and a second local residual model are determined respectively; the response feature vector is input into the first local residual model and the second local residual model respectively to obtain the first residual estimate and the second residual estimate; the interpolation weight is determined based on the position of the current normalized response radius or response direction angle within the boundary transition zone, and the first residual estimate and the second residual estimate are weighted and interpolated to obtain the target residual estimate.

[0068] Specifically, the boundary transition zone is a "fuzzy zone" or "buffer zone" defined between two adjacent partitions. Its core purpose is to avoid errors caused by a one-size-fits-all approach. When the normalized response radius (size) or response direction angle (angle) happens to fall near the partition boundary, even small noise or perturbations can cause the value to oscillate between sides, thus being forcibly assigned to the wrong side. After introducing the boundary transition zone, values ​​falling into this area are no longer forcibly assigned to one side of the partition; instead, both sides are considered to preserve possibilities for subsequent fine-tuning or weighted voting.

[0069] Taking the normalized response radius as an example, for instance, the transition zone between the center region and the effective region, the transition zone and the transition zone, and the edge region all include the edge transition zone; the second boundary radius threshold < the second boundary radius threshold < the third boundary radius threshold < the fourth boundary radius threshold < the fifth boundary radius threshold < the sixth boundary radius threshold; if the normalized response radius is greater than or equal to 0 and less than the first boundary radius threshold, then the spot is determined to be in the center region of the target surface; If the normalized response radius is greater than the second boundary radius threshold and less than the third boundary radius threshold, then the effective area of ​​the light spot on the target surface is determined. If the normalized response radius is greater than the fourth boundary radius threshold and less than the fifth boundary radius threshold, then the transition zone of the light spot on the target surface is determined. If the normalized response radius is greater than the sixth boundary radius threshold and less than or equal to 1, then the spot is determined to be in the edge region of the target surface. If the normalized response radius is greater than or equal to the first boundary radius threshold and less than or equal to the second boundary radius threshold, then the light spot is determined to fall into the boundary transition zone between the center area and the effective area, and the light spot is in the center area and the effective area of ​​the target surface; if the normalized response radius is greater than or equal to the third boundary radius threshold and less than or equal to the fourth boundary radius threshold, then the light spot is determined to fall into the boundary transition zone between the effective area and the transition area, and the light spot is in the effective area and the transition area of ​​the target surface; if the normalized response radius is greater than or equal to the fifth boundary radius threshold and less than or equal to the sixth boundary radius threshold, then the light spot is determined to fall into the boundary transition zone between the transition area and the edge area, and the light spot is in the transition area and the edge area of ​​the target surface.

[0070] Optionally, if the normalized response radius falls within the boundary transition zone between the central region and the effective region, then calculate the first radial region number where the normalized response radius falls within the central region and the second radial region number where the normalized response radius falls within the effective region; determine the first candidate partition number based on the first radial region number and the sector number, and determine the second candidate partition number based on the second radial region number and the sector number.

[0071] The first local residual model is determined based on the first candidate partition number and the operating condition index, and the second local residual model is determined based on the second candidate partition number and the operating condition index.

[0072] Calculate the relative position ratio of the current normalized response radius or response direction angle within the boundary transition zone, and determine the interpolation weights.

[0073] For example, taking the boundary transition zone between the central region and the effective region of the normalized response radius as an example, the interpolation weights... The calculation method for the target residual estimate is as follows: in, The first boundary radius threshold, The second boundary radius threshold, This is the first residual estimator. This is the second residual estimator. This is the target residual estimator.

[0074] It should be noted that if both the normalized response radius and the response direction angle fall within the preset boundary transition zone, then there will be two corresponding partition numbers and operating condition indices. In this case, the partition number and operating condition index will be combined to calculate the target residual estimate.

[0075] In one embodiment, the output current of the four quadrants is input into a pre-calibrated front-end solver to obtain the initial deflection angle of the laser beam relative to the four-quadrant detector; the initial deflection angle and the residual estimate are added together to obtain the corrected deflection angle of the laser beam relative to the four-quadrant detector.

[0076] Corrected laser beam deflection angle relative to the four-quadrant detector ,in, The initial deflection angle, This is the residual estimate. The corrected deflection angle of the laser beam relative to the four-quadrant detector is used for display, recording, or closed-loop control.

[0077] In one embodiment, the construction of the residual model library includes the following steps: S301 collects sample output currents from the four quadrants of the four-quadrant detector target surface under different deflection angles, different environmental interference parameters, and different laser beam physical parameters.

[0078] S302 inputs the sample output current of the four quadrants into the pre-calibrated front-end solver to obtain the initial deflection angle of the laser beam relative to the four-quadrant detector.

[0079] The preceding solver can be a difference ratio method, a polynomial model, a composite algorithm, or a neural network model.

[0080] S303 defines the sample residual based on the difference between the actual deflection angle and the initial deflection angle of the sample.

[0081] The actual standard angle can be provided by a displacement stage, turntable, calibrated mirror, or equivalent spot displacement platform. This residual reflects the local systematic errors that the preceding solver has not yet eliminated in the real system.

[0082] It should be noted that the initial deflection angle includes the initial deflection angles in both the X and Y directions. and The actual deflection angle includes the actual deflection angles in the X and Y directions. and Correspondingly, the sample residuals also include the true residuals in the X and Y directions. and .

[0083] Specifically, , .

[0084] S304 constructs a sample response feature vector, as well as the partition number and operating condition index of each sample, based on the sample output current in the four quadrants.

[0085] Specifically, based on the sample output current in the four quadrants, the sample normalized difference feature, sample diagonal difference feature, and sample power scale feature are calculated to form the sample response feature vector.

[0086] For any given sample, based on the sample normalized difference characteristics, the sample normalized response radius and sample response direction angle of the laser spot on the target surface of the four-quadrant detector are determined. Based on the sample normalized response radius and sample response direction angle, the partition number of the sample spot on the target surface is determined. Based on the environmental interference parameters and laser beam physical parameters of the sample, the operating condition index corresponding to the sample spot is determined to distinguish parameter drift under different conditions.

[0087] The specific calculation method in this embodiment is the same as that in the above embodiments, and will not be repeated here.

[0088] S305, in each combination of partition number and working condition index (p , m Under this condition, a local residual model is constructed using the sample residuals and sample response feature vectors of the corresponding samples.

[0089] The local residual model is constructed using weighted ridge regression, a locally linear model, a low-order polynomial model, or a lightweight neural network. Preferably, it employs a basis function expansion consisting of a constant term, a linear term, a cross term, and a low-order nonlinear term, denoted as […]. . For residual estimators, For the basis functions of the local residual model transpose, These are the model parameters, corresponding to the parameter matrices under different partitions and operating conditions.

[0090] In each ( p , m Within a combination, use basis functions. Establish a two-dimensional residual model. Assume that the combination contains a total of... N For the nth sample, design matrix X of the nth sample... i Behavior The residual matrix E of the first i Behavior The parameter matrix is ​​then:

[0091] in, , These are the residual estimates in the X and Y directions, respectively; This is the regularization matrix, used to constrain model parameters and suppress overfitting; This is a diagonal matrix of sample weights, used to characterize the relative weights of each training sample in the fitting process; It is a non-negative regularization coefficient used to adjust the trade-off between fitting error and parameter complexity.

[0092] If the current operating condition is located between multiple calibration nodes, the nearest one can be selected. L Perform inverse distance weighted interpolation on each working condition node: , , Where L is the number of the most recently calibrated nodes participating in the interpolation; The normalized current continuous operating condition variable; For the first l The normalized process vector corresponding to each calibration node; δ To prevent small amounts from being divided by zero; For the current working condition z To the node The Euclidean distance; For the first l Normalized inverse distance weights of each labeled node, Indicates working conditions s To the node The Euclidean distance; For partition p, operating condition node The corresponding local residual model parameter matrix; This is the model parameter matrix obtained by interpolation for partition p under the current operating condition c.

[0093] S306 stores the model parameters of the local residual model under each combination of partition number and working condition index into the residual model library.

[0094] Parameter storage and confidence record: Stores model parameters for each partition and operating condition index. Sample size Input range, training residual variance, and maximum training residual. The residual variance can be used as a confidence reference when calling the function online.

[0095] Taking a single-axis linear model as an example, the local residual variance for: when When the data is insufficient to support the solution of model parameters, it is preferentially merged with adjacent partitions or reverted to the global residual model. When online samples exceed the calibration range, extrapolation compensation is not performed by default; only the previous stage results are output and a "out of calibration range" flag is generated. Confidence can be determined comprehensively based on the number of samples, local variance, and whether the data is within the calibration range, such as model confidence. ,in and f is set by the calibration data.

[0096] In one embodiment, such as Figure 2 As shown, this invention also provides a residual calibration method for a four-quadrant detector used for laser deflection angle detection. This method includes a residual calibration process and a correction process. The correction process includes:

[0097] 1) Read the raw values ​​from four channels V A , V B , V C , V D Information on current operating conditions, such as the transmittance of the attenuator and the status of ambient lights.

[0098] 2) Press Complete noise floor reduction and channel consistency correction, and perform total signal, saturation, and repeatability checks.

[0099] 3) Calculation , , , , , , r , ψ and optional signal-to-noise ratio SNR , spot size and morphological parameters.

[0100] 4) Call the pre-stage solver to obtain the initial deflection angles in the X and Y directions of the target surface. , .

[0101] 5) Obtain the partition number p according to the simplified five-interval verification example or the general two-dimensional response topology partitioning, and obtain the operating condition index based on transmittance, background noise, light spot morphology, etc. m .

[0102] 6) Calculate the residual estimates in the X and Y directions of the target surface. , At the boundary, the output interpolation of adjacent local residual models is used. 7) The deflection angle of the laser beam, corrected in the X and Y directions on the target surface, relative to the four-quadrant detector. , And calculate the magnitude of the two-dimensional deflection angle. .

[0103] 8) Simultaneously output confidence level, partition number, operating condition number, whether it is out of range, and whether it is saturated, etc., for display, recording or closed-loop control.

[0104] The following simplified verification example based on existing experimental data illustrates the residual calibration method for a four-quadrant detector used for laser deflection angle detection provided by this invention. It should be noted that the aforementioned central region, effective region, transition region, and edge region belong to a general two-dimensional response topology partitioning based on normalized response radius and response direction angle; the following five-interval scheme partitions according to the single-axis pre-stage output, only used to illustrate the feasibility of the "pre-stage solution + partitioned residual correction" concept, and does not correspond one-to-one with the above four radial regions, nor does it limit the number of partitions, partitioning basis, or input form of the local residual model in the general scheme. Both methods determine the applicable scope of the local model based on response characteristics and input the corresponding characteristics into the local residual model; the five-interval scheme is a dimensionality reduction verification form adopted to adapt to existing single-axis experimental data, and therefore does not have a technical contradiction with the general two-dimensional partitioning scheme.

[0105] Global polynomial baseline models for the X-axis and Y-axis are established to obtain baseline residuals. Then, the deflection angle range is divided into 5 intervals, and a linear residual model is fitted to each interval. During the testing phase, the corresponding residual model is selected according to the interval where the baseline deflection angle is located, and the residuals are superimposed on the output of the global model to obtain the compensated deflection angle.

[0106] Calibration data is used to establish the global baseline model and the regional residual model, while independent test data is used to evaluate the model. The test data covers conventional circular light spots, different signal-to-noise ratio conditions, light-on interference conditions, and irregularly shaped light spots. As shown in Table 1, Table 1 shows the data types and uses of the examples.

[0107] Table 1. Instance Data Types and Their Uses X-axis calibration data X-axis normalized output vs. actual displacement Establish the global model and the partitioned residual model of the X-axis. Y-axis calibration data Y-axis normalized output vs. actual displacement Establish the global model and the partitioned residual model along the Y-axis. Circular light spot test data Different transmittance / signal-to-noise ratio Verification of compensation effect under conventional light spot conditions Interference test data when lights are on Ambient light interference Verification of compensation effect under background interference conditions Irregular light spot test data Non-ideal light spot Generalization capability verification In one embodiment, the experimental light source is an LHP-151-230 helium-neon laser with a wavelength of 632.8 nm and an output power of 5 mW; the four-quadrant detector is an OSQ5.8-IT; different transmittances are set by attenuators, the spot morphology is improved by a spatial light filtering unit, and the relative position between the detector and the spot is changed by an MBT616D three-axis displacement stage. This displacement stage has a resolution of 50 μm and a travel of 5 mm. The four analog outputs are buffered, amplified, filtered, and digitized by a multi-channel ADC, read by an STM32 microcontroller, and uploaded to the host computer via USB-to-serial conversion. The CMOS branch after beam splitting is used to observe the spot morphology and provide a reference centroid position.

[0108] Each sample point must record at least: a timestamp t , setpoints Δx and Δy for the displacement stage or deflection mechanism, and four original sampled values V A , V B , V C , V D Attenuation filter transmittance T Ambient light status B Spot morphology label H And the CMOS image or its centroid coordinates. The recommended data row representation is as follows: Continuous data collection at the same location. K frame, K It can be a configurable positive integer; the host computer simultaneously saves the original frame data and the frame average data for subsequent traceability.

[0109] In one embodiment, the acquisition of noise floor, channel gain, and effective signal specifically includes: for each channel when the laser is blocked or does not illuminate the detector. i Continuous data acquisition K 0 frames, obtain the average noise floor of the corresponding channel. and noise standard deviation ,in i ∈{A,B,C,D}.

[0110] in, K 0 represents the number of frames sampled for the noise floor; For the first i The channel is in k The original sampled values ​​of the frame; For the first i The average noise floor of the channel; For the first i Standard deviation of the channel noise floor samples; i ∈{A,B,C,D}.

[0111] The regular circular light spot is adjusted to the geometric center of the four quadrants, theoretically ensuring that each quadrant receives the same amount of energy. The mean value of each channel after deducting background noise is recorded under the center operating condition. q i and the average of the four channels Based on this, calculate the channel gain / response consistency correction factor. The effective four-quadrant signal, after noise floor and gain correction, is used in online or offline calculations. I i :

[0112] in, The original voltage value after ADC conversion or a digital quantity proportional to the photocurrent; The combined noise floor is calculated based on dark current, electron zero bias, and the environmental common baseline. Used to correct inconsistencies in gain and response across four amplification channels. If the hardware has already undergone precise gain calibration, it can be set to... .

[0113] In one embodiment, single-point data averaging, outlier removal, and quality assessment include: for continuously acquired data from the same input location... K For frame data, noise floor and gain corrections are performed frame by frame first, and then a robust average is calculated. Preferably, outlier frames that deviate from the median by more than three times the absolute deviation of the median are first removed, and then the average of the remaining frames is calculated to obtain the value at that position. , , , If no outlier removal is performed, the arithmetic mean is used directly.

[0114] in, Indicates the first i The average output of valid frames at the current input position of the channel; Indicates the number of valid frames after removing abnormal frames; valid Represents the set of valid frame indices; Indicates the first i Channel 1k The effective signal after the frame has undergone noise floor and gain correction.

[0115] Perform a signal quality check at each location point: Total signal S It must be greater than the preset threshold. No channel may reach ADC saturation value. ; number of effective frames The values ​​must not be lower than the preset minimum; the coefficient of variation of repeated sampling of the four signals must not exceed the threshold. Data that fails the check will not be included in the model training, and a low confidence or anomaly flag will be output during the online phase.

[0116] In one embodiment, the actual deflection angle is obtained in three ways, specifically: Method 1 is equivalent calibration of the displacement stage. Taking the center position of the light spot or the center position of the detector as the zero point, the changes in readings (relative displacement) of the three-axis displacement stage in the X and Y directions are recorded as follows: , The system's equivalent focal length is denoted as The actual deflection angles in the X and Y directions are... and for:

[0117] Under small angle conditions: , .

[0118] When Δx, Δy and All times are in mm, and the calculation result is in rad; multiply by 10³ when converting to mrad, and multiply by 10 when converting to μrad. 6 If the experiment only uses displacement as the calibration output, the model can be trained in the displacement domain first, and then the above relationship can be used to uniformly convert it into angle.

[0119] Method two involves input via a reflector or deflector stage. If the mechanical rotation angle of the reflector is α, then the angle of the reflected beam changes by 2α, hence the true value of the standard angle... The displacement of the detection surface satisfies .

[0120] Method 3 uses a CMOS reference branch. The centroid is calculated for the grayscale value G(x,y) after background subtraction and thresholding. Then, based on the pixel size or calibration ratio Converted to surface displacement, and finally to standard angle:

[0121] Where Ω represents the effective spot pixel region participating in centroid calculation after background subtraction and threshold segmentation; x and y represent pixel coordinates; and G(x,y) represents the gray value at pixel (x,y). , These represent the centroid coordinates of the light spot in the X and Y directions, respectively.

[0122] in,( , () is the reference centroid when there is zero deflection. This represents the equivalent focal length of the CMOS branch. The experiment uses a CMOS reference angle to evaluate different pre-stage algorithms.

[0123] In one embodiment, if a CMOS image is used to establish a working condition index, the spot size and morphological parameters can be calculated: , ; Shape irregularity If CMOS is not configured, the operating condition index only uses transmittance, total signal, backlight status, and repetitive sampling noise.

[0124] The signal-to-noise ratio can be calculated from the total signal and the standard deviation of the noise floor: In this invention, the output currents of the four quadrants are input to the pre-stage solver. This includes preprocessing features such as normalized differences calculated based on the four output currents within the pre-stage solver or at its input terminal, and then the specific solution model uses these preprocessed features to perform angle calculations. Therefore, from the system interface perspective, the pre-stage solver receives four output currents. For the preferred embodiment described below, which is consistent with actual experimental data, the 9th-order odd-degree polynomial models for the X-axis and Y-axis use the normalized difference features n_x and n_y as direct independent variables, respectively:

[0125] and Used to absorb zero-point bias; if the zero point is strictly zero after preprocessing, the constant term can be set to 0. Let the X-axis design matrix be... The i-th action The standard angular column vector is Then, least squares or ridge regression can be used to solve the problem.

[0126] in This prevents the constant term from being regularized; The regularization coefficient is . It degenerates into ordinary least squares. The preceding stage can also be replaced with a difference-ratio linear model, a composite model, a backpropagation neural network, or other existing algorithms, while the subsequent residual calculation method remains unchanged.

[0127] In the simplified verification example above, partitioning is performed for both the X-axis and Y-axis. Let the minimum and maximum values ​​of the preceding output in a certain axis calibration set be respectively... and Divide it into 5 equal-width intervals with the following boundaries:

[0128] , k =0,1,…,5 No. p Each interval satisfies , The last interval includes the right endpoint. If the calibration samples are significantly unevenly distributed within the angular range, it can also be... Set as The 20%, 40%, 60%, and 80% quantiles were used to ensure that the sample size in each interval was similar. The experiment was only limited to "dividing into 5 intervals" and no fixed numerical boundaries were given. Therefore, the boundaries should be automatically calculated from the actual calibration range and saved along with the model parameters.

[0129] In the p Within each interval, the previous stage output As the feature input of the local residual model in this simplified validation example, the residual As the dependent variable, fit a linear model. , , Here are the coefficients. The explicit calculation method for ordinary least squares is as follows:

[0130] The summation is performed only on samples belonging to the p-th interval. and These are the mean values ​​of the preceding output and the residual, respectively, within this interval. Ridge regression can also be used to improve stability for small sample intervals. ,in Every action , , This is the sample weight matrix.

[0131] Five sets of parameters were obtained for the X-axis and Y-axis respectively. and During online computation, the results are first obtained from the preceding model. , Then find the corresponding interval, calculate the residual estimate and sum them:

[0132] , , To avoid jumps at interval boundaries, [the following can be done at the boundary] b k A transition zone of width Δ is provided on both sides. When fall into At that time, linear interpolation is output for adjacent intervals: , .

[0133] In one embodiment, the specific calculation method of the evaluation metric includes: for the first independent test set... i For each sample, define the uniaxial error and the two-dimensional vector error: , The number of test samples is N At that time, the various indicators were calculated as follows: , The experimental independent test set contained 437 test points. Using the above calculation method, the 2D vector RMSE decreased from 0.3238 mrad to 0.3147 mrad, an improvement of approximately 2.8%; the maximum 2D vector residual decreased from 2.4712 mrad to 2.4194 mrad; the X-axis RMSE decreased from 0.2995 mrad to 0.2900 mrad, and the Y-axis RMSE decreased from 0.1231 mrad to 0.1223 mrad.

[0134] In one embodiment, calibration data is used only to solve for global model coefficients, partition boundaries, and local residual parameters; independent test data must not participate in parameter fitting. Experimental test data covers different transmittance conditions under circular light spots (including 0.125%, 0.05%, and 0.0625%), two sets of light-on interference data under 0.0625% transmittance, and data for irregularly shaped light spots. X-axis, Y-axis, and two-dimensional indices are calculated for each test file, and finally, the overall indices of all 437 points are summarized. For conditions such as irregularly shaped light spots that exceed the distribution of the calibration samples, a lower confidence level should be output, and priority should be given to adding new condition nodes or recalibrating.

[0135] like Figure 3 As shown, Figure 3 The simulation results for the Partitioned Residual Calibration for Quadrant Detector (PRC-QD) are shown. Figure (a) is a histogram of the angle error norm distribution, with the x-axis representing the angle error norm (μrad) and the y-axis representing the number of samples; blue, orange, and green represent the error distribution after using only the pre-stage solver, using the global residual model, and using PRC-QD, respectively. Figure (b) is a residual error distribution in the angle plane, with the x and y axes representing the true angles (μrad) in the X and Y directions, respectively, and the scatter plot colors corresponding to the PRC-QD error norm (μrad) shown on the right. Figure (c) is a comparison of center / edge robustness, with the x-axis representing the response radius in the feature space and the y-axis representing the angle error norm (μrad); blue and orange scatter plots represent the results of the pre-stage solver and PRC-QD, respectively. (d) The figure shows the relationship between local residual variance and error validity. The horizontal axis represents the predicted local variance, and the vertical axis represents the PRC-QD error norm (μrad). The blue scatter plots represent the results for each sample. Figure 3 It can be seen that, under simulation conditions, PRC-QD partitioned residual calibration can significantly compress the error distribution and improve the robustness of the center / edge.

[0136] like Figure 4 As shown, Figure 4 This is a comparison chart of the residuals before and after partitioning on the calibration set. (a) shows the comparison results in the X direction, and (b) shows the comparison results in the Y direction. The horizontal axis of both charts represents the reference angle (mrad) in the corresponding direction, and the vertical axis represents the residual (mrad). The horizontal zero line indicates that the residual is zero. Blue scatter points represent the residuals before partitioning residual correction, and green scatter points represent the residuals after partitioning residual correction. The corrected scatter points are generally closer to the zero line, indicating that the local residual model can correct systematic errors in different angle intervals. On the calibration set, the distribution of residuals changes before and after partitioning, demonstrating that the local residual model can correct systematic errors in different angle intervals.

[0137] As shown in Table 2, Table 2 presents experimental results data.

[0138] Table 2 Experimental Results Data 2D vector RMSE / mrad 0.3238 0.3147 Approximately 2.8% improvement Maximum 2D vector residual / mrad 2.4712 2.4194 Decrease of 0.0518 mrad X-axis RMSE / mrad 0.2995 0.2900 3.17% improvement Y-axis RMSE / mrad 0.1231 0.1223 0.61% improvement like Figure 5 As shown, Figure 5 A comparison chart of two-dimensional vector RMSE under different test conditions, such as Figure 6 As shown, Figure 6 Heatmap of RMSE improvement rate on the X / Y axis under different test conditions.

[0139] The advantages of this invention compared to the closest prior art are as follows: 1) Improve angle measurement accuracy: In independent testing with real data, the overall two-dimensional vector RMSE decreased from 0.3238 mrad in global polynomial calibration to 0.3147 mrad in partition residual compensation, an improvement rate of approximately 2.8%; the maximum two-dimensional vector residual decreased from 2.4712 mrad to 2.4194 mrad.

[0140] 2) Improved axis error: X-axis RMSE decreased from 0.2995 mrad to 0.2900 mrad, an improvement rate of 3.17%; Y-axis RMSE decreased from 0.1231 mrad to 0.1223 mrad, an improvement rate of 0.61%.

[0141] 3) Enhanced local robustness: Compared to using only the front-end solver or a single global residual model, the partitioned residual model can model the error patterns of the central region, effective region, transition region and edge region separately, so that the errors in the edge region and large displacement region are no longer entirely borne by a unified model.

[0142] 4) Strong compatibility: This invention does not limit the form of the front-end solver and can be superimposed on the difference ratio method, polynomial fitting, composite algorithm or neural network solver, which facilitates software upgrades of existing systems.

[0143] 5) Simple engineering implementation: The offline stage trains and stores local model parameters, while the online stage only needs to calculate a small number of features, query partition / condition indexes and perform low-order model operations, which is suitable for real-time detection and embedded deployment.

[0144] 6) Good interpretability: The partitioning is based on the four-quadrant response topology and operating condition index, and the residual variance can be used as a confidence index, which is beneficial for system debugging, anomaly identification and subsequent parameter maintenance.

[0145] The method provided by this invention can be applied to a laser micro-deflection angle detection system based on a four-quadrant detector, such as... Figure 7 As shown, Figure 7 This is a framework diagram of a laser micro-deflection angle detection system based on a four-quadrant detector. The system includes an optical module, a detection and acquisition module, and a host computer processing and display module. The optical module is used for laser beam guiding, collimation, spatial light filtering, and deflection / equivalent deflection input; the detection and acquisition module is used to acquire the four output signals of the four-quadrant detector; the host computer or main control module is used to perform pre-stage solving, residual partitioning judgment, local residual compensation, and result display.

[0146] Specifically, the optical module includes a laser 1, a filter 2, a first reflecting mirror 3, a second reflecting mirror 4, a first aperture 5, a second aperture 6, a microscope objective 7, a micro-aperture 8, a first achromatic lens 9, a third aperture 10, an adjustable reflecting mirror 11, a second achromatic lens 12, and a beam splitter 13; the detection and acquisition module includes a CMOS camera 14, a four-quadrant detector 15, and a back-end processing circuit 16; the host computer processing and display module includes a computer terminal 17.

[0147] When applying the residual calibration method for the four-quadrant detector used for laser deflection angle detection provided in this manual, it is not necessary to follow the... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this manual does not impose any restrictions on it.

[0148] The above describes one or more embodiments of a four-quadrant detector residual calibration method for laser deflection angle detection provided in this specification. Based on the same idea, this specification also provides a corresponding four-quadrant detector residual calibration device for laser deflection angle detection, which includes: The calculation module is used to calculate the normalized differential features, diagonal differential features, and power scale features based on the output current of the four quadrants of the target surface of the four-quadrant detector, and form a response feature vector. The partitioning module is used to determine the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector based on the normalized differential characteristics, and to determine the partition number of the spot on the target surface based on the normalized response radius and response direction angle; and to determine the operating condition index corresponding to the spot based on the current environmental interference parameters of the four-quadrant detector and the physical parameters of the laser beam. The residual calibration module is used to obtain the local residual model corresponding to the partition number and operating condition index, and input the response feature vector into the local residual model to obtain the target residual estimate; the target residual estimate reflects the local systematic error that the previous solver has not yet eliminated in the real system.

[0149] Specific limitations regarding the residual calibration device for the four-quadrant detector used for laser deflection angle detection can be found in the limitations of the residual calibration method for the four-quadrant detector used for laser deflection angle detection described above, and will not be repeated here. Each module in the aforementioned residual calibration device for the four-quadrant detector used for laser deflection angle detection can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0150] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1A method for residual calibration of a four-quadrant detector for laser deflection angle detection is provided.

[0151] This instruction manual also provides Figure 8 The schematic diagram of the computer device shown is as follows: Figure 8 As shown, at the hardware level, this computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above. Figure 1 A method for residual calibration of a four-quadrant detector for laser deflection angle detection is provided.

[0152] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0153] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for residual calibration of a four-quadrant detector for laser deflection angle detection, characterized in that, include: Based on the output current of the four quadrants of the target surface of the four-quadrant detector, the normalized differential feature, diagonal differential feature, and power scale feature are calculated to form a response feature vector. Based on the normalized differential characteristics, the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector are determined, and the partition number of the spot on the target surface is determined based on the normalized response radius and response direction angle; based on the environmental interference parameters of the current four-quadrant detector and the physical parameters of the laser beam, the operating condition index corresponding to the spot is determined. Obtain the local residual model corresponding to the partition number and operating condition index, and input the response feature vector into the local residual model to obtain the target residual estimate; the target residual estimate reflects the local systematic error that the previous solver has not yet eliminated in the real system.

2. The method according to claim 1, characterized in that, Based on the normalized response radius and response direction angle, the zone numbering of the light spot on the target surface is determined, including: The radial region number of the light spot on the target surface is determined based on the normalized response radius; Based on the response direction angle, the sector number of the light spot on the target surface is determined; the circumference of the target surface is divided into multiple sectors according to the angle. The radial region number and sector number of the light spot on the target surface are encoded to obtain a unique partition number of the light spot on the target surface.

3. The method according to claim 2, characterized in that, The radial region includes the central region, effective region, transition region, and edge region; based on the normalized response radius, the radial region numbering of the light spot on the target surface is determined, including: If the normalized response radius is greater than or equal to 0 and less than the first preset radius threshold, then the light spot is determined to be in the central region of the target surface. If the normalized response radius is greater than or equal to the first preset radius threshold and less than the second preset radius threshold, then the effective area of ​​the light spot on the target surface is determined. If the normalized response radius is greater than or equal to the second preset radius threshold and less than the third preset radius threshold, then the transition zone of the light spot on the target surface is determined. If the normalized response radius is greater than or equal to the third preset threshold and less than or equal to 1, then the spot is determined to be in the edge region of the target surface. The radial region number of the light spot on the target surface is determined based on the radial region of the light spot on the target surface.

4. The method according to claim 1, characterized in that, Based on the environmental interference parameters and laser beam physical parameters of the current four-quadrant detector, the operating condition index corresponding to the light spot is determined, including: The environmental interference parameters and laser beam physical parameters are quantified and classified separately to obtain the level code of each parameter; the environmental interference parameters include the status of ambient lights and background noise; the laser beam physical parameters include optical power, attenuator transmittance, signal-to-noise ratio, spot size and spot morphology characteristics. Based on the level code corresponding to each parameter, a unique working condition index for the light spot is generated.

5. The method according to claim 1, characterized in that, Obtain the local residual model corresponding to the partition number and operating condition index, including: The partition number and the operating condition index are combined to form a two-dimensional joint index vector; Based on the two-dimensional joint index vector, a retrieval is performed in the pre-built residual model library; the residual model library uses each combination of partition number and working condition index as a unique storage address key value, and each combination of partition number and working condition index uniquely corresponds to a set of pre-stored local residual model parameters; different combinations correspond to different local residual model parameters. The retrieved local residual model parameters that match the two-dimensional joint index vector are loaded into the residual estimator to obtain a local residual model that uniquely matches the current spot spatial location and the current working state.

6. The method according to claim 5, characterized in that, The method further includes: When determining the partition number of the light spot, if the normalized response radius or response direction corner falls into the preset boundary transition zone, then the partition numbers adjacent to each other on both sides of the boundary are used as candidate partition numbers. Based on the two candidate partition numbers and the working condition index corresponding to the spot, the first local residual model and the second local residual model are determined. The response feature vector is input into the first local residual model and the second local residual model respectively to obtain the first residual estimate and the second residual estimate; The interpolation weights are determined based on the position of the current normalized response radius or response direction angle within the boundary transition zone, and the first residual estimate and the second residual estimate are weighted and interpolated to obtain the target residual estimate.

7. The method according to claim 5, characterized in that, The methods for constructing residual model libraries include: The sample output current of the four quadrants of the four-quadrant detector target surface was collected under different deflection angles, different environmental interference parameters, and different laser beam physical parameters. The sample output currents of the four quadrants are input into a pre-calibrated front-end solver to obtain the initial deflection angle of the laser beam relative to the four-quadrant detector. The sample residual is defined as the difference between the actual deflection angle and the initial deflection angle of the sample. Based on the sample output current in the four quadrants, construct the sample response feature vector, as well as the partition number and operating condition index for each sample. For each combination of partition number and operating condition index, a local residual model is constructed using the sample residuals and sample response feature vectors of the corresponding samples. Store the model parameters of the local residual model under each combination of partition number and operating condition index into the residual model library.

8. The method according to claim 7, characterized in that, Local residual models are constructed using weighted ridge regression, local linear models, low-order polynomial models, or lightweight neural networks.

9. The method according to claim 1, characterized in that, Based on the output currents in the four quadrants of the four-quadrant detector target surface, normalized differential characteristics, diagonal differential characteristics, and power scale characteristics are calculated to form a response feature vector, including: Calculate the sum of the output currents in the four quadrants to obtain the total photocurrent; Based on the output current in each quadrant, calculate the offset of the light spot in the X and Y directions on the target surface; Based on the total photocurrent, the offset of the light spot in the X and Y directions of the target surface is normalized to obtain the normalized differential characteristics of the light spot on the target surface of the four-quadrant detector. The diagonal current difference is determined based on the difference in current between the diagonal quadrants among the output currents in the four quadrants. The diagonal difference characteristics are obtained by normalizing the diagonal current difference using the total photocurrent. Logarithmic calculation of the total photocurrent yields power-scale characteristics; The normalized difference features, diagonal difference features, and power scale features are combined to form the response feature vector.

10. The method according to claim 1, characterized in that, Normalized difference features include normalized difference features in the X direction and normalized difference features in the Y direction; Based on the normalized difference characteristics, the normalized response radius and response direction angle of the laser spot on the target surface of the four-quadrant detector are determined, including: Calculate the square root of the sum of squares of the normalized difference features in the X direction and the normalized difference features in the Y direction, and divide by . The obtained value is determined as the normalized response radius; By inputting the normalized difference features in the vertical direction as the first variable and the normalized difference features in the horizontal direction as the second variable into the bivariate arctangent function, the response direction angle is obtained.

Images