A method and system for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras.
By combining a visual recognition system and adaptive segmentation technology with a flight self-inspection method, real-time splicing error compensation for LDI machines was achieved, solving the problem of high splicing error during the exposure of large-size substrates and improving production stability and efficiency.
Patent Information
- Application Number
- CN202610997665.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-06
- Publication Date
- 2026-11-17
AI Technical Summary
Existing LDI machines suffer from high splicing errors during the exposure of large-size substrates, and these errors cannot be compensated in real time. This leads to unstable splicing quality during production, affecting mass production consistency and production line cycle time.
A visual recognition system is used for initial calibration to establish a global coordinate system. Adaptive block segmentation and stitching exposure methods are combined with flight self-testing technology to detect stitching errors in real time, and dynamic compensation and secondary calibration are performed to achieve closed-loop control.
Without interrupting production, the system can detect and compensate for splicing deviations caused by factors such as mechanical wear and thermal deformation in real time, thereby improving splicing accuracy and graphic integrity, and enhancing the automation level and production consistency of the equipment.
Smart Images

Figure CN123118353A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of LDI (Light Diffusion Identifier) cameras, and more specifically, to a method and system for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras. Background Technology
[0002] LDI (Laser Direct Imaging) technology eliminates the need for physical masks, projecting patterns directly onto the substrate surface using a digital mask. It is a core method for high-precision patterning in the printed circuit board (PCB) industry. For large substrates, a single exposure cannot cover the entire pattern. Therefore, the pattern to be exposed is typically divided into multiple sub-images, which are then exposed step-by-step using a motion adjustment platform and stitched together to form the complete image. The segmentation strategy of the digital mask and the stitching accuracy of the sub-images directly determine the final exposure quality and are key factors for LDI machines to achieve large-area, high-precision processing.
[0003] The LDI machine employs a "hard splicing" method combining single offline calibration with non-overlapping rigid block segmentation, which is simple, easy to implement, and convenient to use. A global coordinate system is established by identifying the substrate target, and the pattern is segmented into blocks according to fixed dimensions. Splicing relies on the absolute positioning accuracy of the motion platform. During continuous production, disturbances such as mechanical motion errors, system thermal deformation, and optical distortion accumulate, resulting in splicing deviations. Existing technologies lack error perception and compensation methods, making it impossible to dynamically adjust splicing quality during production. Once defects such as misalignment or broken lines occur, the machine must be stopped and recalibrated manually, affecting mass production consistency and production line cycle time. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of existing technologies, such as high splicing errors and the inability to perform real-time compensation during production. It provides an adaptive segmentation, splicing exposure, and flight self-inspection method and system for LDI digital masks, which realizes adaptive segmentation and high-precision splicing, and completes flight detection and secondary calibration during the exposure process to achieve closed-loop dynamic compensation and ensure splicing accuracy in continuous production.
[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: This provides a method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras, including the following steps: S1. The visual recognition system accurately identifies multiple targets evenly distributed in an array within the exposure area, thereby solving the coordinate transformation model, completing the initial calibration of the system, establishing a global coordinate system, and inputting the calibration parameters into the LDI machine. S2. Based on the initial calibration results and the position feedback of the motion adjustment platform, the LDI machine adaptively divides the pattern to be exposed into multiple sub-images, and controls the laser to expose the multiple sub-images sequentially at predetermined positions. S3. The LDI machine simultaneously acquires the exposed image composed of the sub-images during the exposure process using a visual recognition system. It extracts the stitching line edges using an image feature extraction algorithm and compares them with a preset exposed image. The linear stitching error between adjacent sub-images is calculated. The linear stitching error parameters include X-axis position deviation, Y-axis position deviation, angle deviation, and line width consistency deviation. A stitching fidelity index is constructed based on these linear stitching error parameters. If the suture fidelity index If the error is within the preset tolerance range, the self-test process is completed. If the linear splicing error is not within the preset tolerance range, proceed to step S4 for correction. S4. Based on the linear stitching error, dynamically compensate the positioning control parameters of the motion adjustment platform and simultaneously correct the lens distortion parameters of the visual recognition system to complete the secondary calibration; then return to step S3 to re-perform image acquisition, linear stitching error calculation, and stitching fidelity index. Calculate and further test the edge roughness after secondary calibration. and repeatability error If the suture fidelity index Edge roughness Repeatability error If all values are within the corresponding preset tolerance range, the self-inspection process is complete; if the suture fidelity index... Edge roughness Repeatability error If any item still exceeds the corresponding preset tolerance range, then the dynamic compensation and synchronization correction in step S4 are repeated, and then the process returns to step S3. If the suture fidelity index... Edge roughness Repeatability error If any item exceeds the corresponding preset tolerance range, the system will shut down and alarm, prompting manual intervention.
[0006] The LDI machine digital mask adaptive blocking, stitching exposure and in-flight self-inspection method of the present invention firstly collects images of a plurality of high-precision targets distributed in an array within the exposure area through a visual recognition system, solves the coordinate transformation model to complete the initial calibration and establishes a global coordinate system; then, based on the initial calibration result and the position feedback of the motion platform, the system performs adaptive blocking on the pattern to be exposed, divides the large-format pattern into a plurality of sub-images and sequentially performs exposure and stitching; during the continuous exposure process, the system reuses the same visual recognition system to synchronously collect images of the stitching overlapping area in an in-flight detection manner without interrupting the motion, extracts the straight edge features of the stitching, calculates the X / Y direction position deviation, angle deviation and line width consistency deviation, and constructs a stitching fidelity index on this basis as a comprehensive evaluation index of stitching quality; if exceeds the tolerance range, dynamic compensation is performed on the positioning parameters of the motion platform and the lens distortion parameters according to the measured error to complete the secondary calibration, and then the process returns to the detection step for recalculation , and the edge roughness is further detected at the same time and repeatability error , only when , , all three indexes meet the corresponding preset tolerance requirements can the self-inspection be determined as qualified. If any one of the indexes does not meet the requirement, the closed-loop process of calibration and detection is repeated, and the machine stops and gives an alarm after three consecutive failures.
[0007] This method realizes hardware resource sharing for calibration and detection by reusing the same vision system, and combines the adaptive blocking strategy and the multi-index comprehensive evaluation system, which can perceive in real time and perform closed-loop compensation for stitching deviations caused by dynamic factors such as mechanical wear, thermal deformation or optical distortion without stopping the machine or interrupting the production rhythm, significantly suppress defects such as pattern misalignment, rough edges and inconsistent line width, and at the same time pass , , the joint judgment of the three indexes ensures that the stitching quality meets the standards in multiple dimensions, greatly improves the pattern integrity, consistency and long-term process stability of batch exposure, reduces the frequency of manual intervention, and improves the comprehensive production capacity and automation level of the equipment.
[0008] Preferably, in step S1, the identifying the target through the visual recognition system to complete the initial calibration specifically includes: using a high-resolution camera to collect high-precision target images, and solving the rotation matrix in the coordinate transformation model through the least square method optimization algorithm and translation vector T , the coordinate transformation model is:
[0009] wherein, Let be the coordinates of the i-th feature point in the global coordinate system after transformation. For the first i The original coordinates of each feature point in the camera coordinate system are determined. High-precision target images distributed across an array are acquired using a single high-resolution camera, and the rotation matrix is optimized using the least squares method. R With translation vector T This improved the accuracy and robustness of the initial calibration, ensuring a more accurate and reliable establishment of the global coordinate system, and laying a stable spatial reference for subsequent adaptive segmentation and stitching exposure.
[0010] Preferably, in step S2, the LDI machine can adjust the fused grayscale variance before adaptively segmenting the pattern to be exposed. Line width uniformity With edge sharpness Adjust the binarization threshold of the pattern to be exposed, the binarization threshold The calculation method is as follows:
[0011]
[0012]
[0013]
[0014]
[0015] in Threshold T Normalized proportion of foreground pixels in the background Threshold T Background pixel normalization ratio of the background. Threshold T The average grayscale value of the lower foreground pixel. Thresholds T The average grayscale value of the background pixels. The total number of sampled lines, For the first i The actual measured line width of the sampling line. The average line width of all sampled lines. This represents the total number of sampled edge points. Indicates the image at point gradient magnitude at that point For the comprehensive criterion function, To incorporate the weighting coefficients of the gray-level variance term, To integrate grayscale variance The maximum value among all candidate binarization thresholds. For line width uniformity The maximum value among all candidate binarization thresholds. The weighting coefficient for the uniform line width term. For edge sharpness The maximum value among all candidate binarization thresholds. These are the weighting coefficients for the edge sharpness term. The fused grayscale variance is constructed before adaptive segmentation. Line width uniformity With edge sharpness Comprehensive criterion function And solve for Maximize the optimal binarization threshold This invention achieves adaptive optimization of the binarization threshold of the exposed pattern, effectively avoiding image distortion or loss of detail caused by a fixed threshold. It provides a more accurate binarized image input for subsequent image density evaluation and block path planning, thereby improving the accuracy and material adaptability of adaptive block segmentation.
[0016] Preferably, in step S2, the specific process of adaptively dividing the pattern to be exposed into blocks includes: constructing the pattern density function of the region to be exposed. Calculate candidate boundaries Neighborhood Average graphic density within According to the average graphic density Establish a function to minimize the boundary cost. According to the minimized boundary cost function The block path is determined, and the pattern to be exposed is divided into multiple sub-regions. A graphic density function for each sub-region is then constructed. The algorithm calculates the average graphic density in the neighborhood of the candidate boundary and determines the block path with the goal of minimizing the boundary cost function. This enables the automatic tracking and fitting of the block boundary to sparse areas of the graphic, allowing the splicing seam to adaptively avoid fine lines and key graphic areas. This fundamentally reduces the damage to graphic integrity caused by splicing errors and improves the block quality under different pattern types.
[0017] Preferably, in step S2, the average graphic density Minimize the boundary cost function The calculation method is as follows:
[0018]
[0019] in, The set of sub-blocks obtained from the segmentation, For the set of all candidate boundaries, For the boundary The length of the partitioning path planning problem is transformed into an optimization problem of minimizing boundary costs by formulating the average graphic density and minimizing the boundary cost function. This makes the determination of the partitioning boundary have a clear mathematical basis and reproducibility, and can accurately guide the splicing seam to the region with the lowest graphic density while ensuring computational efficiency. This achieves quantitative decision-making and global optimization of the partitioning strategy.
[0020] Preferably, in step S3, a stitch fidelity index is constructed based on the linear stitching error parameter. The specific process is as follows:
[0021]
[0022]
[0023]
[0024]
[0025]
[0026] in, For positional deviation, This refers to the positional deviation in the X direction. For the positional deviation in the Y direction, To match logarithms, For angular deviation, For line width consistency deviation, To design line width, , For the first For the coordinates of the matching feature points in the first sub-image, , For the first For the coordinates of the matching feature points in the second sub-image, For the first For the principal orientation angle of the matching feature points in the first sub-image, For the first For the principal orientation angle of the matched feature points in the second sub-image, For the first The line width on the A side of the subgraph. For the first The line width on the B side of the subgraph. This is determined by constructing a stitch fidelity index. By weighting and integrating the three indicators of X / Y positional deviation, angle deviation, and line width consistency deviation into a single comprehensive evaluation indicator, a holistic and multi-dimensional quantitative assessment of splicing quality is achieved. This avoids the one-sidedness of judgment caused by a single indicator, making the identification of splicing defects more comprehensive and accurate. It provides a unified trigger criterion for subsequent dynamic compensation and secondary calibration, and simplifies the judgment logic of the self-inspection process.
[0027] Preferably, in step S4, the edge roughness and repeatability error The calculation methods are as follows:
[0028]
[0029] in, This represents the total number of edge sampling points. For the first i The edge position coordinates of each sampling point for K Average edge position of each sampling point; The number of times the measurement is repeated. For the first j The splicing error of this measurement. for M The average error of the measurements. This is determined by defining edge roughness. and repeatability error The specific calculation formula expands the assessment of splicing quality from a single positional deviation, angular deviation, and linewidth consistency to two new dimensions: local edge fluctuations and consistency across multiple calibrations. The smoothness of the stitched edges relative to an ideal straight line is quantified by calculating the standard deviation of the edge sampling point positions. The stability of calibration compensation is assessed by calculating the standard deviation of errors from multiple repeated measurements; these two indicators are related to the suture fidelity index. Together, they form a three-tiered quality assessment system encompassing stitch fidelity, edge quality, and repeatability consistency. This system not only identifies macroscopic misalignments in the splicing area but also detects hidden issues at the microscopic scale, such as edge jaggedness, burrs, and compensation value oscillations. This provides a more stringent joint criterion for the qualification judgment of secondary calibration, thereby effectively improving the edge integrity, detail fidelity, and process consistency of the spliced graphics during long-term continuous production.
[0030] Preferably, the secondary calibration in step S4 includes sharing information from the initial and secondary calibrations through a data fusion algorithm to form a unified coordinate mapping model.
[0031] in, For coordinate mapping model, These are the initial calibration results. The initial weights were calibrated, and the secondary calibration results based on linkage detection were as follows: , The secondary calibration weights are defined by using a data fusion algorithm to share information from both the initial and secondary calibrations, forming a unified coordinate mapping model. This achieves the complementary advantages of global calibration and local dynamic compensation, ensuring that the final coordinate mapping retains the stability of the initial global reference while incorporating real-time corrections from online detection, thus improving the overall positioning accuracy of the system.
[0032] Preferably, the initial calibration weights and secondary calibration weights are determined based on the initial calibration confidence level and the secondary calibration confidence level through dynamic normalization:
[0033]
[0034]
[0035]
[0036] in, for The initial weighting of time points, for The initial calibration confidence level at time point, for The second calibration confidence level at time t. The attenuation rate, The growth coefficient, The number of consecutive successful detections is defined. By limiting the weights and dynamically normalizing based on confidence, the initial calibration confidence decreases linearly with runtime, while the secondary calibration confidence increases exponentially with the number of consecutive successful detections. This allows the system to rely more on the rapidly established global calibration in the early stages of production, and gradually transition to a system dominated by local dynamic compensation as online detection data accumulates in the later stages of production. This achieves adaptive optimization of the calibration model, balancing startup efficiency and long-term stability.
[0037] The present invention also provides an adaptive segmentation, stitching exposure and flight self-testing system for digital masks of LDI machine, including a visual recognition system, a working substrate, a motion adjustment platform, an exposure optical system, a data processing and control system and a frame. The visual recognition system, the motion adjustment platform and the exposure optical system are all fixed on the frame. The working substrate and the exposure optical system are both connected to the motion adjustment platform. The visual recognition system and the exposure optical system are both directly opposite the working substrate. The visual recognition system is used to acquire high-precision target images arranged in an array, and works with the data processing and control system to solve the rotation matrix and translation vector, complete the initial calibration of the whole machine and establish a global coordinate system, and acquire the spliced workpiece image online without interrupting exposure, and extract the edge features of the spliced straight line. The motion adjustment platform is used to support the working substrate and the exposure optical system, and to feed back real-time position data to the data processing and control system. The data processing and control system includes an adaptive block operation unit, a stitching error calculation unit, a platform dynamic compensation unit, a lens distortion correction unit, and a fault alarm unit. The adaptive block operation unit is used for adaptive partitioning of the exposure pattern; the stitching error calculation unit is used to calculate the linear stitching error; the platform dynamic compensation unit and the lens distortion correction unit are used to dynamically correct the motion positioning parameters and lens distortion parameters, respectively; the fault alarm unit is used to control the equipment to stop and alarm when the linear stitching error still exceeds the preset tolerance range after multiple compensations.
[0038] Compared with the prior art, the beneficial effects of the present invention are: 1. Achieve adaptive segmentation and high-precision splicing to improve exposure quality and material adaptability. By analyzing the pattern density and feature distribution of the pattern to be exposed, different segmentation strategies are adopted for different types of substrates. Large-size segments are used for the splicing area, and small-size segments are used for the fine area while avoiding key patterns. This can effectively reduce the impact of splicing seams on pattern integrity and significantly improve the exposure quality of different patterns. 2. Closed-loop dynamic compensation is achieved through flight testing and secondary calibration to ensure splicing accuracy in continuous production. The vision system is used to simultaneously acquire images of overlapping areas during the exposure process, calculate positional deviation, angle deviation and line width consistency in real time, and dynamically compensate for the positioning parameters of the motion platform and lens distortion parameters accordingly, forming a closed-loop control of detection-compensation-re-verification, so as to maintain splicing accuracy without interrupting production. 3. Construct a two-level calibration and confidence-weighted fusion mechanism to balance initialization efficiency and long-term stability. The initial calibration quickly establishes the global coordinate system, the secondary calibration refines local errors online, and a unified coordinate mapping model is formed through dynamic confidence-weighted fusion. This ensures rapid equipment startup and production, and allows online detection to gradually lead accuracy correction during continuous production, improving the consistency of batch production and the robustness of long-term operation. Attached Figure Description
[0039] Figure 1 A flowchart for the adaptive segmentation, stitching exposure, and flight self-testing method of digital mask for LDI cameras; Figure 2 This refers to the substrate part number file in Example 1; Figure 3 This is a schematic diagram illustrating the effect of fixed segmentation. Figure 4 This is a schematic diagram of the adaptive block segmentation effect in Example 1; Figure 5 This is a schematic diagram of the positional deviation in Example 1; Figure 6 This is a schematic diagram of the angle deviation in Example 1; Figure 7 This is a schematic diagram illustrating the linewidth consistency in Example 1; Figure 8 This is a schematic diagram of an LDI machine.
[0040] In the attached diagram: 100, visual recognition system; 200, working substrate; 300, motion adjustment platform; 400, exposure optical system. Detailed Implementation
[0041] The present invention will be further described below with reference to specific embodiments. The accompanying drawings are for illustrative purposes only, representing schematic diagrams rather than actual physical objects, and should not be construed as limiting the scope of this patent. To better illustrate the embodiments of the present invention, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0042] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.
[0043] Example 1 This embodiment is the first embodiment of the LDI (Low-Input Digital Mask) adaptive segmentation, stitching exposure, and flight self-test method, including the following steps: S1. The visual recognition system accurately identifies multiple targets evenly distributed in an array within the exposure area, thereby solving the coordinate transformation model, completing the initial calibration of the system, establishing a global coordinate system, and inputting the calibration parameters into the LDI machine. S2. Based on the initial calibration results and the position feedback of the motion adjustment platform, the LDI machine adaptively divides the pattern to be exposed into multiple sub-images and controls the laser to expose the multiple sub-images sequentially at predetermined positions. The S3 and LDI cameras simultaneously acquire images of the exposure image, composed of sub-images, during the exposure process using a visual recognition system. An image feature extraction algorithm is employed to extract the stitching line edges, which are then compared with the preset exposure image. The linear stitching error between adjacent sub-images is calculated, including X-axis positional deviation, Y-axis positional deviation, angle deviation, and line width consistency deviation. A stitching fidelity index is then constructed based on these linear stitching error parameters. If the suture fidelity index If the error is within the preset tolerance range, the self-inspection process is completed. If the linear splicing error is not within the preset tolerance range, proceed to step S4 for correction. S4. Based on the linear stitching error, dynamically compensate the positioning control parameters of the motion adjustment platform and simultaneously correct the lens distortion parameters of the visual recognition system to complete the secondary calibration; then return to step S3 to re-perform image acquisition, linear stitching error calculation, and stitching fidelity index. Calculate and further test the edge roughness after secondary calibration. and repeatability error If the suture fidelity index Edge roughness Repeatability error If all values are within the corresponding preset tolerance range, the self-inspection process is complete; if the suture fidelity index... Edge roughness Repeatability error If any item still exceeds the corresponding preset tolerance range, then repeat the dynamic compensation and synchronization correction in step S4, and then return to step S3. If the suture fidelity index... Edge roughness Repeatability error If any item exceeds the corresponding preset tolerance range, the system will shut down and alarm, prompting manual intervention.
[0044] like Figure 1As shown, in the LDI machine digital mask adaptive blocking, stitching exposure and in-flight self-inspection method of the present invention, first, a vision recognition system is used to collect images of a plurality of high-precision targets distributed in an array within an exposure area, a coordinate transformation model is solved to complete initial calibration and establish a global coordinate system; subsequently, based on the initial calibration result and position feedback of the motion platform, the system performs adaptive blocking on the pattern to be exposed, divides the large-format pattern into a plurality of sub-images and sequentially performs exposure and stitching; during the continuous exposure process, the system reuses the same vision recognition system, synchronously collects images of stitching overlapping regions in an in-flight detection manner without interrupting the motion, extracts stitching straight line edge features, calculates X / Y direction position deviation, angle deviation and line width consistency deviation, and constructs a stitching fidelity index on this basis as a comprehensive evaluation indicator of stitching quality; if exceeds the tolerance range, dynamic compensation is performed on the motion platform positioning parameters and lens distortion parameters according to the measured error to complete secondary calibration, then the process returns to the detection step for recalculation , and edge roughness is further detected at the same time and repeatability error , only when , , all three indicators meet the corresponding preset tolerance requirements, the self-inspection is determined to be qualified; if any one indicator is not satisfied, the closed-loop process of calibration and detection is repeated, and the machine stops and gives an alarm after three consecutive failures. This control method realizes hardware resource sharing for calibration and detection by reusing the same vision system, and combines the adaptive blocking strategy and multi-indicator comprehensive evaluation system, which can perceive in real time and perform closed-loop compensation for stitching deviations caused by dynamic factors such as mechanical wear, thermal deformation or optical distortion without stopping the machine or interrupting the production cycle, significantly suppress defects such as pattern misalignment, rough edges and inconsistent line widths, and at the same time through , , the joint judgment of the three indicators ensures that the stitching quality meets the standards in multiple dimensions, greatly improves the pattern integrity, consistency and long-term process stability of batch exposure, reduces the frequency of manual intervention, and improves the overall production capacity and automation level of the equipment.
[0045] In step S1, the identifying the target by the vision recognition system to complete initial calibration specifically includes: using a high-resolution camera to collect high-precision target images, and solving the rotation matrix in the coordinate transformation model through a least square method optimization algorithm R and translation vector T , the coordinate transformation model is:
[0046] wherein, is the coordinate of the i-th feature point in the global coordinate system after transformation, Let be the original coordinates of the i-th feature point in the camera coordinate system. High-precision target images distributed across an array are acquired using a single high-resolution camera, and the rotation matrix is optimized using the least squares method. R With translation vector T This improved the accuracy and robustness of the initial calibration, ensuring a more accurate and reliable establishment of the global coordinate system, and laying a stable spatial reference for subsequent adaptive segmentation and stitching exposure.
[0047] In step S2, before the LDI machine adaptively segments the pattern to be exposed, it can determine the fusion grayscale variance. Line width uniformity With edge sharpness Adjust the binarization threshold of the pattern to be exposed. The calculation method is as follows:
[0048]
[0049]
[0050]
[0051]
[0052] in Threshold T Normalized proportion of foreground pixels in the background Threshold T Background pixel normalization ratio of the background. Threshold T The average grayscale value of the lower foreground pixel. Thresholds T The average grayscale value of the background pixels. The total number of sampled lines, For the first i The actual measured line width of the sampling line. The average line width of all sampled lines. This represents the total number of sampled edge points. Indicates the image at point gradient magnitude at that point For the comprehensive criterion function, To incorporate the weighting coefficients of the gray-level variance term, To integrate grayscale variance The maximum value among all candidate binarization thresholds. For line width uniformity The maximum value among all candidate binarization thresholds. The weighting coefficient for the uniform line width term. For edge sharpness The maximum value among all candidate binarization thresholds. These are the weighting coefficients for the edge sharpness term. The fused grayscale variance is constructed before adaptive segmentation. Line width uniformity With edge sharpness Comprehensive criterion function And solve for Maximize the optimal binarization threshold This invention achieves adaptive optimization of the binarization threshold of the exposed pattern, effectively avoiding image distortion or loss of detail caused by a fixed threshold. It provides a more accurate binarized image input for subsequent image density evaluation and block path planning, thereby improving the accuracy and material adaptability of adaptive block segmentation.
[0053] In step S2, the specific process of adaptively segmenting the pattern to be exposed includes: constructing the pattern density function of the area to be exposed. Calculate candidate boundaries Neighborhood Average graphic density within According to average graphic density Establish a function to minimize the boundary cost. According to minimizing the boundary cost function The pattern to be exposed is divided into multiple sub-regions by determining the segmentation path. A pattern density function for each sub-region is then constructed. The algorithm calculates the average graphic density in the neighborhood of the candidate boundary and determines the block path with the goal of minimizing the boundary cost function. This enables the automatic tracking and fitting of the block boundary to sparse areas of the graphic, allowing the splicing seam to adaptively avoid fine lines and key graphic areas. This fundamentally reduces the damage to graphic integrity caused by splicing errors and improves the block quality under different pattern types.
[0054] In step S2, the average graphic density Minimize the boundary cost function The calculation method is as follows:
[0055]
[0056] in, The set of sub-blocks obtained from the segmentation, For the set of all candidate boundaries, For the boundary The length of the partitioned path planning problem is determined by formulating the average graphic density and minimizing the boundary cost function. This transforms the problem into an optimization problem of minimizing the boundary cost, providing a clear mathematical basis and reproducibility for determining the partitioned boundaries. It ensures computational efficiency while accurately guiding the seams to the region with the lowest graphic density, achieving quantitative decision-making and global optimization of the partitioning strategy. The PCB part number file used in this embodiment is as follows: Figure 2 As shown, Figure 3 This is a schematic diagram showing the effect of using fixed blocks to divide the pattern to be exposed. The dividing lines between the sub-regions pass through areas with high pattern density, affecting the exposure effect. Figure 4 This is a schematic diagram illustrating the effect of adaptive segmentation in this embodiment, which avoids passing through areas with a large number of patterns, thereby improving the exposure effect.
[0057] In step S3, a stitch fidelity index is constructed based on the linear stitching error parameter. The specific process is as follows:
[0058]
[0059]
[0060]
[0061]
[0062]
[0063] in, For positional deviation, This refers to the positional deviation in the X direction. For the positional deviation in the Y direction, To match logarithms, For angular deviation, For line width consistency deviation, To design line width, , For the first For the coordinates of the matching feature points in the first sub-image, , For the first For the coordinates of the matching feature points in the second sub-image, For the first For the principal orientation angle of the matching feature points in the first sub-image, For the first For the principal orientation angle of the matched feature points in the second sub-image, For the first The line width on the A side of the subgraph. For the first The line width on the B side of the subgraph. This is determined by constructing a stitch fidelity index. This method weights and integrates three indicators—X / Y direction position deviation, angle deviation, and linewidth consistency deviation—into a single comprehensive evaluation index. This achieves a holistic, multi-dimensional quantitative assessment of splicing quality, avoiding the bias of single-indicator judgments. It makes the identification of splicing defects more comprehensive and accurate, providing a unified trigger criterion for subsequent dynamic compensation and secondary calibration, and simplifying the judgment logic of the self-inspection process. In this embodiment, the preset tolerance range can be set to the X-direction position deviation. ≤5μm, Y-axis position deviation ≤5μm, angular deviation ≤1°, line width consistency deviation ≤5μm. For example... Figure 5 As shown, the detection result of the position deviation in this embodiment is 20μm; Figure 6 As shown, the angle deviation in this embodiment The test result was 2°; if Figure 7 As shown, the line width consistency deviation in this embodiment The value is 20μm. Based on the X-axis positional deviation, Y-axis positional deviation, angle deviation, and line width consistency deviation, a set of linear splicing error parameters is formed, which serves as the basis for subsequent dynamic compensation, secondary calibration, and calculation of the stitch fidelity index.
[0064] In step S4, edge roughness and repeatability error The calculation methods are as follows:
[0065]
[0066] in, This represents the total number of edge sampling points. For the first i The edge position coordinates of each sampling point for K Average edge position of each sampling point; The number of times the measurement is repeated. For the first j The splicing error of this measurement. for M The average error of the measurements. This is determined by defining edge roughness. and repeatability error The specific calculation formula expands the assessment of splicing quality from a single positional deviation, angular deviation, and linewidth consistency to two new dimensions: local edge fluctuations and consistency across multiple calibrations. The smoothness of the stitched edges relative to an ideal straight line is quantified by calculating the standard deviation of the edge sampling point positions. The stability of calibration compensation is assessed by calculating the standard deviation of errors from multiple repeated measurements; these two indicators are related to the suture fidelity index. Together, they form a three-tiered quality assessment system encompassing stitch fidelity, edge quality, and repeatability consistency. This system not only identifies macroscopic misalignments in the splicing area but also detects hidden issues at the microscopic scale, such as edge jaggedness, burrs, and compensation value oscillations. This provides a more stringent joint criterion for the qualification judgment of secondary calibration, thereby effectively improving the edge integrity, detail fidelity, and process consistency of the spliced graphics during long-term continuous production.
[0067] In this embodiment, when the effective graphic size of a sub-region is smaller than the exposure field of view and it is a small-area panel (SPB) arranged in a regular array, the unit gaps are extracted as preset channels, the segmentation path is controlled to pass through the gaps, and the first size is used for segmentation. When the effective graphic size of a sub-region is greater than or equal to the exposure field of view, it is a large-area panel (LPB), and segmentation is preferentially performed along the panel gaps. If there are no usable gaps, the path with the lowest graphic density is selected, and the second size is used for segmentation. When the sub-region is a large-area continuous panel (LCB) lacking physical separation, the weak edge areas of the graphic transition are identified, and the second size is used for segmentation to form multiple sub-images. This achieves the goal of using large-size segmentation for sparse graphic areas to reduce the number of segments and improve exposure efficiency, and using small-size segmentation for dense graphic areas to avoid damage to fine lines by the splicing seams, thereby optimizing the production cycle while ensuring splicing quality.
[0068] In step S4, dynamic compensation of the positioning control parameters of the motion adjustment platform includes adjusting the positioning parameters of the motion adjustment platform according to the linear stitching error and updating the geometric correction mapping table of the exposure graphic data. By limiting dynamic compensation to include adjusting the positioning parameters of the motion adjustment platform and updating the geometric correction mapping table of the exposure graphic data, direct correction of mechanical positioning errors is achieved, ensuring the graphic alignment accuracy of the stitching overlap area and avoiding graphic misalignment defects caused by accumulated platform errors.
[0069] In step S4, the synchronous correction of lens distortion parameters of the visual recognition system includes correcting the radial and tangential distortion coefficients based on the image feature points and straight line edge points of the exposed area acquired by the visual recognition system in step S3, and determining the theoretical positional difference distortion parameters between the actual acquisition position and the current distortion parameter projection based on the reprojection error of the image feature points. By limiting the correction of lens distortion parameters to include correcting the radial and tangential distortion coefficients and optimizing them based on the reprojection error of the image feature points, the influence of the optical distortion of the visual recognition system itself on the detection results is effectively eliminated, improving the data authenticity of flight detection and the compensation accuracy of secondary calibration.
[0070] Example 2 This embodiment is the second embodiment of the LDI (Low-Input Digital Mask) adaptive segmentation, stitching exposure, and flight self-test method. Step S4, completing the secondary calibration, includes sharing information from the initial and secondary calibrations through a data fusion algorithm to form a unified coordinate mapping model.
[0071] in, For coordinate mapping model, These are the initial calibration results. The initial weights were calibrated, and the secondary calibration results based on linkage detection were as follows: , The secondary calibration weights are defined by using a data fusion algorithm to share information from both the initial and secondary calibrations, forming a unified coordinate mapping model. This achieves the complementary advantages of global calibration and local dynamic compensation, ensuring that the final coordinate mapping retains the stability of the initial global reference while incorporating real-time corrections from online detection, thus improving the overall positioning accuracy of the system.
[0072] In this embodiment, the initial calibration weights and secondary calibration weights are determined based on the initial calibration confidence level and the secondary calibration confidence level through dynamic normalization.
[0073]
[0074]
[0075]
[0076] in, for The initial weighting of time points, for The initial calibration confidence level at time point, for The second calibration confidence level at time t. The attenuation rate, The growth coefficient, The number of consecutive successful detections is defined. By limiting the weights and dynamically normalizing based on confidence levels, the initial calibration confidence level decreases linearly with runtime, while the secondary calibration confidence level increases exponentially with the number of consecutive successful detections. This allows the system to rely more on rapidly established global calibration in the early stages of production, and gradually transition to a system dominated by local dynamic compensation as online detection data accumulates in the later stages of production. This achieves adaptive optimization of the calibration model, balancing startup efficiency and long-term stability. In this embodiment, the initial state... =1、 =0, relying on initial calibration to quickly establish a baseline, in continuous production Rise to 0.7 =0.3, with accuracy dominated by secondary calibration.
[0077] In step S3, the triggering conditions for synchronous image processing can be configured to trigger after the exposure of a single workpiece is completed, after the exposure of a preset number of workpieces is completed, or when environmental parameters exceed limits. By setting the triggering conditions for flight inspection to a configurable mode—that is, triggering according to the completion cycle of a single workpiece or a preset number of workpieces, or actively triggering when environmental parameters such as temperature, humidity, and vibration exceed limits—the flexibility and intelligence of the inspection strategy are achieved. On the one hand, users can freely adjust the inspection frequency according to production cycle and quality requirements, avoiding over-inspection affecting production capacity or under-inspection leading to missed defects; on the other hand, when abnormal environmental parameters may cause splicing errors, the system can automatically trigger inspection and promptly detect potential problems, thereby optimizing the equipment's production efficiency and adaptability while ensuring exposure quality.
[0078] Example 3 This embodiment is an example of an LDI (Low-Input Digital Mask) machine's adaptive segmentation, stitching exposure, and flight self-test system. Figure 8 As shown, the system includes a visual recognition system 100, a working substrate 200, a motion adjustment platform 300, an exposure optical system 400, a data processing and control system, and a frame. The visual recognition system 100, the motion adjustment platform 300, and the exposure optical system 400 are all fixed on the frame. The working substrate 200 and the exposure optical system 400 are both connected to the motion adjustment platform 300. The visual recognition system 100 and the exposure optical system 400 are both directly opposite the working substrate 200. The visual recognition system 100 is used to acquire high-precision target images arranged in an array, and works with the data processing and control system to solve the rotation matrix and translation vector, complete the initial calibration of the whole machine and establish a global coordinate system, and acquire the spliced workpiece image online without interrupting exposure, and extract the edge features of the spliced straight line. The motion adjustment platform 300 is used to support the working substrate 200 and the exposure optical system 400, and to feed back real-time position data to the data processing and control system. The data processing and control system includes an adaptive block operation unit, a stitching error calculation unit, a platform dynamic compensation unit, a lens distortion correction unit, and a fault alarm unit. The adaptive block operation unit is used for adaptive partitioning of the exposure pattern; the stitching error calculation unit is used to calculate the linear stitching error; the platform dynamic compensation unit and the lens distortion correction unit are used to perform dynamic correction of motion positioning parameters and lens distortion parameters, respectively; the fault alarm unit is used to control the equipment to stop and alarm when the linear stitching error still exceeds the preset tolerance range after multiple compensations.
[0079] In the specific implementation of the above embodiments, the technical features can be combined in any non-contradictory way. For the sake of brevity, not all possible combinations of the above technical features are described. However, as long as the combination of these technical features is not contradictory, it should be considered to be within the scope of this specification.
[0080] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.
Claims
1. A method for LDI machine digital mask adaptive tiling, stitching exposure and flight self-checking, characterized in that, Includes the following steps: S1. Identify multiple high-precision targets evenly distributed in an array within the exposure area using a visual recognition system, thereby solving the coordinate transformation model, completing the initial calibration of the system, establishing a global coordinate system, and inputting the calibration parameters into the LDI machine. S2. Based on the initial calibration results and the position feedback of the motion adjustment platform, the LDI machine adaptively divides the pattern to be exposed into multiple sub-images and controls the laser to expose the multiple sub-images sequentially at predetermined positions. S3. The LDI machine simultaneously acquires the exposed image composed of the sub-images during the exposure process using a visual recognition system. It extracts the stitching line edges using an image feature extraction algorithm and compares them with a preset exposed image. It calculates the linear stitching error parameters between adjacent sub-images. These linear stitching error parameters include X-axis position deviation, Y-axis position deviation, angle deviation, and line width consistency deviation. A stitching fidelity index is then constructed based on these linear stitching error parameters. ; If the suture fidelity index If the error is within the preset tolerance range, the self-test process is completed. If the linear splicing error is not within the preset tolerance range, proceed to step S4 for correction. S4. Based on the linear stitching error, dynamically compensate the positioning control parameters of the motion adjustment platform and simultaneously correct the lens distortion parameters of the visual recognition system to complete the secondary calibration. Then return to step S3 to re-perform image acquisition, linear stitching error calculation, and stitch fidelity index calculation. Calculate and further test the edge roughness after secondary calibration. and repeatability error If the suture fidelity index Edge roughness Repeatability error If all values are within the corresponding preset tolerance range, the self-inspection process is complete; if the suture fidelity index... Edge roughness Repeatability error If any item still exceeds the corresponding preset tolerance range, then the dynamic compensation and synchronization correction in step S4 are repeated, and then the process returns to step S3. If the suture fidelity index... Edge roughness Repeatability error If any item exceeds the corresponding preset tolerance range, the system will shut down and alarm, prompting manual intervention.
2. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 1, characterized in that, In step S1, the initial calibration by identifying the target through a visual recognition system specifically includes: acquiring high-precision target images using a high-resolution camera, and solving the rotation matrix in the coordinate transformation model using a least squares optimization algorithm. R With translation vector T The coordinate transformation model is as follows: in, For the first i The coordinates of each feature point in the global coordinate system after transformation For the first i The original coordinates of each feature point in the camera coordinate system.
3. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 1, characterized in that, In step S2, the LDI machine can adjust the fused grayscale variance before adaptively segmenting the pattern to be exposed. Line width uniformity With edge sharpness Adjust the binarization threshold of the pattern to be exposed, the binarization threshold The calculation method is as follows: in Threshold T Normalized proportion of foreground pixels in the background Threshold T Background pixel normalization ratio of the background. Threshold T The average grayscale value of the lower foreground pixel. Thresholds T The average grayscale value of the background pixels. The total number of sampled lines, For the first i The actual measured line width of the sampling line. The average line width of all sampled lines. This represents the total number of sampled edge points. Indicates the image at point gradient magnitude at that point For the comprehensive criterion function, To incorporate the weighting coefficients of the gray-level variance term, To integrate grayscale variance The maximum value among all candidate binarization thresholds. For line width uniformity The maximum value among all candidate binarization thresholds. The weighting coefficient for the uniform line width term. For edge sharpness The maximum value among all candidate binarization thresholds. This represents the weighting coefficient for the edge sharpness term.
4. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 3, characterized in that, In step S2, the specific process of adaptively dividing the pattern to be exposed into blocks includes: constructing the pattern density function of the area to be exposed. Calculate candidate boundaries Neighborhood Average graphic density within According to the average graphic density Establish a function to minimize the boundary cost. According to the minimized boundary cost function Determine the segmentation path and divide the pattern to be exposed into multiple sub-regions.
5. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 4, characterized in that, In step S2, the average graphic density Minimize the boundary cost function The calculation method is as follows: in, The set of sub-blocks obtained from the segmentation, For the set of all candidate boundaries, For the boundary The length.
6. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 1, characterized in that, In step S3, a stitch fidelity index is constructed based on the linear stitching error parameter. The specific process is as follows: in, For positional deviation, This refers to the positional deviation in the X direction. For the positional deviation in the Y direction, To match logarithms, Due to angular deviation, For line width consistency deviation, To design line width, , For the first For the coordinates of the matching feature points in the first sub-image, , For the first For the coordinates of the matching feature points in the second sub-image, For the first For the principal orientation angle of the matching feature points in the first sub-image, For the first For the principal orientation angle of the matched feature points in the second sub-image, For the first The line width on the A side of the subgraph. For the first The line width on the B side of the subgraph.
7. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 1, characterized in that, In step S4, edge roughness and repeatability error The calculation methods are as follows: in, This represents the total number of edge sampling points. Let be the edge position coordinates of the i-th sampling point. for K Average edge position of each sampling point; The number of times the measurement is repeated. Let j be the splicing error of the j-th measurement. Let be the average error of M measurements.
8. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 1, characterized in that, Step S4 involves completing the secondary calibration by sharing information from the initial and secondary calibrations through a data fusion algorithm to form a unified coordinate mapping model. in, For coordinate mapping model, These are the initial calibration results. The initial weights were calibrated, and the secondary calibration results based on linkage detection were as follows: , For secondary calibration weights.
9. The method for adaptive segmentation, stitching exposure, and flight self-testing of digital masks for LDI cameras according to claim 8, characterized in that, The initial calibration weights and secondary calibration weights are determined based on the initial calibration confidence level and the secondary calibration confidence level through dynamic normalization. in, for The initial weighting of time points, for The initial calibration confidence level at time point, for The second calibration confidence level at time t. The attenuation rate, The growth coefficient, This represents the number of consecutive successful tests.
10. An LDI machine digital mask adaptive segmentation, stitching exposure, and flight self-test system for implementing the method as described in any one of claims 1 to 9, characterized in that, The system includes a visual recognition system (100), a working substrate (200), a motion adjustment platform (300), an exposure optical system (400), a data processing and control system, and a frame. The visual recognition system (100), the motion adjustment platform (300), and the exposure optical system (400) are all fixed on the frame. The working substrate (200) and the exposure optical system (400) are both connected to the motion adjustment platform (300). The visual recognition system (100) and the exposure optical system (400) are both directly opposite the working substrate (200). The visual recognition system (100) is used to acquire high-precision target images arranged in an array, and to solve the rotation matrix and translation vector in conjunction with the data processing and control system to complete the initial calibration of the whole machine and establish a global coordinate system. Under the condition of uninterrupted exposure, it can acquire the spliced workpiece image online and extract the spliced straight line edge features. The motion adjustment platform (300) is used to support the working substrate (200) and the exposure optical system (400), and to feed back real-time position data to the data processing and control system; The data processing and control system includes an adaptive block operation unit, a stitching error calculation unit, a platform dynamic compensation unit, a lens distortion correction unit, and a fault alarm unit. The adaptive block operation unit is used for adaptive partitioning of the exposure pattern; the stitching error calculation unit is used to calculate the linear stitching error; the platform dynamic compensation unit and the lens distortion correction unit are used to dynamically correct the motion positioning parameters and lens distortion parameters, respectively; the fault alarm unit is used to control the equipment to stop and alarm when the linear stitching error still exceeds the preset tolerance range after multiple compensations.
