A multi-state target self-adaptive alignment vision method and system for an ldi device

Abstract

The embodiment of the application provides a multi-state target self-adaptive alignment visual method and system for an LDI device, the method comprising: collecting an original image of a target point, obtaining contrast data, and self-adaptively adjusting the original image; extracting a candidate ROI region in the original image based on a perception strategy; performing feature extraction on the candidate ROI region based on process recipe loading evaluation parameters, obtaining a state vector, and determining a region coverage state of the candidate ROI region based on the state vector and the evaluation parameters; selecting a corresponding image repair scheme to obtain a repaired ROI; determining an edge point cloud based on the repaired ROI and target point shape parameters, fitting based on the edge point cloud to determine the center coordinates of the target point; and combining the repaired ROI and the center coordinates, comparing target point parameter standards of a process recipe, and outputting a target point meeting accuracy requirements.

Description

Technical Field

[0001] This invention relates to the field of LDI visual alignment technology, and in particular to a multi-state target adaptive alignment visual method and system for LDI devices. Background Technology

[0002] Laser Direct Imaging (LDI) technology is widely used in the exposure and alignment processes of inner and outer layers of printed circuit boards (PCBs). Its alignment accuracy directly affects the accuracy of circuit stack-up and the final product yield. In LDI equipment, an alignment target image on the PCB is typically acquired through a vision system, and the exposure coordinates are calculated based on the geometric center of the target, thereby completing the precise alignment of the laser imaging.

[0003] Existing LDI alignment vision systems mostly employ fixed template matching, threshold segmentation, or single feature extraction methods to identify and locate target points. Typical solutions include: extracting target point contours based on grayscale thresholding and morphological processing; obtaining target point centers based on template matching or shape matching; and calculating alignment coordinates based on simple geometric fitting (such as circle fitting or crosshair fitting).

[0004] The above-mentioned solution can meet the basic alignment requirements when the target morphology is consistent and the imaging quality is stable. However, in actual production, PCB targets are often affected by factors such as inconsistent ink coverage, exposure residue, surface reflection, contamination, local defects, and batch differences in the board material, resulting in the target points exhibiting various state characteristics, such as no coverage, half coverage, full coverage, or local damage.

[0005] When the target state changes, existing technical solutions typically suffer from the following accuracy issues: for example, insufficient adaptability to changes in target state, and difficulty in using a single processing flow to accommodate different target morphologies; unstable target contour extraction and decreased center positioning accuracy when ink coverage or contrast is reduced; and a lack of effective discrimination mechanisms for abnormal targets, which can easily introduce incorrect alignment results. Furthermore, the algorithm parameters of the aforementioned existing technologies rely on manual experience for adjustment, resulting in high debugging costs and poor reusability across different machines and batches. Summary of the Invention

[0006] To address the problems existing in the prior art, embodiments of the present invention provide a multi-state target adaptive alignment vision method and system for LDI devices.

[0007] In a first aspect, embodiments of this specification provide a multi-state target adaptive alignment vision method for LDI devices, the method comprising:

[0008] The original image of the target point is acquired, the original image is preprocessed, and the contrast data of the original image is obtained. The original image is then adaptively adjusted based on the contrast data.

[0009] Candidate ROI regions are extracted from the original image based on a perceptual strategy, which includes deep learning, full target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction that resists highlights and contamination.

[0010] Based on the process formulation loading evaluation parameters, feature extraction is performed on the candidate ROI region to obtain a state vector, and the regional coverage status of the candidate ROI region is determined based on the state vector and the evaluation parameters.

[0011] Based on the coverage status of the region, the corresponding image restoration scheme is selected to obtain the restored ROI;

[0012] Based on the repaired ROI and target shape parameters, coarse coordinates are determined and a search line is generated. Based on the coarse coordinates and the search line, an edge point cloud is determined. Based on the edge point cloud fitting, the center coordinates of the target point are determined.

[0013] Based on the repaired ROI and center coordinates, the target parameter standards of the process formulation are compared, the matching score of the target is determined based on the comparison results, and the target with the matching score that meets the accuracy requirements is input.

[0014] Secondly, embodiments of this specification provide a multi-state target adaptive alignment vision system for LDI devices, the system comprising:

[0015] The acquisition module is used to acquire the original image of the target point, preprocess the original image, obtain the contrast data of the original image, and adaptively adjust the original image based on the contrast data.

[0016] The strategy module is used to extract candidate ROI regions from the original image based on a perceptual strategy, which includes deep learning, full target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction that resists highlights and contamination.

[0017] The extraction module is used to extract features from the candidate ROI region based on the process formula and evaluation parameters, obtain a state vector, and determine the regional coverage status of the candidate ROI region based on the state vector and evaluation parameters.

[0018] The repair module is used to select the corresponding image repair scheme based on the coverage status of the region to obtain the repaired ROI;

[0019] The search module is used to determine coarse coordinates and generate a search line based on the repaired ROI and target shape parameters, determine the edge point cloud based on the coarse coordinates and the search line, and determine the center coordinates of the target point based on the edge point cloud fitting.

[0020] The output module is used to combine the repaired ROI and center coordinates, compare the target parameter standards of the process formulation, determine the target matching score based on the comparison results, and output the target points whose matching scores meet the accuracy requirements.

[0021] Thirdly, embodiments of this specification provide an electronic device, including a processor and a memory;

[0022] The processor is connected to the memory;

[0023] The memory is used to store executable program code;

[0024] The processor runs a program corresponding to the executable program code stored in the memory to perform the methods described in one or more embodiments.

[0025] Fourthly, embodiments of this specification provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described multi-state target adaptive alignment vision method for LDI devices.

[0026] In view of the above, the beneficial effects of the technical solutions provided by some embodiments of this specification include at least the following:

[0027] In one or more embodiments of this specification, the original image of the target point is acquired, the original image is preprocessed, and the contrast data of the original image is obtained. The original image is adaptively adjusted based on the contrast data. Candidate ROI regions in the original image are extracted based on a perception strategy. Based on the process formulation loading evaluation parameters, features are extracted from the candidate ROI regions to obtain state vectors. The regional coverage state of the candidate ROI regions is determined based on the state vectors and evaluation parameters. Based on the regional coverage state, a corresponding image restoration scheme is selected to obtain the restored ROI. Based on the restored ROI and the target point shape parameters, coarse coordinates are determined and a search line is generated. The edge point cloud is determined based on the coarse coordinates and the search line. The center coordinates of the target point are determined based on the edge point cloud fitting. Combining the restored ROI and the center coordinates, the target point parameter standard of the process formulation is compared. Based on the comparison results, the matching score of the target point is determined, and the target point whose matching score meets the accuracy requirements is output. By employing differentiated processing strategies for different target states (such as using specific model scaling domains and image enhancement strategies for different film materials), the impact of abnormal target states on positioning accuracy is avoided. Furthermore, it can adapt to various actual production scenarios such as uneven ink coverage, contrast variations, local defects, and high light reflection, greatly reducing misjudgments and rework caused by target anomalies. It also includes multiple adaptive strategy extraction methods, combined with lightweight neural networks to replace the cumbersome parameter adjustments that are highly susceptible to environmental influences, reducing reliance on human experience. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart of a multi-state target adaptive alignment vision method for LDI devices provided in one embodiment of this specification.

[0030] Figure 2 This is a schematic diagram of the structure of a multi-state target adaptive alignment vision system for an LDI device, provided in one embodiment of this specification.

[0031] Figure 3 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this specification. Detailed Implementation

[0032] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed merely to enable those skilled in the art to better understand and implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. The function and arrangement of the elements discussed may be changed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the various examples. For example, the described methods may be performed in a different order than described, and steps may be added, omitted, or combined. Furthermore, features described in some examples may be combined in other examples.

[0033] As used herein, the term "comprising" and its variations are open terms meaning "including but not limited to". The term "based on" means "at least partially based on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other definitions, whether explicit or implicit, may be included below. Unless explicitly indicated by the context, the definition of a term shall remain consistent throughout the specification.

[0034] Please see Figure 1 , Figure 1 This document illustrates an overall flowchart of a multi-state target adaptive alignment vision method for LDI devices, as provided in an embodiment of this specification. The device driver model includes an application service layer, a driver interface layer, and a peripheral layer.

[0035] like Figure 1 As shown, the multi-state target adaptive alignment vision method for LDI devices includes at least the following steps:

[0036] Step S102: Acquire the original image of the target point, preprocess the original image, obtain the contrast data of the original image, and adaptively adjust the original image based on the contrast data.

[0037] Specifically, the configuration type of the PCB is obtained, including but not limited to the PCB board type, target metal layer (copper / gold, etc.), and the preset formula of the current process segment (inner layer / outer layer / solder resist). Grayscale images of the target points in a predetermined area of ​​the PCB are also acquired, and when acquiring grayscale images, other conditions such as lighting and imaging parameters are kept as constant as possible.

[0038] After acquiring the original image, preprocessing is performed, including but not limited to denoising (such as median filtering or Gaussian filtering, with the kernel size automatically determined based on image resolution and target size), removing sensor thermal noise and particle noise. Flat-field correction is performed to eliminate brightness differences between the center and edges of the image caused by uneven lighting or lens vignetting. This results in an original image with low noise levels and a uniform background. Then, the global contrast metric (such as grayscale standard deviation) of the original image is calculated, and combined with local image features of the ROI (Region of Interest), such as local variance and edge density, the global contrast of the original image is adjusted. For example, if the global contrast is high and there are no low-contrast defects in local areas, the original image is directly output; or if the image is generally too dark or too bright, resulting in insufficient dynamic range, global histogram equalization of the original image can be performed; or if only local target areas exhibit low contrast due to ink coverage, or if there are both highlight and dark areas in the image, adaptive histogram equalization with contrast limitation is selected. This ensures that the most suitable enhancement algorithm is automatically triggered under low-contrast conditions.

[0039] Furthermore, when performing global histogram equalization on the original image, the gray-level histogram of the original image is calculated, and a cumulative distribution function is generated to map the gray-level values ​​to the entire 0-255 range, making the histogram of the output image approximately uniformly distributed. When performing contrast-limited adaptive histogram equalization on the original image, the image is divided into multiple local blocks, and the histogram is calculated and equalized independently within each block. The histogram of each block is clipped and limited, and the portion exceeding the limit is evenly distributed across all gray levels to limit excessive amplification of local contrast. Bilinear interpolation is used to merge the boundaries of each block, eliminating block artifacts and outputting an enhanced image. The specific parameters (block size, clipping coefficient) are automatically adjusted based on the estimated size of the target points and the degree of ink coverage: for large target points with severe coverage and blurred texture, larger blocks and lower clipping limits are used; for target points with rich details, smaller blocks and slightly higher limits are used.

[0040] Step S104: Extract candidate ROI regions from the original image based on a perception strategy. The perception strategy includes deep learning, full target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction to resist highlights and contamination.

[0041] Specifically, target region fusion extraction is performed using a multi-heterogeneous sensing strategy. This strategy can include, but is not limited to, deep learning, full-target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction resistant to highlights and contamination. The four branches mentioned above are as follows:

[0042] The deep learning extraction branch loads a pre-trained lightweight object detection model (such as YOLOv5-Nano or MobileNet-SSD), scales the input image to the model-defined size, and normalizes it. Then, it runs the model, outputting predicted bounding boxes, class probabilities, and confidence scores. The model is trained with a large number of diverse target state samples (uncovered, partially covered, fully covered, damaged, specular, etc.). Low-probability predictions are then eliminated based on a confidence threshold, and non-maximum suppression is applied to eliminate overlapping boxes. Finally, the retained bounding boxes are mapped back to the original image resolution, and the box margins are expanded to ensure all target edges are included, generating candidate regions. Each region is accompanied by the model's output confidence score as the confidence level of that ROI.

[0043] Full Target Extraction Based on Extremum Regions and Morphology (MSER): The MSER algorithm is run on the original image, with the parameter Δ (region area change stability threshold) automatically set based on the estimated pixel size of the target (e.g., a smaller Δ for small target points). A set of extremum regions is extracted, each with a connected component of arbitrary shape. Then, the area, roundness, aspect ratio of the minimum bounding rectangle, and convexity are calculated for each region. Obvious non-target regions are removed according to the preset target point geometry. Finally, highly overlapping stable regions are merged, and the center coordinates and major / minor axes are obtained through ellipse fitting as candidate ROIs.

[0044] Sector extraction based on local contours: Gradient magnitude and direction are calculated on the image, strong edges are preserved, and a set of sector structures (e.g., 15°, 30°, 45° arc segments) are predefined, allowing rotation and scaling. Several sector sizes are preset for each target point diameter; morphological hits or gradient direction matching are performed using the sector structures at different angles (e.g., calculating the consistency of the angle between the average gradient direction and the radial direction within the sector region) to detect edge segments in the image that satisfy the characteristics of circular arc segments; then, based on the consistency of the arc center, multiple arc segments belonging to the same circle are clustered to generate a complete circular candidate region; and the boundary is expanded to generate a ROI with the inferred center as the center, and the region is marked as generated by arc segment inference.

[0045] Specific extraction to combat highlights and contamination: Saturation, gradient divergence, or frequency domain analysis is used to locate specular highlight areas (overexposed saturated pixel clusters) and dark contamination areas in the image. Image gradient-based smoothing is applied to the neighborhood of the highlight areas, or a light separation algorithm is used to suppress strong highlights. After processing, roughly circular regions of darkness or light are identified (combining local binarization and connected component analysis). Edge closure algorithms are used to connect target point edge fragments interrupted by highlights. If the highlight completely covers the central part and cannot be recovered, the outer ring or residual arc edge of the target point is used, and the center is inferred from the outer ring or residual arc edge to expand the boundary and generate a Region of Interest (ROI). This region is then marked as having a blind zone risk.

[0046] For the above branches, the following rules can be followed when performing target region fusion extraction: Obtain the original image scene and select different branches for different scene types. For example, for scenes where traditional feature extraction is difficult due to full ink coverage, low contrast, or local deformation, a deep learning branch can be used. For original images with complete morphology, uniform and homogeneous regions (e.g., uncovered copper targets forming stable dark areas in ink, or bright areas under full ink coverage), a full target extraction branch based on extreme regions and morphology can be used. For targets with local damage (e.g., notches, scratches, ink damage), a branch based on extreme regions and morphology can be used. In scenarios involving localized peeling, a fan-shaped extraction branch based on local contours can be used, abandoning the integrity assumption. Only a portion of the arc segment is needed to accurately infer the complete circle, making it the most direct and robust extraction strategy for damaged target points. For original images with high-gloss specular reflections caused by secondary coatings, protective films, etc., and dark spots in the image caused by ink contamination, where these target point areas have defects in consistency and edge continuity, specific extraction against high-gloss and contamination can be used to attempt to recover the occluded information. If recovery is not possible, the core position of the target point can be cleverly locked by using surrounding residual features, avoiding the complete omission of candidate areas by high-gloss.

[0047] Alternatively, when computational resources are sufficient, four branches can be run in parallel. The extraction results from each branch are clustered and merged to generate a comprehensive confidence score for each candidate region. This comprehensive confidence score can integrate the detection confidence score of the deep learning branch, the geometric fit of the MSER branch, the arc segment vote count of the sector branch, and the edge integrity of the anti-highlight branch. The final ROI candidate regions are generated based on the comprehensive confidence score. If the candidate region with the highest comprehensive confidence score is greater than a set threshold, then that group of candidate regions is output as a candidate ROI. If no candidate region with a comprehensive confidence score greater than the set threshold is input from any branch, a rollback is triggered, and parameters or the model are adjusted to retry.

[0048] Step S106: Based on the process formulation loading evaluation parameters, perform feature extraction on the candidate ROI region to obtain a state vector, and determine the regional coverage status of the candidate ROI region based on the state vector and the evaluation parameters.

[0049] Specifically, after identifying candidate ROIs, the process information of the current workpiece is obtained, primarily identifying the actual film material used in the current workpiece process (e.g., liquid photoresist LPI, dry film DFR, solder mask, etc.). Based on the type of film material, evaluation parameter thresholds matching the optical properties of the film material are loaded from a relevant parameter library, including: grayscale range division criteria (defining the grayscale range of dark, bright, and gray), contrast benchmark of normal target template (expected grayscale difference between the target and the background in the uncovered state), and upper limit of noise / texture intensity tolerance, etc. This eliminates the interference of inherent spectral reflectance differences of different film materials on state discrimination.

[0050] Then, the candidate ROIs are divided into regions and their features are extracted. This involves pixel-level analysis of each candidate ROI to determine its state vector, region parameters, and region coverage. The extraction process includes coarse segmentation masks for the target and background using pixel clustering. Then, corresponding parameters are calculated for the target mask, background mask, and their boundaries, including gray-level mean, standard deviation, histogram distribution, and local contrast. Based on these parameters, texture features such as contrast and uniformity of the gray-level co-occurrence matrix (GLCM) are extracted to distinguish between smooth ink coverage (whether the gray-level distribution of the target itself is uniform) and rough metal exposure or contamination. Furthermore, based on the grayscale range of dark, bright, and gray areas, the pixel ratio of dark, bright, and gray areas within the candidate ROI can be statistically analyzed to detect whether there is partial coverage (if the proportion of grayscale area increases significantly, and the standard deviation of the target area is significantly greater than the normal value of no coverage, and the histogram shows multiple peaks or tails, it indicates that there are a large number of intermediate grayscale transition pixels in the target area, corresponding to the half-coverage state of ink blotches and uneven thickness), no coverage (small proportion of grayscale area and excellent contrast), or full coverage (if the contrast between the target and the background is lower than the threshold, and most pixels in the target area fall into a single bright or dark range, and the overall standard deviation is small).

[0051] Step S108: Based on the area coverage status, select the corresponding image restoration scheme to obtain the restored ROI.

[0052] Specifically, based on the state vector of the candidate ROI in step S106, the corresponding processing scheme (repair scheme) is determined, and the repaired ROI is obtained.

[0053] For example, if there is no coverage, it means that the candidate ROI region is in good condition and complete, so no processing is required, or only slight noise reduction with edge preservation is needed.

[0054] If the area is partially covered (where the contrast is moderate compared to the uncovered area, and the grayscale ratio is high), a repair approach of local contrast stretching and edge enhancement can be adopted. For example, in the candidate ROI region, the local contrast of the transition area of ​​the target point edge can be stretched, and an anti-sharpening mask can be applied to enhance the edge gradient blurred by the semi-transparent ink. After the repair is completed, the corresponding ROI region is marked as partially covered and enhanced.

[0055] For regions with full or uniform coverage (where contrast is low and the target and background are almost blended), dynamic range stretching of the entire region and multi-scale matching preparation are required. By applying global histogram equalization or linear contrast stretching to candidate ROIs, the originally extremely narrow grayscale distribution is mapped to the full 0-255 range, forcibly creating a grayscale difference between the target and the background. An image pyramid (e.g., 3 layers, scaling factor 2) is pre-calculated, and the recommended starting matching level is marked as the top layer of the pyramid (coarse scale), because uniformly covered target points at fine scales have almost no texture. The stretched ROI region image patch and pyramid data are output, marked as full-coverage stretching.

[0056] In addition, for areas covered by ink with high noise (where the contrast is low and the texture is coarse), Gaussian filtering or nonlocal mean denoising can be performed to flatten the random texture caused by the coarse ink. Then, the starting matching level of the image pyramid is set to a higher level, and the corresponding search step size is increased to sacrifice a small amount of positioning accuracy in exchange for noise robustness. The denoised ROI region image block and pyramid data are output and marked as high noise compatible.

[0057] Additionally, there are localized highlight / high occlusion areas, which can be repaired through texture inpainting and edge closure restoration. For highlight areas, pixel regions within a candidate ROI whose grayscale exceeds the highlight threshold (e.g., 250) and are spatially connected can be marked as highlight areas. For highlight areas, a sample block-based image inpainting algorithm can be used. During restoration, the algorithm first searches for the optimal matching block outward from the boundary of the damaged area, utilizing surrounding intact target edge fragments and background textures to gradually fill the highlight area. For edge closure restoration, a circular constraint can be applied during the restoration process to guide the algorithm to generate imaginary edges that conform to the curvature trend of the target point. After restoration, the corresponding boundaries are merged to eliminate stitching marks and are marked as highlight restoration.

[0058] Step S110: Based on the repaired ROI and the target shape parameters, determine the coarse coordinates and generate a search line; determine the edge point cloud based on the coarse coordinates and the search line; and determine the center coordinates of the target point based on the edge point cloud fitting.

[0059] Specifically, in step S106, the relevant parameters of the target point include its geometric type. Based on the geometric type of the target point, a corresponding shape template image or gradient template is loaded from the template library, thereby quickly locking the approximate location of the target point, i.e., its coarse coordinates. Then, using the coarse coordinates as the origin, N corresponding search rays are generated according to the target point's geometric model. For example, for circles, radial rays are generated at equal angular intervals, and for rectangles, vertical search lines are generated along their edges. The length and step size of the search rays are adjusted according to the estimated size of the target point and the repair area (search lines closer to the repair area have shorter search ranges or reduced subsequent weights). The pixel intensity gradient is calculated along each ray, and the gradient peak position is found. Then, sub-pixel interpolation is used to accurately locate the edge point positions at the fractional pixel level. For each edge point, it is checked whether its gradient magnitude is significantly higher than the local noise level, and whether the gradient direction is consistent with the expected target point boundary normal (e.g., within the radial tolerance). Weak response points and directional anomalies are removed to generate a set of reliable edge points. Thus, the coarse localization is transformed into a high-precision edge coordinate point cloud.

[0060] Based on the established fitting model, initial fitting and residual calculation are performed on the set of reliable edge points. This involves running least-squares fitting on all reliable edge points to obtain initial model parameters (center, radius, etc.) and the fitting residual for each point. Then, a robust regression algorithm is used (each point is assigned a weight, which decreases as the residual increases; points exceeding a set threshold are considered outliers and their weights are reset to zero) for robust weight iteration. Weighted least-squares fitting is then performed again, iterating until convergence, and finally outputting the precise center coordinates and key dimensions of the target point, such as radius and roundness errors. In other words, robust statistical methods automatically identify and remove outlier edge points caused by factors such as contamination, ink edges, and highlight restoration artifacts, ensuring that the final center coordinates are not affected by these artifacts.

[0061] Additionally, if the target point is extremely incomplete, meaning that the ROI has insufficient inlier ratio during the target point feature detection and repair process in steps S106 or S108, a feature rollback mechanism is triggered. This mechanism calculates the relative center coordinates of concentric and parallel satellite circles (connected component extraction and algebraic distance clustering) to deduce the location of the main target center. Specifically, by using connected component extraction based on grayscale thresholds or MSER within the expanded ROI region, small, circular, or specifically shaped satellite patterns are found. These patterns typically have a fixed spatial relationship with the main target point (e.g., concentric rings, symmetrically distributed points). The center coordinates of the extracted satellite patterns are then calculated, and algebraic distance clustering is used to verify whether they conform to predefined concentric relationships or parallel layouts. Isolated interference points are excluded. Furthermore, if a group of satellite patterns satisfying geometric constraints is found (e.g., multiple small circles sharing a common circle, whose centers should be located on the same circle centered on the center of the larger circle), the theoretical location of the main target center is calculated by least-squares fitting of the common circle centers of the satellite circles, or by using known fixed offsets.

[0062] Step S112: Combining the repaired ROI and center coordinates, compare the target parameter standards of the process formulation, determine the target matching score based on the comparison results, and output the target points whose matching scores meet the accuracy requirements.

[0063] Specifically, the system receives the precise center coordinates and geometric dimensions of the target point from the preceding steps. It also acquires the standard geometric parameters of the target point (standard circle center position, standard radius, standard rectangle length and width, etc.) obtained from the process formulation. These two parameters are compared, and error quantization is calculated. Based on the error quantization results, the target point's matching score is determined. According to the matching score, the target point center coordinates that meet the accuracy requirements are output to the LDI control system for laser imaging alignment calculations.

[0064] This invention provides a multi-state target adaptive alignment vision method for LDI devices. The method involves: acquiring original images of the target; preprocessing the original images and obtaining their contrast data; adaptively adjusting the original images based on the contrast data; extracting candidate Regions of Interest (ROIs) from the original images using a perception strategy; extracting features from the candidate ROIs based on process recipe loading evaluation parameters to obtain state vectors; determining the regional coverage state of the candidate ROIs based on the state vectors and evaluation parameters; selecting a corresponding image restoration scheme based on the regional coverage state to obtain the restored ROI; determining coarse coordinates and generating search lines based on the restored ROI and target shape parameters; determining edge point clouds based on the coarse coordinates and search lines; determining the center coordinates of the target based on edge point cloud fitting; combining the restored ROI and center coordinates with the target parameter standards of the process recipe; determining the target matching score based on the comparison results; and outputting target points whose matching scores meet the accuracy requirements. By employing differentiated processing strategies for different target states (such as using specific model scaling domains and image enhancement strategies for different film materials), the impact of abnormal target states on positioning accuracy is avoided. Furthermore, it can adapt to various actual production scenarios such as uneven ink coverage, contrast variations, local defects, and high light reflection, greatly reducing misjudgments and rework caused by target anomalies. It also includes multiple adaptive strategy extraction methods, combined with lightweight neural networks to replace the cumbersome parameter adjustments that are highly susceptible to environmental influences, reducing reliance on human experience.

[0065] Please refer to the following. Figure 2 , Figure 2 A schematic diagram of a multi-state target adaptive alignment vision system for an LDI device, as provided in an embodiment of this specification, is shown. It should be noted that... Figure 2 The multi-state target adaptive alignment vision system shown for LDI devices is used to perform the functions described in this specification. Figure 1The methods shown in the embodiments are illustrated for ease of explanation, showing only the parts related to the embodiments of this specification. For specific technical details not disclosed, please refer to this specification. Figure 1 The example shown.

[0066] like Figure 2 As shown, the multi-state target adaptive alignment vision system for LDI devices may include at least:

[0067] The acquisition module S202 is used to acquire the original image of the target point, preprocess the original image, obtain the contrast data of the original image, and adaptively adjust the original image based on the contrast data.

[0068] The strategy module S204 is used to extract candidate ROI regions in the original image based on a perception strategy, wherein the perception strategy includes deep learning, full target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction that resists highlights and contamination.

[0069] Extraction module S206 is used to extract features from the candidate ROI region based on the process formula loading evaluation parameters, obtain a state vector, and determine the regional coverage state of the candidate ROI region based on the state vector and evaluation parameters.

[0070] The repair module S208 is used to select a corresponding image repair scheme based on the area coverage status to obtain the repaired ROI;

[0071] The search module S210 is used to determine coarse coordinates and generate a search line based on the repaired ROI and the target shape parameters, determine the edge point cloud based on the coarse coordinates and the search line, and determine the center coordinates of the target point based on the edge point cloud fitting.

[0072] The output module S212 is used to combine the repaired ROI and the center coordinates, compare the target parameter standard of the process formula, determine the target matching score based on the comparison result, and output the target points whose matching scores meet the accuracy requirements.

[0073] In another embodiment, a multi-state target adaptive alignment vision system for an LDI device further includes:

[0074] The acquisition module is used to acquire the global contrast measurement of the original image and, combined with the local image features of the ROI region as the adjustment basis, adaptively adjust the original image, including:

[0075] The first contrast module is used to directly output the original image when the global contrast is high and there are no low contrast defects in the local area.

[0076] The second contrast module is used to perform global histogram equalization on the original image when the global contrast is low.

[0077] The third contrast module is used to select an adaptive histogram balance that limits contrast when only the target area presents low contrast, or when both highlight and dark areas exist in the original image.

[0078] Those skilled in the art will clearly understand that the technical solutions of the embodiments in this specification can be implemented by means of software and / or hardware. In this specification, "unit" and "module" refer to software and / or hardware capable of independently performing or cooperating with other components to perform a specific function. The hardware may be, for example, a Field-Programmable Gate Array (FPGA), an Integrated Circuit (IC), etc.

[0079] Each processing unit and / or module in the embodiments of this specification can be implemented by an analog circuit that implements the functions described in the embodiments of this specification, or by software that executes the functions described in the embodiments of this specification.

[0080] See Figure 3 It shows a schematic diagram of the structure of an electronic device according to an embodiment of this specification, which can be used to implement... Figure 1 The method in the illustrated embodiment. (As shown) Figure 3 As shown, the electronic device 300 may include: at least one central processing unit 301, at least one network interface 304, user interface 303, memory 305, and at least one communication bus 302.

[0081] The communication bus 302 is used to enable communication between these components.

[0082] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0083] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0084] The processor 301 may include one or more processing cores. The processor 301 connects to various parts within the electronic device 300 using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 305, and by calling data stored in the memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0085] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory 305 may include a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage system located remotely from the aforementioned processor 301. Figure 3 As shown, the memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and program instructions.

[0086] exist Figure 3In the illustrated electronic device 300, the user interface 303 is mainly used to provide an input interface for the user and to acquire user input data; while the processor 301 can be used to call the image-based interactive application stored in the memory 305 and specifically perform the following operations:

[0087] The original image of the target point is acquired, the original image is preprocessed, and the contrast data of the original image is obtained. The original image is then adaptively adjusted based on the contrast data.

[0088] Candidate ROI regions are extracted from the original image based on a perceptual strategy, which includes deep learning, full target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction that resists highlights and contamination.

[0089] Based on the process formulation loading evaluation parameters, feature extraction is performed on the candidate ROI region to obtain a state vector, and the regional coverage status of the candidate ROI region is determined based on the state vector and the evaluation parameters.

[0090] Based on the coverage status of the region, the corresponding image restoration scheme is selected to obtain the restored ROI;

[0091] Based on the repaired ROI and target shape parameters, coarse coordinates are determined and a search line is generated. Based on the coarse coordinates and the search line, an edge point cloud is determined. Based on the edge point cloud fitting, the center coordinates of the target point are determined.

[0092] By combining the repaired ROI and center coordinates, and comparing the target parameter standards of the process formulation, the matching score of the target is determined based on the comparison results, and the target with the matching score that meets the accuracy requirements is output.

[0093] As an optional embodiment of this specification, the adaptive adjustment of the original image based on contrast data includes:

[0094] Adaptive adjustment of the original image based on contrast data includes:

[0095] Obtain the global contrast metric of the original image and combine it with the local image features of the ROI region as the adjustment basis to adaptively adjust the original image, including:

[0096] When the global contrast is high and there are no low-contrast defects in the local area, the original image is directly output.

[0097] When the global contrast is low, global histogram equalization is performed on the original image;

[0098] When only the target area exhibits low contrast, or when both highlight and dark areas exist in the original image, an adaptive histogram balance that limits contrast is selected.

[0099] As an optional embodiment of this specification, the extraction of candidate ROI regions from the original image based on a perception strategy includes:

[0100] The image scene of the original image is obtained, a corresponding perception strategy is selected based on the image scene, and candidate ROI regions are extracted based on the perception strategy.

[0101] Alternatively, multiple sensing strategies can be implemented in parallel. The extraction results of each sensing strategy are clustered and merged. Based on the comprehensive confidence score after clustering and merging, candidate ROI regions are selected.

[0102] As an optional embodiment of this specification, the method further includes:

[0103] Based on the process formulation loading evaluation parameters, feature extraction is performed on the candidate ROI region to obtain a state vector;

[0104] The membrane material type is obtained based on the process formulation, and an evaluation parameter threshold corresponding to the membrane material type is loaded. The evaluation parameter threshold includes a grayscale range division standard and a contrast benchmark.

[0105] Pixel analysis is performed on the candidate ROI region to obtain region parameters, and the region coverage status of the candidate ROI region is determined by combining the grayscale interval division standard. The ink-covered region and the highlight / occlusion region are distinguished by combining the contrast benchmark, and the corresponding integrity mark is generated.

[0106] As an optional embodiment of this specification, the step of selecting a corresponding image inpainting scheme based on the region coverage state to obtain the repaired ROI includes:

[0107] Based on the region coverage status and integrity markers in the state vector, the corresponding image restoration scheme is triggered by looking up the table: texture repair is performed on the highlight / occluded areas, noise reduction is performed on the high-noise areas and the matching pyramid level is adjusted, and robust fitting parameters are passed to the damaged areas.

[0108] As an optional embodiment of this specification, the method further includes:

[0109] If the proportion of interior points is insufficient when repairing the ROI, the process will revert to the main target position in the satellite circular connected domain clustering inference, and the main target position will be used as the center coordinates.

[0110] This specification also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0111] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this specification is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this specification. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this specification.

[0112] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0113] In the embodiments provided in this specification, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between systems or units may be electrical or other forms.

[0114] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0115] Furthermore, the functional units in the various embodiments of this specification can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0116] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this specification, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this specification. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0117] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0118] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A multi-state target adaptive alignment vision method for LDI devices, characterized in that, The method includes: The original image of the target point is acquired, the original image is preprocessed, and the contrast data of the original image is obtained. The original image is then adaptively adjusted based on the contrast data. Candidate ROI regions are extracted from the original image based on a perceptual strategy, which includes deep learning, full target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction that resists highlights and contamination. Based on the process formulation loading evaluation parameters, feature extraction is performed on the candidate ROI region to obtain a state vector, and the regional coverage status of the candidate ROI region is determined based on the state vector and the evaluation parameters. Based on the coverage status of the region, the corresponding image restoration scheme is selected to obtain the restored ROI; Based on the repaired ROI and target shape parameters, coarse coordinates are determined and a search line is generated. Based on the coarse coordinates and the search line, an edge point cloud is determined. Based on the edge point cloud fitting, the center coordinates of the target point are determined. By combining the repaired ROI and center coordinates, and comparing the target parameter standards of the process formulation, the matching score of the target is determined based on the comparison results, and the target with the matching score that meets the accuracy requirements is output.

2. The multi-state target adaptive alignment vision method for LDI devices according to claim 1, characterized in that, The adaptive adjustment of the original image based on contrast data includes: Adaptive adjustment of the original image based on contrast data includes: Obtain the global contrast metric of the original image and combine it with the local image features of the ROI region as the adjustment basis to adaptively adjust the original image, including: When the global contrast is high and there are no low-contrast defects in the local area, the original image is directly output. When the global contrast is low, global histogram equalization is performed on the original image; When only the target area exhibits low contrast, or when both highlight and dark areas exist in the original image, an adaptive histogram balance that limits contrast is selected. The selection of adaptive histogram balance that limits contrast includes: dividing the original image into multiple local blocks, independently calculating and equalizing the histogram in each block, clipping and limiting the histogram of each block, and uniformly distributing the portion exceeding the limit to all gray levels.

3. The multi-state target adaptive alignment vision method for LDI devices according to claim 1, characterized in that, The extraction of candidate ROI regions from the original image based on a perceptual strategy includes: The image scene of the original image is obtained, a corresponding perception strategy is selected based on the image scene, and candidate ROI regions are extracted based on the perception strategy. Alternatively, multiple sensing strategies can be implemented in parallel. The extraction results of each sensing strategy are clustered and merged. Based on the comprehensive confidence score after clustering and merging, candidate ROI regions are selected.

4. The multi-state target adaptive alignment vision method for LDI devices according to claim 1, characterized in that, The method further includes: Based on the process formulation loading evaluation parameters, feature extraction is performed on the candidate ROI region to obtain a state vector; The membrane material type is obtained based on the process formulation, and an evaluation parameter threshold corresponding to the membrane material type is loaded. The evaluation parameter threshold includes a grayscale range division standard and a contrast benchmark. Pixel analysis is performed on the candidate ROI region to obtain region parameters, and the region coverage status of the candidate ROI region is determined by combining the grayscale interval division standard. The ink-covered region and the highlight / occlusion region are distinguished by combining the contrast benchmark, and the corresponding integrity mark is generated.

5. The multi-state target adaptive alignment vision method for LDI devices according to claim 4, characterized in that, The step of selecting a corresponding image inpainting scheme based on the region coverage status to obtain the repaired ROI includes: Based on the region coverage status and integrity markers in the state vector, the corresponding image restoration scheme is triggered by looking up the table: texture repair is performed on the highlight / occluded areas, noise reduction is performed on the high-noise areas and the matching pyramid level is adjusted, and robust fitting parameters are passed to the damaged areas.

6. The multi-state target adaptive alignment vision method for LDI devices according to claim 1, characterized in that, The method further includes: If the proportion of interior points is insufficient when repairing the ROI, the process will revert to the main target position in the satellite circular connected domain clustering inference, and the main target position will be used as the center coordinates.

7. A multi-state target adaptive alignment vision system for LDI devices, characterized in that, The system includes: The acquisition module is used to acquire the original image of the target point, preprocess the original image, obtain the contrast data of the original image, and adaptively adjust the original image based on the contrast data. The strategy module is used to extract candidate ROI regions from the original image based on a perceptual strategy, which includes deep learning, full target extraction based on extreme regions and morphology, fan-shaped extraction based on local contours, and specific extraction that resists highlights and contamination. The extraction module is used to extract features from the candidate ROI region based on the process formula and evaluation parameters, obtain a state vector, and determine the regional coverage status of the candidate ROI region based on the state vector and evaluation parameters. The repair module is used to select the corresponding image repair scheme based on the coverage status of the region to obtain the repaired ROI; The search module is used to determine coarse coordinates and generate a search line based on the repaired ROI and target shape parameters, determine the edge point cloud based on the coarse coordinates and the search line, and determine the center coordinates of the target point based on the edge point cloud fitting. The output module is used to combine the repaired ROI and center coordinates, compare the target parameter standards of the process formulation, determine the target matching score based on the comparison results, and output the target points whose matching scores meet the accuracy requirements.

8. The multi-state target adaptive alignment vision system for LDI devices according to claim 7, characterized in that, The system also includes: The acquisition module is used to acquire the global contrast measurement of the original image and, combined with the local image features of the ROI region as the adjustment basis, adaptively adjust the original image, including: The first contrast module is used to directly output the original image when the global contrast is high and there are no low contrast defects in the local area. The second contrast module is used to perform global histogram equalization on the original image when the global contrast is low. The third contrast module is used to select an adaptive histogram balance that limits contrast when only the target area presents low contrast, or when there are both highlight areas and dark areas in the original image. The selection of the adaptive histogram balance that limits contrast includes: dividing the original image into multiple local blocks, calculating the histogram independently in each block and equalizing it, clipping and limiting the histogram of each block, and uniformly distributing the portion exceeding the limit to all gray levels.

9. An electronic device, comprising a processor and a memory; The processor is connected to the memory; The memory is used to store executable program code; The processor runs a program corresponding to the executable program code stored in the memory to perform the method as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of claims 1-6.

Images