Optical illumination and visual recognition collaborative precision improvement method and system

Abstract

The application relates to a kind of optical illumination and visual recognition collaborative precision promotion method and system, the method comprises: the full-link error transmission model of glass-based mark point visual identification is constructed;Based on the parameter optimization and hardware construction of integrated directional lighting system of error transmission model;Based on the imaging quality pre-regulation and collaborative pretreatment of lighting parameter optimization;Subpixel edge detection algorithm is used to realize subpixel level edge positioning, and residual distortion and warping deviation are compensated;Establish positioning accuracy evaluation system, when positioning error or edge gradient triggers threshold, automatically adjust lighting parameters, and repeat the pretreatment and positioning steps, form closed loop optimization;Complete the visual system calibration and stability adaptation under mass production condition.The application realizes submicron level high-precision positioning, solves the imaging pain point of glass substrate, and has fast adaptation, long-term stability and strong versatility, provides a high-precision visual solution for various glass substrates precision mounting and advanced packaging.

Description

Technical Field

[0001] This invention relates to the fields of machine vision and precision manufacturing technology, and more specifically, to a method and system for improving the accuracy of optical illumination and visual recognition in synergy. Background Technology

[0002] With the rapid development of industries such as naked-eye 3D display and ultra-high-speed co-packaged optics (CPO) optical engine packaging, high-precision mounting of glass substrates has become a core manufacturing link in the fields of new display and optical communication semiconductors.

[0003] Current research and industrialization solutions for visual recognition technology of glass-based Mark points mainly fall into two categories: lighting hardware structure optimization and visual recognition algorithm optimization. Lighting hardware structure optimization focuses on improvements to hardware structures such as coaxial polarized illumination and ring-coaxial composite illumination. However, these solutions primarily address localized improvements to the hardware structure, targeting single-layer glass and single-sided Mark point imaging designs. They cannot adapt to cross-layer scenarios in flip-chip mounting where imaging of Mark points on the back side requires penetrating the glass substrate. Furthermore, they lack a quantitative relationship between lighting parameters and imaging and positioning errors, making parameter optimization highly dependent on trial and error based on human experience. Visual recognition algorithm optimization focuses on improvements to algorithms such as edge detection, sub-pixel fitting, and template matching. However, these solutions can only correct the image after imaging and cannot fundamentally suppress optical errors caused by the high transmittance and strong reflection of the glass substrate. When image contrast is insufficient or edges are severely blurred, the accuracy improvement effect of algorithm optimization is extremely limited, and may even introduce additional fitting errors, failing to meet sub-micron level stable positioning requirements.

[0004] Therefore, there is a need to design a method and system for improving the accuracy of optical illumination and visual recognition in collaboration with the lighting parameters and recognition algorithm to meet the cross-layer imaging requirements of Mark points on the back of glass substrates in flip-chip mounting scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for improving the accuracy of optical illumination and visual recognition in coordination, so as to solve the problem that the existing technology cannot achieve coordinated matching of illumination parameters and recognition algorithms, and adapt to the cross-layer imaging requirements of Mark points on the back of glass substrates in flip-chip mounting scenarios.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for improving the accuracy of optical illumination and visual recognition in synergy, comprising: Construct a full-link error propagation model for visual recognition of glass-based Mark points; Parameter optimization and hardware construction of an integrated directional lighting system based on an error propagation model; Imaging quality pre-control and collaborative preprocessing based on illumination parameter optimization; A subpixel edge detection algorithm is used to achieve subpixel-level edge localization, and the error propagation model is used to compensate for residual distortion and warping deviation. Establish a positioning accuracy evaluation system. When the positioning error or edge gradient triggers the threshold, automatically adjust the lighting parameters and repeat the preprocessing and positioning steps to form a closed-loop optimization. Complete the calibration and stability adaptation of the vision system under mass production conditions.

[0007] In one embodiment, the construction of the end-to-end error propagation model for glass-based Mark point visual recognition includes: Identify the error sources throughout the entire visual recognition process; Construct the transfer function from optical error to imaging error; Construct a transfer function from imaging error to pixel extraction error; Construct a transfer function from pixel extraction error to spatial positioning error; The influence weight of core lighting parameters on spatial positioning error is quantified.

[0008] In one embodiment, the parameter optimization and hardware construction of the integrated directional lighting system based on the error propagation model includes: Based on the error propagation model, the selection and matching of visual imaging hardware is completed; Based on the error propagation model, the core parameters of the integrated directional lighting system were quantitatively optimized. Construct an integrated directional lighting system.

[0009] In one embodiment, the core parameter quantification optimization includes: spectral parameter optimization, incident angle optimization, lamp bead spacing optimization, and brightness level optimization.

[0010] In one embodiment, the imaging quality pre-control and collaborative preprocessing based on illumination parameter optimization includes: The original image of the Mark point was acquired; The original images of the acquired Mark points are preprocessed.

[0011] In one embodiment, the acquired raw image of the Mark points is preprocessed, including: Based on the image of the red light channel, perform single-channel grayscale extraction; Median filtering algorithm is used to remove salt-and-pepper noise from the image, and the filter kernel size is matched with the illumination uniformity; An adaptive histogram equalization algorithm with limited contrast is used to enhance the grayscale difference between the Mark point edge and the background.

[0012] In one embodiment, a subpixel edge detection algorithm is used to achieve subpixel-level edge localization, and the error propagation model is used to compensate for residual distortion and warping deviation, including: A subpixel-level edge localization is achieved using a subpixel edge detection algorithm based on the Canny operator; The center coordinates of the Mark point are fitted using the least squares method. At the same time, the error propagation model is combined to compensate for the imaging deviation caused by lens residual distortion and glass warping in real time, so as to obtain the accurate pixel coordinates of the Mark point. Based on camera calibration parameters, the pixel coordinates of the Mark point are converted into physical space coordinates to achieve high-precision positioning of the glass-based Mark point.

[0013] In one embodiment, a positioning accuracy evaluation system is established. When a positioning error or edge gradient triggers a threshold, the lighting parameters are automatically adjusted, and the preprocessing and positioning steps are repeated to form a closed-loop optimization, including: Establish a positioning accuracy evaluation system; Set closed-loop optimization trigger conditions; After adjusting the lighting parameters, repeat the steps of pre-controlling and co-processing of imaging quality based on lighting parameter optimization and achieving sub-pixel level edge localization using the sub-pixel edge detection algorithm, and use the error propagation model to compensate for residual distortion and warping deviation until the localization accuracy meets the set requirements, thereby achieving bidirectional co-optimization of lighting and recognition. Perform this closed-loop optimization once every 90-110 products to automatically correct accuracy drift caused by changes in ambient light and light source attenuation.

[0014] In one embodiment, the calibration and stability adaptation of the vision system under mass production conditions includes: A glass calibration plate with a precision of 0.5μm was used to complete the one-time calibration of the vision system; A lighting parameter formula library was established based on glass substrates of different thicknesses and materials; Establish an equipment status monitoring mechanism.

[0015] The present invention also provides a system for improving the accuracy of optical illumination and visual recognition in conjunction, comprising: The model building module is used to build a full-link error propagation model for glass-based Mark point visual recognition; The model optimization module is used for parameter optimization and hardware construction of an integrated directional lighting system based on an error propagation model. The preprocessing module is used for pre-processing and collaborative pre-imaging quality control based on illumination parameter optimization. A high-precision positioning module is used to achieve sub-pixel-level edge positioning using a sub-pixel edge detection algorithm, and to compensate for residual distortion and warping deviation using the error propagation model. The closed-loop optimization module is used to establish a positioning accuracy evaluation system. When the positioning error or edge gradient triggers the threshold, the lighting parameters are automatically adjusted, and the preprocessing and positioning steps are repeated to form a closed-loop optimization. The system calibration module is used to complete the calibration and stability adaptation of the vision system under mass production conditions.

[0016] As can be seen from the above technical solutions, the present invention has at least the following advantages and positive effects compared with the prior art: The optical illumination and visual recognition collaborative accuracy improvement method and system of the present invention achieves sub-micron level high-precision positioning through deep collaboration between optics and algorithms, solves the pain point of glass substrate imaging, and has rapid adaptation, long-term stability and strong versatility, providing a high-precision visual solution for precision mounting and advanced packaging of various glass substrates. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0018] Figure 1 This is a flowchart illustrating the method for improving the accuracy of optical illumination and visual recognition in conjunction with the present invention. Figure 2 This is a flowchart illustrating the process of constructing the end-to-end error propagation model for visual recognition of glass-based Mark points in step S1 of this invention. Figure 3 This is a flowchart illustrating the parameter optimization and hardware construction of the integrated directional lighting system based on the error propagation model in step S2 of the present invention. Figure 4 This is a schematic diagram of the imaging quality pre-control and collaborative preprocessing based on illumination parameter optimization in step S3 of the present invention. Figure 5 This is a schematic diagram of the process of preprocessing the original image of the acquired Mark points in step S302 of the present invention. Figure 6 This is a schematic diagram of the process in step S4 of the present invention, which uses a subpixel edge detection algorithm to achieve subpixel-level edge localization and uses the error propagation model to compensate for residual distortion and warping deviation. Figure 7The establishment of a positioning accuracy evaluation system in step S5 of this invention involves automatically adjusting lighting parameters and repeating preprocessing and positioning steps when positioning error or edge gradient trigger thresholds are reached, forming a flowchart of closed-loop optimization. Figure 8 This is a flowchart illustrating the process of visual system calibration and stability adaptation under mass production conditions in step S6 of the present invention. Figure 9 This is a schematic diagram of the structure of an optical illumination and visual recognition collaborative accuracy improvement system according to the present invention.

[0019] The annotations in the attached figures are explained as follows: Among them, 210 is the model building module; 220 is the model optimization module; 230 is the preprocessing module; 240 is the high-precision positioning module; 250 is the closed-loop optimization module; and 260 is the system calibration module. Detailed Implementation

[0020] To more clearly explain the purpose, technical solutions, and advantages of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein. On the contrary, these embodiments are provided so that the present invention will be more comprehensive and complete, and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0021] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of the invention. However, those skilled in the art will recognize that the technical solutions of the invention can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the invention.

[0022] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0023] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0024] The present invention will now be described in detail with reference to specific embodiments: Example 1

[0025] This invention provides a method for improving the accuracy of optical illumination and visual recognition in conjunction with each other, with reference to... Figure 1 , Figure 1 This is a flowchart illustrating the method for improving the accuracy of optical illumination and visual recognition in conjunction with the present invention.

[0026] Depend on Figure 1 It is known that a method for improving the accuracy of optical illumination and visual recognition through collaboration includes the following steps: S1. Construct a full-link error propagation model for visual recognition of glass-based Mark points.

[0027] The core of step S1 is to establish a quantitative mapping relationship between lighting parameters and final positioning accuracy, thus breaking free from the limitations of traditional trial-and-error based on experience.

[0028] refer to Figure 2 , Figure 2 This is a schematic diagram of the process of constructing the end-to-end error propagation model for visual recognition of glass-based Mark points in step S1 of the present invention.

[0029] The construction of a full-link error propagation model for glass-based Mark point visual recognition includes the following steps: S101. Identify the error sources throughout the entire visual recognition process.

[0030] Errors affecting positioning accuracy can be categorized into four main types: defocusing errors caused by insufficient depth of field, geometric distortion errors caused by optical links, grayscale errors caused by uneven illumination and stray light, and extraction errors caused by image acquisition and algorithms.

[0031] S102. Construct the transfer function from optical error to imaging error.

[0032] Let the total optical error of the system be E. o Due to depth error E d Lens distortion error E l Cylinder refraction error E r Lighting error E i Composition, namely: E o =E d +E l +Er +E i Optical errors are projected onto the image plane and converted into imaging errors E. m E m The formula for the pixel deviation between the actual and ideal edge positions of the Mark point is: , Where (u0,v0) are the ideal image edge pixel coordinates of the Mark point, and (u,v) are the actual edge pixel coordinates under the influence of optical errors.

[0033] S103. Construct the transfer function from imaging error to pixel extraction error.

[0034] Let the ideal pixel coordinates of the center of Mark point be (u c ,v c ), pixel extraction error E p for: , in, Ep represents the actual center pixel coordinates extracted by the algorithm. Ep is negatively correlated with the imaging edge gradient and image contrast, and positively correlated with the proportion of stray light.

[0035] S104. Construct the transfer function from pixel extraction error to spatial positioning error.

[0036] Let the scaling factor from pixel coordinates to physical coordinates be k. x k y Spatial positioning error E s for: , in, , .

[0037] S105, Quantify the influence weight of core lighting parameters on spatial positioning error.

[0038] The core lighting parameters include: incident angle of the light source, spacing between LEDs, brightness level of the light source, and wavelength of the light source spectrum. The optimization priority and target range of each parameter are clearly defined.

[0039] It should be noted that the input data for the error propagation model includes static physical parameters, dynamic environmental parameters, and real-time image features. Static physical parameters include glass thickness, refractive index, lens distortion coefficient, and pixel equivalent. Dynamic environmental parameters include ambient light intensity and current operating temperature. Real-time image features include the contrast of the current frame image, edge gradient, and stray light ratio.

[0040] It should be noted that the output data of the error propagation model includes intermediate layer output, final layer output, and control command output. The intermediate layer output includes the predicted pixel extraction error and the predicted optical distortion field. The final layer output includes the predicted spatial positioning error. The control command output includes the optimal illumination parameter combination, target brightness, and target spectrum.

[0041] S2. Parameter optimization and hardware construction of an integrated directional lighting system based on an error propagation model.

[0042] The core of step S2 is to complete the quantitative optimization of the visual hardware and lighting system based on the error propagation model of step S1, realize the pre-control of imaging quality, and suppress optical errors from the source.

[0043] refer to Figure 3 , Figure 3 This is a flowchart illustrating the parameter optimization and hardware construction of the integrated directional lighting system based on the error propagation model in step S2 of this invention.

[0044] The parameter optimization and hardware construction of an integrated directional lighting system based on an error propagation model includes the following steps: S201. Based on the error propagation model, complete the selection and matching of visual imaging hardware.

[0045] Visual imaging hardware selection includes, but is not limited to: camera selection, lens selection, and depth-of-field optimization.

[0046] Camera selection: A 24-megapixel global shutter CMOS industrial camera is selected, with a single pixel physical size of 3.45μm×3.45μm, a frame rate of ≥35fps, and an imaging linearity of ≤0.1%, ensuring no motion blur under high-speed motion conditions, while meeting the resolution requirements for sub-micron level positioning.

[0047] Lens selection: A dual-sided telecentric lens with an optical magnification of 0.38x, optical distortion ≤0.015%, working distance of 120mm±10mm, aperture range of F4-F16, fully covering the camera sensor surface, with no vignetting.

[0048] Depth of field optimization: Based on the object-side depth of field theory formula, the optimal aperture of the lens is determined to be F8. Under the condition of grating, the effective depth of field is ≥0.4mm, which can cover the object distance fluctuation requirements of ±0.2mm in the production line and avoid edge blur caused by defocus.

[0049] S202. Based on the error propagation model, complete the quantitative optimization of the core parameters of the integrated directional lighting system.

[0050] The core parameter quantification optimization includes, but is not limited to, spectral parameter optimization, incident angle optimization, LED spacing optimization, and brightness level optimization.

[0051] Spectral parameter optimization: Based on the glass substrate and the chromium mark point, the imaging effects of 450nm blue light, 520nm green light, 620nm red light, and white light were compared, and 620nm red light was determined to be the optimal spectrum. This band has moderate transmittance to optical glass, minimal refractive influence from the parallax barrier grating structure, and the greatest difference in reflectivity between the mark point and the glass substrate, resulting in an imaging grayscale contrast ratio ≥45:1. In glass substrates with parallax barrier cylindrical lens structures, long-wavelength red light has lower sensitivity to diffraction and refraction at the microstructure interface, effectively reducing multiple stray light interference caused by the cylindrical lens structure.

[0052] Incident Angle Optimization: Through single-factor control experiments, a composite incident angle of 0° coaxial illumination and 15° annular oblique illumination was determined. This scheme ensures uniform illumination of the overall outline of the Mark point while highlighting the edge features of the etched Mark point through oblique illumination. Measured stray light ratio is ≤1.8%, resulting in Mark point imaging without halo, shadow, or sharp edges. 0° coaxial illumination ensures the integrity of the overall outline, while 15° annular oblique illumination highlights edge details. The power ratio of these two elements achieves both overall uniformity and localized sharpening in the imaging effect. Actual measurements show that this scheme improves the Mark point edge gradient by approximately 2.3 times and controls the stray light ratio to within 1.8%, fundamentally solving the core pain points of blurred edges and insufficient contrast in Mark point imaging on the back of the glass substrate.

[0053] LED spacing optimization: The optimal spacing of the LEDs was determined to be 5mm. 0402 packaged LEDs were used to form a square closed-loop light-emitting array to ensure continuous light emission without dark areas, full-field light uniformity ≥96.7%, and edge brightness attenuation ≤3%.

[0054] Brightness level optimization: 50%-70% is determined as the optimal brightness range. At this time, the grayscale value of the camera image is in the optimal dynamic range of 120-200, the peak signal-to-noise ratio is ≥43dB, there are no overexposure or underexposure issues, and the edge feature extraction is guaranteed to be stable.

[0055] Based on the aforementioned visual imaging hardware, the hardware parameters determined after quantitative optimization in this embodiment are as follows: coaxial light incident angle 0°, ring light incident angle 15°, spectral center wavelength 620nm, LED spacing 5mm, brightness level 50%-70%, and light-emitting surface distance 50mm. Under the conditions of 0.7mm thick glass substrate and chromium mark point imaging on the back side, this solution achieves: image contrast ratio ≥45:1, stray light ratio ≤1.8%, average edge gradient improvement of 2.3 times, and camera grayscale value within the optimal dynamic range of 120-200. This provides high-quality, low-noise image input for subsequent sub-pixel edge detection and closed-loop optimization, with a final spatial positioning accuracy ≤0.2μm.

[0056] Through the aforementioned hardware solution that combines composite angle and spectrum, this invention fundamentally solves the industry pain points of blurred Mark point edges, insufficient contrast, and halo interference when imaging through glass in the prior art, and achieves deep collaboration between lighting parameters and visual recognition algorithms.

[0057] S203. Construct an integrated directional lighting system.

[0058] The lighting system and imaging unit are coaxially mounted, using an optical-grade PMMA acrylic integral injection molded bracket with no splicing gaps; an integrated cylindrical light-concentrating strip is set on the inner side of the bracket, which can shape the LED Lambertian divergent light into parallel directional light with a divergence angle of ≤3°; the distance between the light-emitting surface of the light source and the glass substrate is set at 50mm, with no mechanical interference from the 120mm working distance of the lens, and the optical path is completely coaxial, ensuring a perfect match between the incident angle of the light and the receiving angle of the lens.

[0059] S3. Imaging quality pre-control and collaborative preprocessing based on illumination parameter optimization.

[0060] The core of step S3 is to perform preprocessing on the image based on the illumination optimization results of step S2, which is coordinated with the illumination parameters, to further amplify the imaging advantages and lay the foundation for subsequent high-precision recognition.

[0061] refer to Figure 4 , Figure 4 This is a schematic diagram of the process of pre-processing and collaborative preprocessing of imaging quality based on illumination parameter optimization in step S3 of the present invention.

[0062] Imaging quality pre-control and collaborative preprocessing based on illumination parameter optimization specifically include the following steps: S301. Acquire the original image of the Mark point.

[0063] Based on the glass cross-layer imaging requirements in the flip-chip mounting scenario, the composite angle directional illumination optimized in step S2 is used to suppress stray light from the double specular reflection on the upper and lower surfaces of the glass and the refraction of the cylindrical lens structure, thereby eliminating imaging halo and edge blurring issues and acquiring the original image of the Mark point.

[0064] S302. Preprocess the original images of the acquired Mark points.

[0065] refer to Figure 5 , Figure 5 This is a schematic diagram of the process of preprocessing the original image of the acquired Mark points in step S302 of the present invention.

[0066] The preprocessing of the acquired Mark point raw images includes the following steps: S3021. Based on the image of the red light channel, perform single-channel grayscale extraction to avoid noise interference caused by multi-channel fusion.

[0067] S3022. Median filtering algorithm is used to remove salt-and-pepper noise from the image, and the filter kernel size is matched with the illumination uniformity. When the illumination uniformity is ≥95%, a 3×3 filter kernel is selected; when the illumination uniformity is <95%, a 5×5 filter kernel is selected.

[0068] S3023. A contrast-limited adaptive histogram equalization algorithm is adopted to enhance the grayscale difference between the Mark point edge and the background. The clipLimit parameter of the algorithm is adjusted in conjunction with the image contrast: when the image contrast is ≥40:1, clipLimit is set to 2.0; when the contrast is <40:1, clipLimit is set to 3.0.

[0069] S4. A subpixel-level edge localization is achieved using a subpixel edge detection algorithm, and the residual distortion and warping deviation are compensated using the error propagation model.

[0070] The core of step S4 is to establish a linkage mechanism between the recognition algorithm and the lighting parameters, and to achieve sub-micron level precise positioning of the Mark points based on the high-quality image optimized in advance.

[0071] refer to Figure 6 , Figure 6 This is a schematic diagram of the process in step S4 of the present invention, which uses a subpixel edge detection algorithm to achieve subpixel-level edge localization and uses the error propagation model to compensate for residual distortion and warping deviation.

[0072] A subpixel edge detection algorithm is used to achieve subpixel-level edge localization, and the error propagation model is used to compensate for residual distortion and warping deviation. Specifically, the steps include: S401. A sub-pixel-level edge localization is achieved using a sub-pixel edge detection algorithm based on the Canny operator.

[0073] High threshold T h With low threshold T l Based on the image peak signal-to-noise ratio (PSNR): when PSNR ≥ 40dB, T h =0.7 × maximum grayscale value, T l =0.3 × maximum grayscale value; when PSNR < 40dB, T h =0.6 × maximum grayscale value, T l =0.2 × maximum gray value.

[0074] For the detected pixel-level edges, a quadratic polynomial fitting method is used to achieve sub-pixel-level edge localization, with an edge localization resolution of 0.01 pixels.

[0075] S402. The center coordinates of the Mark point are fitted using the least squares method. At the same time, the error propagation model in step S1 is used to compensate for the imaging deviation caused by lens residual distortion and glass warping in real time, so as to obtain the accurate pixel coordinates of the Mark point.

[0076] S403. Based on camera calibration parameters, the pixel coordinates of the Mark point are converted into physical space coordinates to complete the high-precision positioning of the glass-based Mark point.

[0077] S5. Establish a positioning accuracy evaluation system. When the positioning error or edge gradient triggers the threshold, automatically adjust the lighting parameters and repeat the preprocessing and positioning steps to form a closed-loop optimization.

[0078] The core of step S5 is to establish a closed-loop mechanism for optimizing lighting parameters in reverse based on the recognition results, thereby achieving bidirectional collaboration between lighting and recognition and solving the accuracy drift problem caused by environmental changes and light source attenuation during mass production.

[0079] refer to Figure 7 , Figure 7 The diagram illustrates the process of establishing a positioning accuracy evaluation system in step S5 of this invention. When the positioning error or edge gradient triggers the threshold, the lighting parameters are automatically adjusted, and the preprocessing and positioning steps are repeated to form a closed-loop optimization process.

[0080] Establish a positioning accuracy evaluation system. When positioning error or edge gradient triggers a threshold, automatically adjust the lighting parameters and repeat the preprocessing and positioning steps to form a closed-loop optimization. Specifically, this includes the following steps: S501. Establish a positioning accuracy evaluation system.

[0081] The high-precision positioning results obtained in step S4 are evaluated for accuracy. The core evaluation indicators include: standard deviation of positioning error, mean edge gradient, image contrast, and stray light ratio.

[0082] S502, Set closed-loop optimization trigger conditions.

[0083] When the standard deviation of the positioning error is greater than 0.2μm, or the mean edge gradient is less than the set threshold, the adaptive adjustment of the lighting parameters is triggered. The specific adjustment rules are as follows: if the stray light ratio is greater than 3%, the incident angle of the ring light is finely adjusted in a step of 1° until the stray light ratio is less than 2%; if the image contrast is less than 30:1, the brightness level of the light source is adjusted in a step of 5% until the contrast is greater than or equal to 40:1; if the edges are blurred or the gradient is insufficient, the power ratio of the coaxial light and the ring light is adjusted to improve the sharpness of the edge features.

[0084] It should be noted that this is based on the signal-to-noise ratio (SNR) theory of visual algorithms. Experiments show that when the grayscale contrast between the Mark point edge and the background is less than 30:1, the interference of image noise on the sub-pixel edge extraction algorithm increases exponentially, causing the standard deviation of the positioning error to rapidly exceed the 0.2μm accuracy threshold. Therefore, setting the image contrast to <30:1 ensures that the algorithm always operates within the linear region with a sufficient SNR.

[0085] It should be noted that this analysis is based on edge gradient analysis. Stray light fills in the dark details of the Mark points, causing the edge gradient to decrease. When the proportion of stray light exceeds 3%, the decrease in edge gradient leads to a significant increase in the fitting residual of the sub-pixel interpolation algorithm, introducing systematic bias.

[0086] It should be noted that, based on optical sensitivity analysis, even small changes in the incident angle have a significant impact on the optical path in cylindrical glass imaging. Experimental data show that when the angle change exceeds 2°, the position of the reflected light spot may jump, leading to abrupt changes in image features; while adjustments of less than 0.5° are not effective in eliminating stray light and are too time-consuming. A 1° adjustment step is the optimal solution that balances adjustment accuracy with response speed.

[0087] It should be noted that, based on the linear response range of the camera sensor, LED light sources exhibit a non-linear region at low duty cycles and are prone to saturation at high duty cycles. A 5% step size can effectively avoid the non-linear region while ensuring that the contrast ratio can be increased from the threshold (30:1) to above the target value (40:1) within 3-5 adjustment cycles, meeting the production line cycle time requirements.

[0088] S503. After adjusting the illumination parameters, repeat the steps of pre-controlling and co-processing of imaging quality based on illumination parameter optimization and achieving sub-pixel-level edge positioning using the sub-pixel edge detection algorithm, and use the error propagation model to compensate for residual distortion and warping deviation until the positioning accuracy meets the set requirements, thereby achieving bidirectional co-optimization of illumination and recognition.

[0089] S504. Perform this closed-loop optimization once every 90-110 products to automatically correct accuracy drift caused by changes in ambient light and light source attenuation.

[0090] S6. Complete the calibration and stability adaptation of the vision system under mass production conditions.

[0091] refer to Figure 8 , Figure 8 This is a schematic diagram of the process for calibrating and adapting the vision system to stability under mass production conditions in step S6 of the present invention.

[0092] The calibration and stability adaptation of the vision system under mass production conditions includes the following steps: S601 uses a glass calibration plate with a precision of 0.5μm to complete the one-time calibration of the vision system.

[0093] The calibration includes lens distortion, pixel equivalent, and camera intrinsic parameters. After calibration, the residual distortion across the entire field of view is ≤0.5μm.

[0094] S602. Establish a lighting parameter formula library based on glass substrates of different thicknesses and materials.

[0095] Based on the optimization results of step S2, the optimal lighting parameter combination for the corresponding glass specifications is stored and can be directly called during mass production and model changeover without repeated debugging.

[0096] S603. Establish an equipment status monitoring mechanism.

[0097] It monitors parameters such as light source brightness, camera image quality, and ambient temperature and humidity in real time, and issues early warnings when parameters exceed thresholds to ensure the stability of the mass production process.

[0098] In summary, this application achieves sub-micron level high-precision positioning through deep collaboration between optics and algorithms, solving the pain points of glass substrate imaging. It also features rapid adaptation, long-term stability, and strong versatility, providing a high-precision vision solution for precision mounting and advanced packaging of various glass substrates.

[0099] This invention achieves deep synergy between optics and algorithms, resulting in an order-of-magnitude improvement in recognition accuracy. It utilizes a full-link error propagation model to establish a collaborative optimization channel between illumination and recognition, rather than relying on local optimization of independent modules.

[0100] Perfectly adapted to the needs of flip-chip mounting and cross-layer imaging, this invention solves the pain points of the glass substrate imaging industry. Through composite angle directional illumination, spectral optimization, and parallel light shaping design, it can effectively suppress double-sided reflections of glass and stray light from cylindrical lens structures, improving image contrast to 47:1 and reducing stray light ratio to 1.8%. This fundamentally solves the core pain points of existing technologies, such as blurred Mark point edges, insufficient contrast, and halo interference, when imaging through glass.

[0101] This invention establishes a standardized parameter optimization method, significantly improving mass production adaptability. Based on an error propagation model, it establishes a quantitative mapping relationship between lighting parameters and positioning accuracy. Without the need for manual experience-based adjustments, it can quickly complete parameter optimization and adaptation for glass substrates of different thicknesses and materials. Product changeover and debugging time is reduced from the traditional 30 minutes to less than 10 minutes, greatly improving production line changeover efficiency.

[0102] End-to-end error control ensures long-term stability in mass production. This invention suppresses errors at the source of the optical link, combining front-end illumination optimization with back-end algorithm collaboration. At the same time, through a two-way closed-loop optimization mechanism, it automatically corrects accuracy drift caused by environmental changes and light source attenuation. Example 2

[0103] The error propagation model is structurally represented as an "end-to-end" serial hierarchical architecture, with the logical connection relationship as follows: optical error source - imaging error - pixel extraction error - spatial positioning error.

[0104] At the system module level, the model is generated by the model building module and interacts bidirectionally with the model optimization module, high-precision positioning module, and closed-loop optimization module. It not only serves as a forward guide connecting the hardware physical layer (lighting system) and the software algorithm layer (image processing), but also serves as a reverse basis connecting the accuracy evaluation results and lighting parameter adjustments, thereby achieving full lifecycle management from error generation, transmission, quantification to final elimination.

[0105] The construction and training process of the error propagation model is essentially a systematic process of physical parameter calibration, sensitivity analysis, and establishment of mapping relationships. This process aims to fit the propagation coefficients of errors at each level using experimental data. Specific implementation details are as follows: Step 1: Calibration of basic physical parameters.

[0106] First, the physical properties of the glass substrate and the geometric characteristics of the Mark points are input. Optical experiments are then conducted to determine the lens's distortion coefficient and depth of field at different working distances, establishing the first level: the fundamental physical field of optical error - imaging error.

[0107] It should be noted that the physical properties of the glass substrate include, but are not limited to, thickness and refractive index.

[0108] Step 2: Spatial scan of lighting parameters.

[0109] Under controlled experimental conditions, a full-parameter spatial scan of the integrated directional lighting system was performed. The incident angle of the light source, the spacing between the LEDs, the spectral wavelength, and the brightness level were changed sequentially, and original images of the Mark points were acquired under different combinations.

[0110] Step S3: Image quality feature extraction.

[0111] The acquired images are processed using algorithms to extract key feature metrics, including mean edge gradient, image contrast, signal-to-noise ratio, and stray light ratio. This step is used to construct the second level: the quantization relationship between imaging error and pixel extraction error.

[0112] Step S4: Error sensitivity measurement.

[0113] The aforementioned image features are input into the localization algorithm to calculate the final spatial localization error. Regression analysis is then used to calculate the partial derivatives of each lighting parameter with respect to the spatial localization error.

[0114] Step S5: Generate the optimal parameter mapping table.

[0115] Based on the sensitivity analysis results, parameter ranges with high error sensitivity are eliminated, and an illumination parameter-optimal imaging quality mapping table is generated. This table is the final product of model training. Example 3

[0116] refer to Figure 9 This invention provides a system for improving the accuracy of optical illumination and visual recognition in synergy. Figure 9 This is a schematic diagram of the structure of an optical illumination and visual recognition collaborative accuracy improvement system according to the present invention.

[0117] Depend on Figure 9 An optical illumination and visual recognition collaborative accuracy improvement system is known to include: a model building module 210, a model optimization module 220, a preprocessing module 230, a high-precision positioning module 240, a closed-loop optimization module 250, and a system calibration module 260.

[0118] Among them, the model building module 210 is used to build a full-link error propagation model for visual recognition of glass-based Mark points.

[0119] The model optimization module 220 is used for parameter optimization and hardware construction of an integrated directional lighting system based on an error propagation model.

[0120] The preprocessing module 230 is used for pre-imaging quality control and collaborative preprocessing based on illumination parameter optimization.

[0121] The high-precision positioning module 240 is used to achieve sub-pixel-level edge positioning using a sub-pixel edge detection algorithm, and to compensate for residual distortion and warping deviation using the error propagation model.

[0122] The closed-loop optimization module 250 is used to establish a positioning accuracy evaluation system. When the positioning error or edge gradient triggers the threshold, the lighting parameters are automatically adjusted, and the preprocessing and positioning steps are repeated to form a closed-loop optimization.

[0123] The system calibration module 260 is used to complete the calibration and stability adaptation of the vision system under mass production conditions.

[0124] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims. It should be understood that the invention is not limited to the precise structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for improving the accuracy of optical illumination and visual recognition in conjunction, characterized in that, include: Construct a full-link error propagation model for visual recognition of glass-based Mark points; Parameter optimization and hardware construction of an integrated directional lighting system based on an error propagation model; Imaging quality pre-control and collaborative preprocessing based on illumination parameter optimization; A subpixel edge detection algorithm is used to achieve subpixel-level edge localization, and the error propagation model is used to compensate for residual distortion and warping deviation. Establish a positioning accuracy evaluation system. When the positioning error or edge gradient triggers the threshold, automatically adjust the lighting parameters and repeat the preprocessing and positioning steps to form a closed-loop optimization. Complete the calibration and stability adaptation of the vision system under mass production conditions.

2. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 1, characterized in that, The construction of the end-to-end error propagation model for glass-based Mark point visual recognition includes: Identify the error sources throughout the entire visual recognition process; Construct the transfer function from optical error to imaging error; Construct a transfer function from imaging error to pixel extraction error; Construct a transfer function from pixel extraction error to spatial positioning error; The influence weight of core lighting parameters on spatial positioning error is quantified.

3. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 1, characterized in that, The parameter optimization and hardware construction of the integrated directional lighting system based on the error propagation model include: Based on the error propagation model, the selection and matching of visual imaging hardware is completed; Based on the error propagation model, the core parameters of the integrated directional lighting system were quantitatively optimized. Construct an integrated directional lighting system.

4. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 3, characterized in that, The core parameter quantification optimization includes: spectral parameter optimization, incident angle optimization, lamp bead spacing optimization, and brightness level optimization.

5. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 1, characterized in that, The imaging quality pre-control and collaborative preprocessing based on illumination parameter optimization includes: The original image of the Mark point was acquired; The original images of the acquired Mark points are preprocessed.

6. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 5, characterized in that, The acquired raw images of the Mark points are preprocessed, including: Based on the image of the red light channel, perform single-channel grayscale extraction; Median filtering algorithm is used to remove salt-and-pepper noise from the image, and the filter kernel size is matched with the illumination uniformity; An adaptive histogram equalization algorithm with limited contrast is used to enhance the grayscale difference between the Mark point edge and the background.

7. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 1, characterized in that, A subpixel-level edge localization is achieved using a subpixel edge detection algorithm, and the residual distortion and warping deviation are compensated using the aforementioned error propagation model, including: A subpixel-level edge localization is achieved using a subpixel edge detection algorithm based on the Canny operator; The center coordinates of the Mark point are fitted using the least squares method. At the same time, the error propagation model is combined to compensate for the imaging deviation caused by lens residual distortion and glass warping in real time, so as to obtain the accurate pixel coordinates of the Mark point. Based on the camera calibration parameters, the pixel coordinates of the Mark point are converted into physical space coordinates to achieve high-precision positioning of the glass-based Mark point.

8. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 1, characterized in that, Establish a positioning accuracy evaluation system. When positioning errors or edge gradients trigger thresholds, automatically adjust lighting parameters and repeat preprocessing and positioning steps to form a closed-loop optimization, including: Establish a positioning accuracy evaluation system; Set closed-loop optimization trigger conditions; After adjusting the lighting parameters, repeat the steps of pre-controlling and co-processing of imaging quality based on lighting parameter optimization and achieving sub-pixel level edge localization using the sub-pixel edge detection algorithm, and use the error propagation model to compensate for residual distortion and warping deviation until the localization accuracy meets the set requirements, thereby achieving bidirectional co-optimization of lighting and recognition. Perform this closed-loop optimization once every 90-110 products to automatically correct accuracy drift caused by changes in ambient light and light source attenuation.

9. The method for improving the accuracy of optical illumination and visual recognition in collaboration according to claim 1, characterized in that, Complete the calibration and stability adaptation of the vision system under mass production conditions, including: A glass calibration plate with a precision of 0.5μm was used to complete the one-time calibration of the vision system; A lighting parameter formula library was established based on glass substrates of different thicknesses and materials; Establish an equipment status monitoring mechanism.

10. A system for improving the accuracy of optical illumination and visual recognition in conjunction, characterized in that, include: The model building module is used to build a full-link error propagation model for glass-based Mark point visual recognition; The model optimization module is used for parameter optimization and hardware construction of an integrated directional lighting system based on an error propagation model. The preprocessing module is used for pre-processing and collaborative pre-imaging quality control based on illumination parameter optimization. A high-precision positioning module is used to achieve sub-pixel-level edge positioning using a sub-pixel edge detection algorithm, and to compensate for residual distortion and warping deviation using the error propagation model. The closed-loop optimization module is used to establish a positioning accuracy evaluation system. When the positioning error or edge gradient triggers the threshold, the lighting parameters are automatically adjusted, and the preprocessing and positioning steps are repeated to form a closed-loop optimization. The system calibration module is used to complete the calibration and stability adaptation of the vision system under mass production conditions.

Images