Machine Vision-Based Optimization Method and System for Precise Positioning Control of Chip Insertion Pins

Abstract

This invention discloses a machine vision-based method and system for precise positioning control optimization of chip solder ball arrays, relating to the field of precise solder ball positioning technology. The method includes: obtaining initial optical coaxial uniformity based on multi-angle brightness distribution consistency analysis; generating several optical coaxial uniformities to be screened by neighborhood expansion sampling of the initial optical coaxial uniformity; mapping the optical coaxial uniformity to a control target group based on the illumination-imaging coupling response curve; performing solder ball positioning using the mapped optical coaxial uniformities to be screened; acquiring solder ball positioning data and generating uniformity scores based on machine learning training; constructing performance curves based on the optical coaxial uniformities to be screened and their corresponding uniformity scores; analyzing and selecting the optimal optical coaxial uniformity for solder ball positioning; and solving the problems of uneven brightness, blurred contours, and insufficient solder ball positioning accuracy in BGA solder ball array imaging.

Description

Technical Field

[0001] This invention relates to the field of pin precision positioning technology, and more specifically, to a method and system for optimizing pin precision positioning control based on machine vision. Background Technology

[0002] As chip packaging rapidly evolves from traditional metal-leaded packages like DIP and QFP to high-density BGA array packages, the bottom of the chip no longer has insertable metal leads. Instead, it uses regularly arranged tiny solder balls as contacts. Therefore, in surface mount and bonding processes, machine vision is needed for precise array identification, geometric positioning, and alignment control of the BGA solder ball array with the substrate pads. This process is known in industry as "array-type pin precision positioning." Existing machine vision-based chip pin precision positioning methods typically rely on steps such as bottom solder ball imaging, solder ball mesh structure extraction, solder ball center positioning, and array registration to achieve precise alignment between chip contacts and pads.

[0003] In machine vision positioning methods for precise positioning of array-type pins, optical imaging is the core underlying step. Optical coaxial uniformity affects the brightness consistency of coaxial illumination across the entire imaging field of view, directly determining the clarity of the solder balls' development, the stability of their reflection characteristics, and the complete representation of their edge contours. Because BGA solder balls have high reflectivity and a spherical mirror structure, insufficient coaxial uniformity can lead to bright spots, dark areas, or localized overexposure in the solder ball's reflective region, significantly reducing the accuracy of subsequent steps such as solder ball contour extraction, sub-pixel center positioning, and array orientation recognition.

[0004] Optical coaxial uniformity is neither always better the higher nor the lower the better; rather, there exists an optimal range suitable for BGA solder ball development. When coaxial uniformity is too low, the brightness difference between the solder ball's edge and the reflection center area widens, leading to instability in the visual model's judgment of the solder ball's geometric features. Conversely, when coaxial uniformity is too high, the reflection channels on the solder ball surface are enhanced, potentially resulting in overall overexposure or excessive smoothing of the reflection peak distribution, which in turn reduces the discriminability of the solder ball's center point and contour features. This means that different uniformity control levels will induce different types of positioning errors, collectively limiting the machine vision pin positioning accuracy.

[0005] Specifically, when the optical coaxial uniformity is low, uneven brightness or localized illumination attenuation will occur in the imaging field of view, preventing solder balls located at the edge of the field of view or in the array offset area from obtaining stable reflective bright spots and clear outlines. In this case, the grayscale gradient at the edge of the solder ball is discontinuous, causing deviations in edge detection and spherical contour fitting; furthermore, the sub-pixel center extraction results of low-brightness solder balls will show significant drift, resulting in systematic errors in the overall array orientation angle, array center centroid, and solder ball spacing measurements. Ultimately, when aligning the chip with the pads, alignment misalignment, solder ball misalignment, or localized warping can easily occur, reducing pin insertion accuracy.

[0006] When the optical coaxial uniformity is too high, the strong reflection from the solder ball surface will create a large area of ​​high brightness in the image. This causes a "spot diffusion" effect between the true outline of the solder ball and the distribution of reflection peaks, resulting in problems such as blurred edges, overexposure of the reflection center, and weakened boundaries between solder balls. In this case, the outline fitting of the solder balls will be distorted, the diameter will be identified as too small or too large, and even adjacent solder balls may be incorrectly identified as connected areas. Furthermore, excessively high uniformity will cause the overall brightness of the array to tend to be uniform, reducing the sensitivity of the vision system to local morphological differences (such as eccentricity, deformation, missing balls, etc.), affecting defect recognition and pin stability.

[0007] However, current technologies for controlling optical coaxial uniformity in BGA solder ball development mostly rely on fixed illumination parameters or simple increases or decreases in light intensity. They lack dynamic control methods for different array densities, solder ball reflectivity, and substrate absorbance, making it impossible to find the optimal balance between high and low uniformity to suit the solder ball imaging characteristics. As a result, machine vision inevitably encounters problems such as solder ball center drift, unstable array orientation recognition, and inaccurate contour segmentation when performing precise chip pin positioning, thus limiting overall positioning accuracy and process adaptability.

[0008] To address the above problems, this invention proposes a solution. Summary of the Invention

[0009] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a machine vision-based chip pin precision positioning control optimization method and system. By using quantization mapping based on optical coaxial uniformity and machine vision-based chip pin positioning optimization method, the problems of uneven brightness, blurred outlines, and insufficient pin positioning accuracy in BGA solder ball array imaging are solved.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A machine vision-based method for precise positioning and control optimization of chip inserts includes the following steps: obtaining initial optical coaxial uniformity based on multi-angle brightness distribution consistency analysis; generating several optical coaxial uniformities to be screened by neighborhood expansion sampling of the initial optical coaxial uniformity; mapping the optical coaxial uniformity to a control target group based on the illumination-imaging coupling response curve, wherein the control target group includes coaxial illumination numerical aperture adjustment and optical axis eccentricity compensation displacement; performing insert positioning by applying the mapped optical coaxial uniformities to be screened respectively; acquiring insert positioning data and generating uniformity scores based on machine learning training; constructing performance curves based on the optical coaxial uniformities to be screened and their corresponding uniformity scores, and analyzing and selecting the best optical coaxial uniformity for insert positioning.

[0012] A machine vision-based chip pin precision positioning control optimization system includes an optical coaxial uniformity initialization module, a neighborhood expansion sampling module, an optical coaxial uniformity control mapping module, a pin positioning module, a uniformity scoring module, and an optical coaxial uniformity screening module. The optical coaxial uniformity initialization module obtains initial optical coaxial uniformity based on multi-angle brightness distribution consistency analysis. The neighborhood expansion sampling module performs neighborhood expansion sampling on the initial optical coaxial uniformity to generate several optical coaxial uniformities to be screened. The optical coaxial uniformity control mapping module maps the optical coaxial uniformity to a control target group based on the illumination-imaging coupling response curve. The control target group includes coaxial illumination numerical aperture adjustment and optical axis eccentricity compensation displacement. The pin positioning module applies the mapped optical coaxial uniformities to be screened to perform pin positioning. The uniformity scoring module acquires pin positioning data and generates uniformity scores based on machine learning training. The optical coaxial uniformity screening module constructs performance curves based on the optical coaxial uniformities to be screened and their corresponding uniformity scores, and analyzes and screens the optimal optical coaxial uniformity for pin positioning.

[0013] The technical effects and advantages of the machine vision-based chip pin precise positioning control optimization method and system of this invention are as follows:

[0014] 1. This invention obtains an initial optical coaxial uniformity and further generates several optical coaxial uniformities to be screened based on neighborhood expansion sampling. Combined with the illumination-imaging coupling response curve, the optical coaxial uniformity is mapped to controllable actual optical parameters (including illumination numerical aperture adjustment and optical axis eccentricity compensation displacement), achieving a quantitative transformation from abstract optical indicators to operable controllable variables. Through this mapping method, the system can precisely adjust optical conditions for different chip arrays, solder ball reflection characteristics, and substrate light absorption, optimizing the uniformity of solder ball brightness within the imaging field of view. This significantly improves the clarity of solder ball outline development, edge gradient continuity, and reflection peak distribution stability. This technique effectively solves the problem that traditional methods of fixing illumination parameters or simply increasing or decreasing light intensity cannot dynamically adapt to different optical conditions, achieving controllable and customized optimization of optical coaxial uniformity.

[0015] 2. This invention applies several optical coaxial uniformities to chip pin positioning, and collects pin imaging data, extracts center offset features, contour gradient features, reflection peak features, and contrast features. A uniformity score is generated based on a machine learning model, and performance curves are constructed for analysis and selection to obtain the optimal optical coaxial uniformity and corresponding control target parameters. This method finds the optimal balance between high and low uniformity, maximizing the accuracy of solder ball sub-pixel center positioning, array direction recognition accuracy, and contour fitting reliability, thereby significantly improving the overall stability and accuracy of chip pin positioning under machine vision. This technical solution solves the problems of solder ball edge drift, reflection peak overexposure, and unstable array recognition caused by fixed or coarse optical conditions in existing technologies, and is suitable for high-precision pin positioning scenarios of high-density BGA packaged chips. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the chip pin precision positioning control optimization method based on machine vision according to the present invention.

[0017] Figure 2 This is a schematic diagram of the structure of the chip pin precision positioning control optimization system based on machine vision according to the present invention. Detailed Implementation

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

[0019] Example 1, Figure 1The present invention provides a machine vision-based method for precise positioning and control optimization of chip pins, comprising the following steps:

[0020] S1, Initial optical coaxial uniformity is obtained based on multi-angle brightness distribution consistency analysis;

[0021] In this embodiment, obtaining the initial optical coaxial uniformity based on multi-angle brightness distribution consistency analysis specifically involves:

[0022] Acquire the brightness distribution image of the light source under a preset angle set, wherein the angle set consists of multiple observation angles acquired with the optical axis as the center at a fixed angle step;

[0023] For each observation angle, calculate the radial gradient sequence at the corresponding angle in the brightness distribution image, and generate a set of radial gradient sequences indexed by angle;

[0024] Calculate the relative amplitude of the radial gradient sequence corresponding to each angle in the radial gradient sequence set, and generate a radial gradient relative amplitude sequence indexed by angle;

[0025] The maximum relative amplitude is selected from the radial gradient relative amplitude sequence to generate the radial gradient dispersion.

[0026] In the brightness distribution image corresponding to each observation angle, select the circumferential brightness value at a preset radius to generate a circumferential brightness sequence;

[0027] Calculate the relative circumferential amplitude of the circumferential brightness sequence to generate the circumferential brightness variability.

[0028] The initial optical coaxial uniformity is obtained based on the analysis of radial gradient dispersion and circumferential brightness fluctuation.

[0029] In this embodiment, the relative amplitude of the radial gradient is used to describe the fluctuation of brightness changes within the radial gradient sequence. It is obtained as follows: First, the position with the most significant brightness change in the radial gradient sequence is found, and the gradient amplitude at that position is recorded. Then, the position with the most gradual change in the same sequence is found, and the gradient amplitude at that position is recorded. Finally, the difference between the two amplitudes is used as the relative amplitude of the sequence. The larger the difference, the more significant the discontinuous change in illumination in the radial direction at that observation angle; the smaller the difference, the more stable and consistent the brightness change with the radial direction.

[0030] In this embodiment, the radial gradient dispersion is used to describe the degree of instability in radial brightness variation across all observation angles. It is obtained by first selecting a value representing a significant brightness change from the relative amplitude of the radial gradient at each observation angle; then, finding the largest value among these values ​​for all angles, and taking this maximum value as the most significant instability in the radial direction of the light field. This dispersion reflects the abrupt change in radial brightness of the light source in the most unfavorable direction and is a key indicator for evaluating the radial consistency of the light field.

[0031] In this embodiment, the circumferential brightness fluctuation is used to describe whether the brightness of the light source is uniform in the circumferential direction. It is obtained as follows: First, samples are taken point-by-point along the circumference at a preset radius of the brightness distribution image, forming a brightness sequence arranged according to angular variations. Then, the positions with the highest and lowest brightness are found in this sequence, and their brightness values ​​are recorded respectively. Finally, the difference between the two brightness values ​​is used as the fluctuation characteristic of that position. Combined with the continuity of brightness changes in the sequence, the difference with the most significant brightness change is taken as the circumferential brightness fluctuation. A larger value indicates that there are significant brightness fluctuations in the light field in the circumferential direction, while a smaller value indicates that the circumferential brightness of the light field is more uniform.

[0032] In this embodiment, the brightness distribution image is the base image used to carry the above data. It is acquired by the imaging device at different observation angles, and the brightness value serves as the direct source for all gradient, amplitude, and fluctuation calculations.

[0033] In this embodiment, the set of observation angles is used to provide data sources for multiple illumination directions. The step size between each angle is consistent, so that the radial and circumferential variation trends can be compared across angles, ensuring that the radial gradient relative amplitude and circumferential brightness fluctuation have a consistent data reference structure.

[0034] In this embodiment, optical coaxial uniformity is the degree of light field consistency obtained by comprehensively analyzing radial gradient dispersion and circumferential brightness fluctuation. It is formed by combining the most significant brightness abrupt change in the radial direction with the most obvious brightness fluctuation in the circumferential direction to evaluate the overall brightness balance of the light source in different spatial directions. When both indicators are at low levels, the coaxial uniformity is high; when either indicator is large, the coaxial uniformity is low. This uniformity is used as a comprehensive measure of illumination quality for subsequent pin positioning illumination control.

[0035] It should be noted that the above three values ​​(radial gradient relative amplitude, radial gradient dispersion, and circumferential brightness fluctuation) are all obtained by comparing the image brightness changes. The whole process does not depend on the absolute luminous flux, but on the relative trend of brightness changes in the image. Therefore, it can avoid the influence of instantaneous power fluctuations of the light source on the results and make the calculation results more stable.

[0036] It should be noted that, in order to ensure the robustness of the above features, the brightness distribution image can be sampled with a fixed step size, a fixed radius, or repeatedly sampled in practice to eliminate the interference of local outliers, making the obtained radial gradient dispersion and circumferential brightness fluctuation more stable and reliable, thereby making the calculation of optical coaxial uniformity more adaptable to different environmental conditions.

[0037] In this embodiment, the initial optical coaxial uniformity is obtained based on the analysis of radial gradient dispersion and circumferential brightness fluctuation, specifically as follows:

[0038] The radial gradient dispersion and circumferential brightness fluctuation are converted into brightness deviation pairs, which consist of the radial deviation corresponding to the radial gradient dispersion and the circumferential deviation corresponding to the circumferential brightness fluctuation.

[0039] A brightness deviation feature sequence is generated based on the brightness deviation amount, and the brightness deviation feature sequence is formed by arranging the radial deviation amount and the circumferential deviation amount in a preset order;

[0040] The component with the larger deviation in the brightness deviation feature sequence is selected as the main deviation, and the main deviation is obtained.

[0041] Substitute the principal deviation into the preset deviation grading interval to obtain the grading position of the principal deviation in the deviation grading interval.

[0042] The corresponding uniformity percentage is determined based on the grade position of the main deviation, and used as the initial optical coaxial uniformity.

[0043] In this embodiment, the brightness deviation pair refers to the integration of radial gradient dispersion and circumferential brightness fluctuation into a set of numerical pairs describing the optical field deviation in a one-to-one correspondence. The numerical pair consists of radial deviation, which represents the degree of difference in the radial direction, and circumferential deviation, which represents the degree of difference in the circumferential direction. The two together characterize the deviation of the optical field in two orthogonal directions under the same observation conditions, and serve as a unified data carrier for subsequent sorting and hierarchical mapping.

[0044] In this embodiment, the radial deviation refers to a comparable value obtained from the radial gradient dispersion to represent the intensity of the radial brightness deviation. The steps for obtaining it are as follows: for each observation angle, the radial gradient sequence first locates the position representing significant change in the sequence and records the local change amplitude at that position. Then, the position representing stable change is located in the same sequence and the local change amplitude at that position is recorded. Subsequently, the difference between the two is used as the radial deviation for that angle. Among the radial deviations of all angles, the value that can represent the overall radial non-uniformity is selected as the source of the radial gradient dispersion. This process is based on the relative change of image pixel brightness and does not depend on external absolute measures.

[0045] In this embodiment, the circumferential deviation refers to a comparable value obtained by converting the circumferential brightness fluctuation to represent the intensity of the brightness deviation in the circumferential direction. The acquisition steps are as follows: the circumferential brightness sequence is obtained by sampling along the circumferential direction at a preset radius, the highest point and the lowest point in the sequence are identified and their brightness difference is recorded, and the representative fluctuation amplitude is extracted by combining the continuity characteristics of local fluctuations in the circumferential sequence. Finally, the representative fluctuation amplitude is used as the circumferential deviation to reflect the non-uniformity of the light field in the circumferential direction.

[0046] In this embodiment, the brightness deviation feature sequence refers to an ordered data column formed by splicing radial deviation and circumferential deviation in a preset order. The preset order is to list the radial deviation first and then the circumferential deviation, or to arrange them cyclically according to the observation angle, forming a unified sequence structure so as to perform unified size comparison and identification of the main deviation. Each item in the sequence clearly corresponds to the source direction and sampling angle, which is convenient for subsequent tracing and fault location.

[0047] In this embodiment, the principal deviation refers to the item that represents the most significant deviation of the overall light field, selected from the brightness deviation feature sequence by comparing each item. The selection step is to perform a one-way comparison of the value of each item in the brightness deviation feature sequence to find the component with the largest value. This largest component is regarded as the principal deviation and used as the input for subsequent hierarchical mapping. The principal deviation can come from either the radial deviation or the circumferential deviation. The selection rules are clear and do not introduce artificial weights.

[0048] In this embodiment, the deviation grading interval refers to a series of continuous partitions pre-defined for the main deviation quantity, used to discretize the main deviation quantity according to its position in all samples. The interval consists of several non-overlapping grade segments, each corresponding to a specific deviation grade description. The deviation grading interval is set based on historical verification data and imaging acceptability criteria, and grading is achieved through discrete segment boundaries given in this embodiment. The grading process involves comparing the main deviation quantity with the preset segment boundaries segment by segment to determine its corresponding segment.

[0049] In this embodiment, the uniformity percentage refers to converting the deviation level determined within the deviation grading interval into an intuitive percentage form to represent the initial optical coaxial uniformity. The conversion step is to first determine the grading segment where the main deviation amount is located, then determine the corresponding percentage interval according to the sequence position of the segment in all grading segments, and determine a unique percentage value within the interval according to the segment position as the final uniformity output. This percentage value reflects the consistency level of the light field under the current detection conditions.

[0050] It should be noted that the boundary values ​​of the deviation grading intervals and the number of grade segments need to be determined through a preliminary calibration procedure during implementation. This calibration includes collecting sample data under different lighting and device conditions, performing the above-mentioned deviation extraction process on the sample data, and recording the correspondence with the actual positioning performance. This determines the mapping relationship between each grade segment and imaging acceptability, thereby making it possible to directly map the main deviation amount to the uniformity percentage using the grading rules in the subsequent process with engineering reproducibility.

[0051] It should be noted that, in order to improve the robustness of the above-mentioned mapping between deviation and uniformity, the brightness distribution image should be repeatedly sampled with a fixed step size and averaged over multiple frames during implementation. Abnormal single-frame values ​​should be removed or smoothed before sequence generation and selection of the main deviation to reduce the impact of noise and occasional events on deviation determination. The original source information of each step should be retained for backtracking analysis when necessary.

[0052] S2, perform neighborhood expansion sampling on the initial optical coaxial uniformity to generate several optical coaxial uniformities to be screened;

[0053] In this embodiment, the step of performing neighborhood expansion sampling on the initial optical coaxial uniformity to generate several optical coaxial uniformities to be screened specifically involves:

[0054] The neighborhood expansion step size set is determined based on the initial optical coaxial uniformity. The neighborhood expansion step size set includes multiple expansion step sizes generated in an arithmetic manner based on a preset minimum step size.

[0055] The initial optical coaxial uniformity is added to and subtracted from the positive and negative expansion step sizes in the neighborhood expansion step size set to generate a sequence of candidate uniformity values ​​arranged by step size index.

[0056] Apply a preset uniformity effective interval constraint to the candidate uniformity value sequence, remove candidate uniformity values ​​that exceed the uniformity effective interval, and generate an effective candidate uniformity sequence.

[0057] The valid candidate uniformity sequence is checked for duplicate values ​​and sorted in ascending order to generate an ordered candidate uniformity sequence.

[0058] Each uniformity value in the ordered candidate uniformity sequence is combined with the initial optical coaxial uniformity to form a neighborhood extended uniformity set. Each uniformity value in the neighborhood extended uniformity set is an optical coaxial uniformity to be screened.

[0059] In this embodiment, the neighborhood expansion step size set refers to a set of step size values ​​generated in the positive and negative directions with the initial optical coaxial uniformity as the center and a preset minimum step size as the reference. The step size is increased sequentially in an arithmetic manner to form a set of reference values ​​that can be used to fine-tune the initial uniformity. This is used to form a sampling space near the uniformity to ensure that subsequent screening can cover the possible offset range of uniformity.

[0060] In this embodiment, the forward expansion step size refers to the step size value that shifts from the initial optical coaxial uniformity to a higher uniformity direction. It is obtained by selecting a step size value greater than zero from the neighborhood expansion step size set and adding it to the initial uniformity one by one to generate a candidate uniformity value sequence that is higher than the initial value. This is used to evaluate the impact of the increase in uniformity on the pin positioning performance.

[0061] In this embodiment, the negative expansion step size refers to the step size value that shifts from the initial optical coaxial uniformity to a lower uniformity. It is obtained by selecting a step size value less than zero from the neighborhood expansion step size set and adding it to the initial uniformity one by one or subtracting it directly to generate a candidate uniformity value sequence that is lower than the initial value. This is used to evaluate the impact of the reduction in uniformity on the pin positioning performance.

[0062] In this embodiment, the candidate uniformity value sequence refers to the ordered numerical sequence formed by combining the initial optical coaxial uniformity with all positive and negative expansion step sizes. The sequence is arranged in step size order, and each value represents a possible optical coaxial uniformity to be screened, which is used for subsequent validity checks and sorting.

[0063] In this embodiment, the effective uniformity range refers to the preset range of allowable optical coaxial uniformity values, which is used to eliminate candidate uniformity values ​​that are too high or too low and may cause unstable pin positioning performance. The determination is based on the imaging tolerance of the optical system, the solder ball development characteristics, and historical experimental data. All candidate uniformity values ​​that exceed this range are eliminated.

[0064] In this embodiment, the effective candidate uniformity sequence refers to the candidate uniformity value sequence retained after screening the effective uniformity interval. These values ​​are all within the allowable range and can be used for subsequent sorting, duplicate value removal, and generation of neighborhood expansion uniformity sets.

[0065] In this embodiment, the neighborhood expansion uniformity set refers to the set formed by combining each value in the effective candidate uniformity sequence with the initial optical coaxial uniformity. Each item in the set represents a single optical coaxial uniformity to be screened, which is used for subsequent pin positioning experiments and uniformity scoring to ensure sufficient test coverage in the vicinity of the initial uniformity.

[0066] It should be noted that when generating the neighborhood expansion step size set, it can be fine-tuned according to the sensitivity and imaging resolution of the optical system to ensure that the step size can reflect the impact of small changes in uniformity on positioning performance without excessively increasing the number of experiments, thereby improving experimental efficiency and data quality.

[0067] It should be noted that, in order to further improve the representativeness of the neighborhood expansion uniformity set, random perturbation or uniform sampling can be added during the step size selection process, so that the generated uniformity to be screened covers more possible optical field deviations, so that the machine learning model can fully capture the various effects of uniformity on pin positioning during the training process.

[0068] S3, based on the illumination-imaging coupling response curve, the optical coaxial uniformity is mapped to a control target group, the control target group including the coaxial illumination numerical aperture adjustment amount and the optical axis eccentricity compensation displacement amount;

[0069] In this embodiment, the mapping of optical coaxial uniformity to the control target group based on the illumination-imaging coupling response curve specifically involves:

[0070] Several numerical aperture adjustment values ​​are obtained from the illumination end as numerical aperture sampling points;

[0071] Several optical axis eccentricity compensation displacements are obtained from the imaging end as eccentricity displacement sampling points;

[0072] A set of illumination-imaging sampling points arranged in two-dimensional indexes is generated based on numerical aperture sampling points and eccentric displacement sampling points;

[0073] Adjust the illumination numerical aperture and optical axis eccentricity compensation displacement at each sampling point in the illumination-imaging sampling point set, and acquire the corresponding imaging brightness distribution image to generate a brightness distribution image set indexed by sampling point.

[0074] For each brightness distribution image in the brightness distribution image set, obtain the radial brightness response value and the circumferential brightness response value, and combine the two to form a coaxial response value, generating a set of coaxial response values ​​indexed by sampling point;

[0075] Two-dimensional interpolation is performed based on the set of illumination-imaging sampling points and the corresponding set of coaxial response values ​​to generate an illumination-imaging coupling response curve with the numerical aperture adjustment of illumination and the optical axis eccentricity compensation displacement as input variables and the coaxial response value as the output variable.

[0076] Each optical coaxial uniformity to be screened in the neighborhood extended uniformity set is associated with the coupling response curve to determine the coaxial response value corresponding to the corresponding optical coaxial uniformity. The illumination numerical aperture adjustment amount and optical axis eccentricity compensation displacement amount that make the coaxial response value consistent with the corresponding optical coaxial uniformity are found in the coupling response curve to generate a set of control pairs indexed by uniformity.

[0077] The control targets are combined to form a set of control targets, which includes the numerical aperture adjustment of illumination and the eccentricity compensation displacement of the optical axis.

[0078] In this embodiment, the numerical aperture sampling points refer to a series of adjustable optical numerical aperture values ​​obtained from the illumination end. Each sampling point represents the light cone angle output by the light source under a specific aperture setting, which is used to evaluate the influence of different light cone angles on the imaging brightness distribution and optical coaxial uniformity. These sampling points are repeatedly adjustable through mechanical or electronically controlled optical elements.

[0079] In this embodiment, the eccentric displacement sampling points refer to a series of optical axis eccentricity compensation displacements obtained from the imaging end. Each sampling point represents a small displacement adjustment of the camera or imaging optical system relative to the optical axis, which is used to simulate and compensate for changes in imaging brightness distribution caused by optical axis misalignment or lens eccentricity. These displacements are applied through an adjustable mechanical platform or a micro-drive mechanism.

[0080] In this embodiment, the illumination-imaging sampling point set refers to a set composed of all numerical aperture sampling points and eccentric displacement sampling points arranged and combined according to a two-dimensional index. Each sampling point represents an independent combination of illumination and imaging states, which is used to systematically collect the imaging brightness response under various states to ensure comprehensive coverage of optical characteristics.

[0081] In this embodiment, the brightness distribution image set refers to the collection of imaging brightness images acquired at each illumination-imaging sampling point. Each image reflects the light intensity distribution under the current illumination numerical aperture and optical axis eccentricity compensation state, and is used for subsequent extraction of radial brightness response and circumferential brightness response.

[0082] In this embodiment, the radial brightness response value refers to the brightness variation collected from the brightness distribution image along the radial direction from the center of the optical axis. It is obtained by analyzing the brightness value variation at each radial position and is used to reflect the uniformity and attenuation characteristics of the beam in the radial direction.

[0083] In this embodiment, the circumferential brightness response value refers to the brightness variation collected along a circular path centered on the optical axis in the brightness distribution image. It is obtained by statistically analyzing the brightness value changes at each position in the circumferential direction and is used to reflect the uniformity fluctuation of the light field in the circumferential direction.

[0084] In this embodiment, the coaxial response value refers to the comprehensive quantity formed by combining the radial brightness response value and the circumferential brightness response value. Each sampling point corresponds to a coaxial response value, which is used to represent the overall optical coaxial uniformity characteristics under the current illumination-imaging state.

[0085] In this embodiment, the illumination-imaging coupling response curve refers to the curve or surface formed by mapping the illumination numerical aperture adjustment and optical axis eccentricity compensation displacement of the sampling point to the corresponding coaxial response value through a two-dimensional interpolation method. It is used to look up the illumination and imaging adjustment parameters from the specified optical coaxial uniformity.

[0086] In this embodiment, the set of control pairs refers to the combination of illumination numerical aperture adjustment amount and optical axis eccentricity compensation displacement amount that makes the coaxial response value consistent with the target uniformity for each optical coaxial uniformity to be screened. Each combination is a control pair used to precisely control the optical state.

[0087] In this embodiment, the control target group refers to the set formed by combining the corresponding illumination numerical aperture adjustment amount and optical axis eccentricity compensation displacement amount in all control pairs. It is used to directly set the optical system in experiments or production to achieve precise control of the coaxial uniformity of the optical system to be screened.

[0088] It should be noted that, in order to ensure the accuracy of the illumination-imaging coupling response curve, the sampling points should be selected to cover the possible optical adjustment range of the system and be distributed as evenly as possible, so as to fully reflect the impact of changes in optical coaxial uniformity on the imaging effect under different combination states, while also considering the mechanical adjustment accuracy and repeatability.

[0089] It should be noted that the generation of control target groups can not only be used for single optical uniformity control, but also be extended to the basis for automatic optimization of system parameters. By combining multiple control targets to form a control strategy, the machine vision system can quickly adjust and stably position the pin under different uniformity conditions, thereby further improving the overall performance and adaptability of the system.

[0090] S4, apply the mapped optical coaxial uniformity of the target to perform pin positioning respectively;

[0091] In this embodiment, the step of applying the mapped optical coaxial uniformity to be screened to perform pin positioning specifically involves:

[0092] The numerical aperture adjustment amount and optical axis eccentricity compensation displacement corresponding to the optical coaxial uniformity to be screened are obtained respectively, and the two are applied to form a positioning light field for the corresponding uniformity.

[0093] Under the positioning light field, the full field of view of the array of pins is exposed point by point, and the corresponding pin imaging frame sequence is acquired to generate a pin imaging sequence for corresponding optical coaxial uniformity.

[0094] Background suppression and brightness normalization are performed on the pin imaging sequence to obtain a pin contour image sequence with uniform brightness;

[0095] Extract the pin boundaries in each pin outline image, perform sub-pixel center fitting based on the boundary shape to obtain the center coordinates of each pin, and form a set of pin center coordinates for corresponding uniformity.

[0096] Based on the set of pin center coordinates and the standard arrangement coordinates of the array pins, the center offset of each pin is calculated point by point to generate a set of pin offsets;

[0097] Identify the maximum offset, average offset, and offset distribution in the set of pin offsets to complete the positioning.

[0098] In this embodiment, the positioning light field refers to the illumination distribution state formed after the actual adjustment of the illumination numerical aperture adjustment amount and the optical axis eccentricity compensation displacement amount corresponding to the mapped optical coaxial uniformity to be screened in the optical system. This light field covers the entire array of pins and is used to ensure that each pin is uniformly and controllably illuminated in the field of view during the imaging process.

[0099] In this embodiment, point-by-point exposure refers to sequentially performing imaging acquisition operations on each position or sub-region of the array of pins under the positioning light field to ensure that the image information of each pin is completely recorded, generating a continuous pin imaging frame sequence for subsequent extraction of contours and center coordinates.

[0100] In this embodiment, the pin imaging sequence refers to a set of image frames obtained by point-by-point exposure. Each frame corresponds to the imaging state of a certain pin or combination of pins in the array. This sequence reflects the brightness distribution and contour features of the pins under different optical uniformity conditions.

[0101] In this embodiment, background suppression refers to removing brightness interference from non-pin areas in the field of view during the pin imaging sequence. By analyzing the gray level of the image background and performing subtraction or smoothing, the brightness features of the pin area are made more prominent, which facilitates accurate contour extraction.

[0102] In this embodiment, brightness normalization processing refers to uniformly scaling the pixel brightness values ​​of each pin outline image to ensure that the overall brightness level of different images is consistent, thereby eliminating the influence of differences in optical systems or exposure conditions on pin outline feature analysis.

[0103] In this embodiment, the pin outline image refers to the image after background suppression and brightness normalization processing, in which the pin boundaries are clearly distinguishable, and is used to extract the shape and position features of the pin from the image.

[0104] In this embodiment, subpixel center fitting refers to refining the boundary point positions and performing shape fitting based on the extracted pin boundaries to obtain the center coordinates of each pin with subpixel accuracy, thereby improving positioning accuracy.

[0105] In this embodiment, the set of pin center coordinates refers to the set formed by summing the center coordinates of all pins in the array through sub-pixel fitting. Each center coordinate is used to describe the precise position of the pin in the field of view.

[0106] In this embodiment, the standard arrangement coordinates refer to the ideal arrangement position coordinates specified in the design or manufacturing process of the array pins. These coordinates are used to compare with the actual measured center coordinates of the pins and to evaluate the pin positioning error.

[0107] In this embodiment, the pin offset set refers to the set formed by summing the offsets between the center coordinates of each pin and the standard layout coordinates. By analyzing this set, the degree and direction of deviation of each pin in the field of view can be obtained.

[0108] It should be noted that when generating the set of pin offsets, the local deviations that may be caused by uneven illumination in the array areas where different pins are located should be considered. Therefore, the entire field of view needs to be sampled and the offset characteristics of each pin need to be recorded for subsequent uniformity scoring.

[0109] It should be noted that this step is not only used to evaluate the positioning effect of each optical coaxial uniformity to be screened individually, but also to provide input features for subsequent machine learning models. By analyzing the combination of maximum offset, average offset and offset distribution, the influence of optical uniformity on the positioning stability and accuracy of the insert can be quantified, and closed-loop control of optical parameter optimization and insert accuracy improvement can be achieved.

[0110] S5, acquire pin positioning data, and generate uniformity scores based on machine learning training;

[0111] In this embodiment, the step of acquiring pin positioning data and generating a uniformity score based on machine learning training specifically involves:

[0112] Based on the set of pin offsets, the pin center offset features are extracted to obtain the maximum offset, average offset, and offset variance used to characterize the degree of offset diffusion under low uniformity conditions, forming a center offset feature set.

[0113] The contour gradient magnitude of each pin is calculated based on the pin contour image sequence, and the minimum, maximum and gradient magnitude change rate of the gradient magnitude within the entire field of view are statistically analyzed to obtain a contour gradient feature set used to characterize the discontinuity of contour gray-level gradient caused by low uniformity.

[0114] In each pin profile image, the brightness value of the solder ball reflection peak region is measured, and the peak area, peak overexposure ratio and brightness difference between adjacent peaks are calculated to form a reflection peak feature group to characterize the degree of overexposure and spot diffusion in the reflection area caused by high uniformity.

[0115] The contrast between solder balls is calculated based on the brightness difference between adjacent solder balls in the full field of view pin outline image. The average contrast and contrast decay rate are statistically analyzed in each image to generate a contrast feature group to characterize the weakening of solder ball boundaries caused by high uniformity.

[0116] The center offset feature group, contour gradient feature group, reflection peak feature group, and contrast feature group are combined in a fixed order to form a positioning feature vector for this uniformity.

[0117] Collect all the positioning feature vectors corresponding to the optical coaxial uniformity to be screened, construct a positioning feature training set, and label each feature vector with positioning quality labels based on pin offset stability and contour fidelity to generate a positioning quality label set.

[0118] A regression model is trained based on a training set of positioning features and a set of positioning quality labels. The weighting coefficients of each positioning feature in the uniformity score are determined through training, resulting in a machine learning model for calculating the optical coaxial uniformity score. The machine learning model is a weighted form of all features, and the training object is the weight value of each feature. The trained machine learning model outputs the uniformity score.

[0119] In this embodiment, the pin center offset feature refers to the feature quantity extracted from the pin offset set that describes the degree of position deviation of the pin within the field of view, including the maximum offset, the average offset, and the offset variance. These features can reflect the degree of diffusion and non-uniformity of the pin arrangement under low optical coaxial uniformity conditions and are used to quantify the stability of the center position.

[0120] In this embodiment, the contour gradient feature refers to the gradient magnitude calculated based on the gray-scale change of each pin boundary in the pin contour image sequence. The minimum, maximum and gradient magnitude change rate are statistically analyzed in the entire field of view to reflect the discontinuity of the contour gray-scale gradient under low uniformity conditions, thereby characterizing the reliability of edge recognition and contour fitting.

[0121] In this embodiment, the reflection peak feature refers to measuring the brightness value of the bright area on the surface of the solder ball in each pin outline image, and obtaining the peak area, overexposure ratio, and brightness difference between adjacent peaks. These features are used to characterize the overexposure of the reflection area and the light spot diffusion phenomenon that may occur under high uniformity conditions, which affects the resolvability of the center and outline of the solder ball.

[0122] In this embodiment, the contrast feature refers to the feature formed by calculating the brightness difference between adjacent solder balls in the full field of view pin outline image and statistically analyzing the average value and attenuation rate. This feature is used to reflect the problem of solder ball boundary weakening and local brightness difference reduction caused by high uniformity, thereby evaluating the outline resolution capability.

[0123] In this embodiment, the positioning feature vector refers to the feature vector formed by combining the center offset feature group, the contour gradient feature group, the reflection peak feature group, and the contrast feature group in a fixed order. It is used to comprehensively describe the combined effect of the optical coaxial uniformity to be screened on the positioning of the insert, and serves as the input to the machine learning model.

[0124] In this embodiment, the positioning feature training set refers to the set formed by collecting all positioning feature vectors corresponding to the optical coaxial uniformity to be screened. Each feature vector is labeled with a corresponding positioning quality label, which is determined based on the pin offset stability and contour fidelity, and serves as the data basis for training the machine learning regression model.

[0125] In this embodiment, the machine learning model refers to a model constructed in the form of regression. By training the positioning feature training set and the positioning quality label set, it automatically learns the weight coefficient of each feature in the uniformity score, thereby outputting a uniformity score for each optical coaxial uniformity, which is used to quantify the overall impact of uniformity on the pin positioning accuracy.

[0126] It should be noted that when training machine learning models, it is important to ensure that the features cover different types of localization defects under both low and high uniformity conditions. This is so that the model can simultaneously reflect influencing factors such as offset diffusion, contour grayscale discontinuity, overexposure, and boundary weakening, thereby generating a comprehensive and stable uniformity score.

[0127] It should be noted that this step not only provides a basis for subsequent uniformity screening, but also allows us to understand the sensitivity of each feature to positioning accuracy through model weight analysis, guiding the adjustment and optimization of the optical system, and achieving synergistic optimization of optical coaxial uniformity and pin accuracy.

[0128] S6. Based on the optical coaxial uniformity to be screened and the corresponding uniformity score, construct a performance curve, analyze and screen the best optical coaxial uniformity for pin positioning.

[0129] In this embodiment, the step of constructing a performance curve based on the optical coaxial uniformity to be screened and the corresponding uniformity score, and analyzing and screening the optimal optical coaxial uniformity for pin positioning, specifically involves:

[0130] All the optical coaxial uniformity to be screened are sorted by numerical value to form a uniformity sequence arranged in ascending order of uniformity.

[0131] Based on the order of the uniformity sequence, the corresponding uniformity scores are arranged in the same order to form a score sequence;

[0132] A performance curve is constructed to characterize the relationship between uniformity changes and score changes, with the uniformity sequence as the horizontal axis and the score sequence as the vertical axis.

[0133] Calculate the difference between adjacent scores in the performance curve, identify the position of the maximum gain of the score difference, and determine the uniformity corresponding to the position as the candidate optimal uniformity;

[0134] Read the corresponding uniformity scores at the previous and next uniformity near the candidate optimal uniformity to form a candidate interval score group;

[0135] Compare the scores in the candidate interval score group and select the uniformity with the highest score as the final optimal optical coaxial uniformity.

[0136] The final optimal optical coaxial uniformity is input into the illumination-imaging coupling response curve to obtain the corresponding illumination numerical aperture adjustment amount and optical axis eccentricity compensation displacement amount, thereby generating the optimal positioning control target for chip pin positioning.

[0137] In this embodiment, the uniformity sequence refers to an ordered sequence formed by arranging all the optical coaxial uniformities to be screened from low to high values. This sequence is used to clarify the horizontal position of different uniformity levels in the performance curve and ensure that the scoring sequence corresponds one-to-one with the uniformity value.

[0138] In this embodiment, the scoring sequence refers to the set of uniformity scores corresponding to each uniformity, arranged in order according to the uniformity sequence. Each score reflects the comprehensive influence of the pin positioning feature on the overall positioning accuracy under the uniformity condition, and is used to represent the longitudinal trend on the performance curve.

[0139] In this embodiment, the performance curve refers to a curve constructed with the uniformity sequence as the horizontal axis and the scoring sequence as the vertical axis. It is used to visually demonstrate the influence of uniformity changes on the pin positioning accuracy. By observing the curve shape, the sensitive range and potential optimal point for uniformity optimization can be identified.

[0140] In this embodiment, the adjacent score difference refers to the difference in scores corresponding to adjacent uniformity in the performance curve. It is used to quantify the performance improvement or decline caused by uniformity changes. Identifying the position of maximum gain helps to find the interval where uniformity changes have the most significant effect on improving positioning accuracy.

[0141] In this embodiment, the candidate optimal uniformity refers to the uniformity value determined by analyzing the position with the largest score difference in the performance curve. This uniformity is considered to be the optical coaxial uniformity that may provide the best positioning performance and is used to further refine the final optimal value.

[0142] In this embodiment, the candidate interval score group refers to the set of uniformity scores read at the adjacent uniformity positions before and after the candidate optimal uniformity. It is used to compare the score size within a local range to ensure that the finally selected uniformity has the highest comprehensive positioning performance.

[0143] In this embodiment, the final optimal optical coaxial uniformity refers to selecting the uniformity with the highest score in the candidate interval scoring group, which is used as the optimal control optical condition for chip pin positioning. This uniformity can maximize the pin center stability, outline clarity, and boundary discernibility.

[0144] In this embodiment, the optimal positioning control target refers to the combination parameters of the corresponding illumination numerical aperture adjustment amount and optical axis eccentricity compensation displacement obtained after inputting the final optimal optical coaxial uniformity into the illumination-imaging coupling response curve. These parameters are used to generate a positioning light field for the chip insert, ensuring that the insert positioning accuracy reaches the optimal level.

[0145] It should be noted that when constructing performance curves, the uniformity sequence and scoring sequence should be designed to fully cover the entire range from low to high uniformity, so that the performance curves can comprehensively reflect the trend of positioning accuracy changes under different uniformity conditions and avoid missing key optimal intervals.

[0146] It should be noted that the performance curve is not only used to screen for the final optimal uniformity, but also as a reference for the debugging and optimization of the optical system. By analyzing the curve shape, the neighborhood expansion step size or optical axis eccentricity compensation strategy can be adjusted to achieve synergistic optimization of optical coaxial uniformity and pin positioning accuracy.

[0147] Example 2, Figure 2 This invention presents a machine vision-based chip pin precision positioning control optimization system, comprising an optical coaxial uniformity initialization module, a neighborhood expansion sampling module, an optical coaxial uniformity control mapping module, a pin positioning module, a uniformity scoring module, and an optical coaxial uniformity screening module. The optical coaxial uniformity initialization module obtains initial optical coaxial uniformity based on multi-angle brightness distribution consistency analysis. The neighborhood expansion sampling module generates several optical coaxial uniformities to be screened by expanding the neighborhood sampling of the initial optical coaxial uniformity. The optical coaxial uniformity control mapping module maps the optical coaxial uniformity to a control target group based on the illumination-imaging coupling response curve, wherein the control target group includes coaxial illumination numerical aperture adjustment and optical axis eccentricity compensation displacement. The pin positioning module performs pin positioning by applying the mapped optical coaxial uniformities to be screened. The uniformity scoring module acquires pin positioning data and generates uniformity scores based on machine learning training. The optical coaxial uniformity screening module constructs performance curves based on the optical coaxial uniformities to be screened and their corresponding uniformity scores, and analyzes and selects the optimal optical coaxial uniformity for pin positioning.

[0148] In the embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0149] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0150] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0151] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[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. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0153] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0154] In the embodiments provided in this disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0155] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0156] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A machine vision-based method for precise positioning and control optimization of chip pins, characterized in that, Includes the following steps: Initial optical coaxial uniformity is obtained based on multi-angle brightness distribution consistency analysis; The initial optical coaxial uniformity is sampled by neighborhood expansion to generate several optical coaxial uniformities to be screened. The optical coaxial uniformity is mapped to a set of control targets based on the illumination-imaging coupling response curve. The set of control targets includes the coaxial illumination numerical aperture adjustment amount and the optical axis eccentricity compensation displacement amount. The mapped optical coaxial uniformity of the target to be screened is applied to perform pin positioning; Acquire pin positioning data and generate uniformity scores based on machine learning training; Based on the optical coaxial uniformity to be screened and the corresponding uniformity score, a performance curve is constructed, and the optimal optical coaxial uniformity is selected for pin positioning. The initial optical coaxial uniformity is obtained based on multi-angle brightness distribution consistency analysis, specifically as follows: Acquire the brightness distribution image of the light source under a preset angle set, wherein the angle set consists of multiple observation angles acquired with the optical axis as the center at a fixed angle step; For each observation angle, calculate the radial gradient sequence at the corresponding angle in the brightness distribution image, and generate a set of radial gradient sequences indexed by angle; Calculate the relative amplitude of the radial gradient sequence corresponding to each angle in the radial gradient sequence set, and generate a radial gradient relative amplitude sequence indexed by angle; The maximum relative amplitude is selected from the radial gradient relative amplitude sequence to generate the radial gradient dispersion. In the brightness distribution image corresponding to each observation angle, select the circumferential brightness value at a preset radius to generate a circumferential brightness sequence; Calculate the relative circumferential amplitude of the circumferential brightness sequence to generate the circumferential brightness variability. The initial optical coaxial uniformity is obtained based on the analysis of radial gradient dispersion and circumferential brightness fluctuation.

2. The machine vision-based chip pin precise positioning control optimization method according to claim 1, characterized in that, The initial optical coaxial uniformity is obtained based on the analysis of radial gradient dispersion and circumferential brightness fluctuation, specifically as follows: The radial gradient dispersion and circumferential brightness fluctuation are converted into brightness deviation pairs, which consist of the radial deviation corresponding to the radial gradient dispersion and the circumferential deviation corresponding to the circumferential brightness fluctuation. A brightness deviation feature sequence is generated based on the brightness deviation amount, and the brightness deviation feature sequence is formed by arranging the radial deviation amount and the circumferential deviation amount in a preset order; The component with the larger deviation in the brightness deviation feature sequence is selected as the main deviation, and the main deviation is obtained. Substitute the principal deviation into the preset deviation grading interval to obtain the grading position of the principal deviation in the deviation grading interval. The corresponding uniformity percentage is determined based on the grade position of the main deviation, and used as the initial optical coaxial uniformity.

3. The machine vision-based chip pin precise positioning control optimization method according to claim 2, characterized in that, The process of generating several optical coaxial uniformities to be screened by neighborhood expansion sampling of the initial optical coaxial uniformity is as follows: The neighborhood expansion step size set is determined based on the initial optical coaxial uniformity. The neighborhood expansion step size set includes multiple expansion step sizes generated in an arithmetic manner based on a preset minimum step size. The initial optical coaxial uniformity is added to and subtracted from the positive and negative expansion step sizes in the neighborhood expansion step size set to generate a sequence of candidate uniformity values ​​arranged by step size index. Apply a preset uniformity effective interval constraint to the candidate uniformity value sequence, remove candidate uniformity values ​​that exceed the uniformity effective interval, and generate an effective candidate uniformity sequence. The valid candidate uniformity sequence is checked for duplicate values ​​and sorted in ascending order to generate an ordered candidate uniformity sequence. Each uniformity value in the ordered candidate uniformity sequence is combined with the initial optical coaxial uniformity to form a neighborhood extended uniformity set. Each uniformity value in the neighborhood extended uniformity set is an optical coaxial uniformity to be screened.

4. The machine vision-based chip pin precise positioning control optimization method according to claim 3, characterized in that, The mapping of optical coaxial uniformity to the control target group based on the illumination-imaging coupling response curve is specifically as follows: Several numerical aperture adjustment values ​​are obtained from the illumination end as numerical aperture sampling points; Several optical axis eccentricity compensation displacements are obtained from the imaging end as eccentricity displacement sampling points; A set of illumination-imaging sampling points arranged in two-dimensional indexes is generated based on numerical aperture sampling points and eccentric displacement sampling points; Adjust the illumination numerical aperture and optical axis eccentricity compensation displacement at each sampling point in the illumination-imaging sampling point set, and acquire the corresponding imaging brightness distribution image to generate a brightness distribution image set indexed by sampling point; For each brightness distribution image in the brightness distribution image set, obtain the radial brightness response value and the circumferential brightness response value, and combine the two to form a coaxial response value, generating a set of coaxial response values ​​indexed by sampling point; Two-dimensional interpolation is performed based on the set of illumination-imaging sampling points and the corresponding set of coaxial response values ​​to generate an illumination-imaging coupling response curve with the numerical aperture adjustment of illumination and the optical axis eccentricity compensation displacement as input variables and the coaxial response value as the output variable. Each optical coaxial uniformity to be screened in the neighborhood extended uniformity set is associated with the coupling response curve to determine the coaxial response value corresponding to the corresponding optical coaxial uniformity. The illumination numerical aperture adjustment amount and optical axis eccentricity compensation displacement amount that make the coaxial response value consistent with the corresponding optical coaxial uniformity are found in the coupling response curve to generate a set of control pairs indexed by uniformity. The control targets are combined to form a set of control targets, which includes the numerical aperture adjustment of illumination and the eccentricity compensation displacement of the optical axis.

5. The machine vision-based chip pin precise positioning control optimization method according to claim 4, characterized in that, The step of applying the mapped optical coaxial uniformity of the object to be screened to perform pin positioning is specifically as follows: The numerical aperture adjustment amount and optical axis eccentricity compensation displacement corresponding to the optical coaxial uniformity to be screened are obtained respectively, and the two are applied to form a positioning light field for the corresponding uniformity. Under the positioning light field, the full field of view of the array of pins is exposed point by point, and the corresponding pin imaging frame sequence is acquired to generate a pin imaging sequence for corresponding optical coaxial uniformity. Background suppression and brightness normalization are performed on the pin imaging sequence to obtain a pin contour image sequence with uniform brightness; Extract the pin boundaries in each pin outline image, perform sub-pixel center fitting based on the boundary shape to obtain the center coordinates of each pin, and form a set of pin center coordinates for corresponding uniformity. Based on the set of pin center coordinates and the standard arrangement coordinates of the array pins, the center offset of each pin is calculated point by point to generate a set of pin offsets; Identify the maximum offset, average offset, and offset distribution in the set of pin offsets to complete the positioning.

6. The machine vision-based chip pin precise positioning control optimization method according to claim 5, characterized in that, The process of acquiring pin positioning data and generating a uniformity score based on machine learning training specifically involves: Based on the set of pin offsets, the pin center offset features are extracted to obtain the maximum offset, average offset, and offset variance used to characterize the degree of offset diffusion under low uniformity conditions, forming a center offset feature set. The contour gradient magnitude of each pin is calculated based on the pin contour image sequence, and the minimum, maximum and gradient magnitude change rate of the gradient magnitude within the entire field of view are statistically analyzed to obtain a contour gradient feature set used to characterize the discontinuity of contour gray-level gradient caused by low uniformity. In each pin profile image, the brightness value of the solder ball reflection peak region is measured, and the peak area, peak overexposure ratio and brightness difference between adjacent peaks are calculated to form a reflection peak feature group to characterize the degree of overexposure and spot diffusion in the reflection area caused by high uniformity. The contrast between solder balls is calculated based on the brightness difference between adjacent solder balls in the full field of view pin outline image. The average contrast and contrast decay rate are statistically analyzed in each image to generate a contrast feature group to characterize the weakening of solder ball boundaries caused by high uniformity. The center offset feature group, contour gradient feature group, reflection peak feature group, and contrast feature group are combined in a fixed order to form a positioning feature vector for this uniformity. Collect all the positioning feature vectors corresponding to the optical coaxial uniformity to be screened, construct a positioning feature training set, and label each feature vector with positioning quality labels based on pin offset stability and contour fidelity to generate a positioning quality label set. A regression model is trained based on a training set of positioning features and a set of positioning quality labels. The weighting coefficients of each positioning feature in the uniformity score are determined through training, resulting in a machine learning model for calculating the optical coaxial uniformity score. The machine learning model is a weighted form of all features, and the training object is the weight value of each feature. The trained machine learning model outputs the uniformity score.

7. The machine vision-based chip pin precise positioning control optimization method according to claim 6, characterized in that, The process involves constructing a performance curve based on the optical coaxial uniformity to be screened and the corresponding uniformity score, analyzing and selecting the optimal optical coaxial uniformity for pin positioning, specifically as follows: All the optical coaxial uniformity to be screened are sorted by numerical value to form a uniformity sequence arranged in ascending order of uniformity. Based on the order of the uniformity sequence, the corresponding uniformity scores are arranged in the same order to form a score sequence; A performance curve is constructed to characterize the relationship between uniformity changes and score changes, with the uniformity sequence as the horizontal axis and the score sequence as the vertical axis. Calculate the difference between adjacent scores in the performance curve, identify the position of the maximum gain of the score difference, and determine the uniformity corresponding to the position as the candidate optimal uniformity; Read the corresponding uniformity scores at the previous and next uniformity near the candidate optimal uniformity to form a candidate interval score group; Compare the scores in the candidate interval score group and select the uniformity with the highest score as the final optimal optical coaxial uniformity. The final optimal optical coaxial uniformity is input into the illumination-imaging coupling response curve to obtain the corresponding illumination numerical aperture adjustment amount and optical axis eccentricity compensation displacement amount, thereby generating the optimal positioning control target for chip pin positioning.

8. A system using the machine vision-based chip pin precise positioning control optimization method as described in any one of claims 1-7, characterized in that, It includes an optical coaxial uniformity initialization module, a neighborhood expansion sampling module, an optical coaxial uniformity control mapping module, a pin positioning module, a uniformity scoring module, and an optical coaxial uniformity screening module. The optical coaxial uniformity initialization module is used to obtain the initial optical coaxial uniformity based on the consistency analysis of multi-angle brightness distribution. The neighborhood expansion sampling module is used to perform neighborhood expansion sampling on the initial optical coaxial uniformity to generate several optical coaxial uniformities to be screened. An optical coaxial uniformity control mapping module is used to map optical coaxial uniformity to a control target group based on the illumination-imaging coupling response curve. The control target group includes the coaxial illumination numerical aperture adjustment amount and the optical axis eccentricity compensation displacement amount. The pin positioning module is used to perform pin positioning by applying the mapped optical coaxial uniformity of the object to be screened; The uniformity scoring module is used to acquire pin positioning data and generate uniformity scores based on machine learning training. The optical coaxial uniformity screening module is used to construct performance curves based on the optical coaxial uniformity to be screened and the corresponding uniformity scores, and to analyze and screen the best optical coaxial uniformity for pin positioning.

Images