A method and device for segmenting and geometrically constructing a gold wire microscopic image

Abstract

This application discloses a method and apparatus for segmentation and geometric construction of gold wire microscopic images, relating to the fields of machine vision and semiconductor packaging inspection technology. The method includes: performing image enhancement and edge-preserving denoising processing on the gold wire microscopic image, binarizing and optimizing the enhanced image, obtaining linear geometric feature extraction based on the binary results to obtain a candidate line segment set, and performing spatial analysis on the candidate reference region information to obtain a reference point set; performing geometric constraint matching and association processing on the candidate line segment set according to the reference point set to obtain the target line segment and associated position corresponding to each reference point in the reference point set; and determining and outputting a structured result set based on the geometric relationship between the associated positions corresponding to each target line segment. This application does not rely on model training conditions; it only performs instance-level construction on the gold wire microscopic image to output gold wire instance results with geometric semantics and start / end information.

Description

Technical Field

[0001] This application relates to the fields of machine vision and semiconductor packaging inspection technology, and in particular to a method and apparatus for segmentation and geometry construction of gold wire micrographs. Background Technology

[0002] In semiconductor packaging manufacturing, the internal pads and external pins of a chip are typically electrically connected via bonding wires. The integrity and spatial morphology of these gold wires directly affect the chip's electrical performance and long-term reliability. Therefore, automated optical inspection of the gold wires after packaging is a crucial step in semiconductor manufacturing quality control. With the continuous increase in chip integration and packaging density, the diameter of bonding wires is gradually decreasing, their arrangement is becoming increasingly dense, and they exhibit highly reflective and high-contrast metallic properties under microscopic imaging conditions. This often results in gold wires appearing in images with significant uneven illumination and localized saturated background texture interference, posing a significant challenge to image-based automated inspection.

[0003] In existing technologies, image analysis methods for gold wire can be mainly divided into two categories. One category is deep learning-based segmentation methods, which typically construct convolutional neural networks to perform semantic or instance segmentation of the gold wire region. This type of method can achieve high detection accuracy under specific pattern conditions, but it relies on a large amount of manually labeled data, resulting in high training costs. It also has poor adaptability to different chip structures, imaging conditions, and process variations, and the model inference process lacks interpretability, which is not conducive to stable deployment in industrial settings. The other category is based on traditional image processing and geometric analysis methods, such as using threshold segmentation techniques to extract gold wire contours or line segment features. However, this type of method is easily affected by noise in complex background conditions, and when the gold wire is broken, intersecting, or partially missing, it is often difficult to reliably associate multiple local features as the same physical instance. Especially in instance-level analysis scenarios, simply obtaining the pixel-level foreground region or undirected line segment set of the gold wire is usually insufficient to support subsequent quality judgment requirements. For example, applications such as circuit detection, length measurement, and direction offset analysis all require the ability to clearly distinguish different gold wire instances and accurately determine the start and end positions and spatial orientation of each instance. However, existing line detection-based techniques often output non-directional line segment results, lacking effective instance association and direction discrimination mechanisms. This can easily lead to instance confusion or directional ambiguity when gold lines are densely packed or partially broken. Furthermore, in real-world industrial images, gold lines are often not continuous, complete line segments, but rather elongated structures composed of multiple irregular segments. Limited by imaging noise, illumination variations, and detection parameter settings, there is often a discrepancy between the extracted line segment endpoints and the physical connection points of the gold lines, making traditional association methods based on point-to-line distance or global optimal matching difficult to operate stably.

[0004] Therefore, there is an urgent need for a technical solution that can stably extract gold wire features in complex noise backgrounds and achieve instance-level segmentation and structured description of bonding gold wires without relying on large-scale training data, so as to meet the comprehensive requirements of accuracy, robustness and engineering deployability in semiconductor packaging inspection. Summary of the Invention

[0005] The purpose of this application is to address at least one of the aforementioned technical deficiencies.

[0006] On one hand, embodiments of this application provide a method for segmentation and geometric construction of gold wire micrographs, the method comprising: Acquire the gold wire microscopic image and candidate reference region information of the chip to be processed. The candidate reference region information is used to indicate the location of the starting end of the gold wire instance. The candidate reference region and the gold wire microscopic image are in the same image coordinate system. Image enhancement and edge-preserving denoising were performed on the gold wire microscopic image to obtain an enhanced image. The enhanced image was then binarized and structurally optimized to obtain a binary result. Linear geometric feature extraction is performed based on the binary results to obtain a set of candidate line segments, and spatial analysis is performed on the candidate reference area information to obtain a set of reference points. Based on the set of reference points, the candidate line segment set is subjected to geometric constraint matching and association processing to obtain the target line segment corresponding to each reference point in the set of reference points, and the associated position of each target line segment is determined. Based on the geometric relationships between the associated positions of each target line segment, a structured result set is determined and output. The structured result set includes the start and end positions of each gold line instance.

[0007] Optionally, image enhancement and edge-preserving denoising are performed on the gold wire micrograph to obtain an enhanced image. The enhanced image is then binarized and structurally optimized to obtain a binary result, including: Edge-preserving smoothing was applied to the gold wire micrograph to obtain an enhanced image; The enhanced image is processed using an adaptive thresholding method based on local region statistical features to obtain the binary result to be optimized. The binary result to be optimized is then combined with morphological optimization to obtain the final binary result.

[0008] Optionally, the linear geometric feature extraction process can be one of Hough line segment detection or skeleton-based linear segment construction. Based on the binary results, linear geometric feature extraction is performed to obtain a candidate line segment set, including: Linear geometric feature extraction is performed based on the binary results to obtain each initial candidate line segment; Each initial candidate line segment is filtered based on preset filtering conditions, and a candidate line segment set is formed based on the filtered initial candidate line segments; wherein, the preset filtering conditions include at least one of the minimum line segment length threshold and the maximum break gap threshold.

[0009] Optionally, a candidate line segment set is formed based on the filtered initial candidate line segments, including: For each filtered initial candidate line segment, the direction angle of the filtered initial candidate line segment is determined based on the endpoint coordinates of the filtered initial candidate line segment; Obtain a preset direction angle range, and perform direction filtering on each filtered initial candidate line segment based on the direction angle range and the direction angle of each filtered initial candidate line segment to obtain a set of candidate line segments.

[0010] Optionally, the candidate reference region information can be represented in any of the following forms: binary mask, region marker, or pixel coordinates. The candidate reference region information is obtained in the following ways: When the candidate reference region information is represented in either a binary mask or pixel coordinates, the binary mask image corresponding to the gold wire microscopic image is obtained, and the coordinate set corresponding to the non-zero pixels in the binary mask image is extracted as the foreground pixel set to obtain the candidate reference region information. When the candidate reference region information is represented in the form of region labeling, the enhanced image corresponding to the gold line microscopic image is obtained, and the local bright region is extracted based on the pixel brightness or local contrast in the enhanced image to generate the initial candidate binary mask. Connected component labeling is performed on the initial candidate binary mask to obtain at least one connected component, and the gray-level statistical features and morphological features of each connected component are calculated. Each connected component is filtered based on its gray-scale statistical and morphological characteristics, and the filtered connected components are used as candidate reference region information.

[0011] Optionally, spatial analysis is performed on the candidate reference region information to obtain a set of reference points, including: When the candidate reference region information is represented in either a binary mask or pixel coordinates, spatial aggregation processing is performed on the foreground pixels corresponding to the candidate reference region information to form at least one aggregation cluster; for each aggregation cluster, the geometric center of the pixel coordinates within the aggregation cluster is calculated, and the geometric center or the pixel with the smallest distance from the geometric center is taken as the reference point of the aggregation cluster; a reference point set is formed based on the reference point of each aggregation cluster. When candidate reference region information is represented in the form of region labels, connected component labeling is performed on the binary mask corresponding to the candidate reference region to obtain at least one region component, and each region component corresponds to a set of pixels; spatial merging is performed on the set of pixels corresponding to each region component to form at least one merged region; for each merged region, the centroid position of the merged region is determined, and the centroid position or the pixel with the smallest distance from the centroid position is taken as the reference point of the merged region; a set of reference points is formed based on the reference points of each merged region.

[0012] Optionally, geometric constraint matching and association processing are performed on the candidate line segment set based on the reference point set to obtain the target line segment corresponding to each reference point in the reference point set, including: For any reference point in the set of reference points and any candidate line segment in the set of candidate line segments, construct the projection relationship from the reference point to the candidate line segment to obtain the projection parameters and the initial projection position; when the initial projection position is outside the spatial range of the candidate line segment, constrain and correct the initial projection position according to the projection parameters to obtain the corrected projection position. Using the squared distance from the reference point to the corrected projection position as the matching cost, we traverse all candidate segments in the candidate segment set and select the candidate segment with the smallest matching cost as the target segment corresponding to the reference point.

[0013] Optionally, determine the associated position corresponding to each target line segment, including: For each target line segment, the corresponding corrected projection position on the target line segment is determined as the associated position of the target line segment.

[0014] Optionally, based on the geometric relationship between each target line segment and its corresponding associated location, a structured result set is determined and output, including: For each target line segment, the reference point corresponding to the target line segment is used as the starting position of the gold line instance corresponding to the target line segment; Compare the squared distances from the associated position of the target line segment to the two endpoints of the target line segment, and determine the endpoint with the largest squared distance as the termination position of the gold line instance corresponding to the target line segment. The two endpoints of the target line segment are the start endpoint and the end endpoint of the target line segment. A structured result set is constructed based on the start and end positions of the gold line instance corresponding to each target line segment.

[0015] On the other hand, embodiments of this application provide a device for segmenting and geometrically constructing gold wire micrographs, characterized in that it includes: The image acquisition module is used to acquire the gold wire microscopic image of the chip to be processed and the candidate reference region information. The candidate reference region information is used to indicate the location information of the starting end of the gold wire instance. The candidate reference region and the gold wire microscopic image are in the same image coordinate system. The preprocessing module is used to perform image enhancement and edge-preserving denoising on the gold wire microscopic image to obtain an enhanced image, and then perform binarization and structure optimization on the enhanced image to obtain a binary result; The reference point generation module is used to perform linear geometric feature extraction based on the binary results to obtain a set of candidate line segments, and to perform spatial analysis on the candidate reference area information to obtain a set of reference points. The instance construction module is used to perform geometric constraint matching and association processing on the candidate line segment set based on the reference point set, to obtain the target line segment corresponding to each reference point in the reference point set, and to determine the associated position of each target line segment; based on the geometric relationship between each target line segment and its corresponding associated position, a structured result set is determined. The results output module is used to output a structured result set, which includes the start and end positions of each gold line instance.

[0016] Furthermore, embodiments of this application provide an electronic device, including a processor and a memory: The memory is configured to store machine-readable instructions that, when executed by the processor, cause the processor to perform any of the methods in a gold wire micrograph segmentation and geometry construction approach.

[0017] The beneficial effects of the technical solutions provided in this application include at least the following: In this application, a set of reference points and a set of candidate line segments can be calculated based on the acquired microscopic image of gold wire and candidate reference region information. Then, geometric constraint matching and association processing are performed on the candidate line segment set based on the set of reference points to obtain a structured result set including the start and end positions of each gold wire instance. In this process, there is no need to rely on the conditions of model training. The gold wire microscopic image is constructed at the instance level based solely on image processing and geometric constraints. Gold wire instance results with geometric semantics and start and end information can be output. Furthermore, it is no longer limited by the influence of imaging noise, illumination changes, and detection parameters. Even in cases of gold wire breakage, intersection, or complex backgrounds, gold wire instance results can still be output. This can meet the application requirements for instance-level analysis in semiconductor packaging inspection scenarios and is more suitable for industrial applications such as semiconductor packaging inspection.

[0018] Furthermore, after obtaining the microscopic image of the gold wire, this application can first perform image enhancement and edge-preserving denoising processing on the gold wire microscopic image. At this time, while suppressing background noise and reflection interference, the slender edge structure of the gold wire can be preserved as much as possible. Further, in order to reduce the impact of the background structure on the subsequent construction of gold wire instances, each initial candidate line segment can be screened according to preset filtering conditions. At this time, excessively short line segments, excessively fragmented line segments, and noisy line segments can be prevented from entering the candidate line segment set. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a method for segmenting and geometrically constructing a gold wire micrograph provided in this application embodiment; Figure 2 This application provides schematic diagrams of input data and gold wire examples for embodiments of the present application. Figure 3 A schematic diagram of the geometric matching principle provided for an embodiment of this application; Figure 4 This is a schematic diagram of a structured result set provided in the embodiments of this application; Figure 5 A flowchart illustrating another method for segmenting and geometrically constructing gold wire micrographs provided in this application embodiment; Figure 6 A schematic diagram of the system structure provided in the embodiments of this application; Figure 7 A schematic diagram of a device for segmenting and geometric construction of gold wire micrographs provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0021] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting the invention.

[0022] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0024] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0025] Specifically, such as Figure 1 As shown, the method may include: Step S101: Obtain the gold wire microscopic image of the chip to be processed and the candidate reference region information. The candidate reference region information is used to indicate the location information of the starting end of the gold wire instance. The candidate reference region and the gold wire microscopic image are in the same image coordinate system.

[0026] Optionally, the gold wire micrograph can be the original grayscale image of the chip to be processed, or a grayscale image obtained through color space conversion. Candidate reference region information is used to indicate the local reference position of the gold wire instance in the physical structure. In this application, the local reference position specifically refers to the local region at the starting end where the gold wire connects to the pad / solder joint, i.e., the location information of the starting end of the gold wire instance, which can be used as an anchor point for subsequent geometric matching and instance construction. The gold wire micrograph can be as follows: Figure 2 As shown in (a) above, the corresponding candidate reference region information is as follows: Figure 2 The red pixel in (b) is shown.

[0027] Optionally, the candidate reference region information is located in the same image coordinate system as the gold line micrograph. The source of the candidate reference region information can be selected according to the actual situation, and this application embodiment limits this. For example, the candidate reference region information can be obtained by manual annotation, rule extraction, model prediction, or device-side output.

[0028] In optional embodiments of this application, the candidate reference region information is represented in any one of the following forms: binary mask, region marker, or pixel coordinates. The candidate reference region information is obtained in the following ways: When the candidate reference region information is represented in either a binary mask or pixel coordinates, the binary mask image corresponding to the gold wire microscopic image is obtained, and the coordinate set corresponding to the non-zero pixels in the binary mask image is extracted as the foreground pixel set to obtain the candidate reference region information. When the candidate reference region information is represented in the form of region labeling, the enhanced image corresponding to the gold line microscopic image is obtained, and the local bright region is extracted and processed according to the pixel brightness or local contrast in the enhanced image to generate the initial candidate binary mask. Connected component labeling is performed on the initial candidate binary mask to obtain at least one connected component, and the gray-level statistical features and morphological features of each connected component are calculated. Each connected component is filtered based on its gray-scale statistical and morphological characteristics, and the filtered connected components are used as candidate reference region information.

[0029] Optionally, candidate reference region information can be represented in the form of a binary mask, region markers, or a set of pixel coordinates. The generation method of candidate reference region information differs depending on the representation. For example, when candidate reference region information is represented by either a binary mask or pixel coordinates, a binary mask image corresponding to the gold wire micrograph can be obtained. Then, the coordinate set corresponding to the non-zero pixels in the binary mask image can be extracted as a foreground pixel set, and the coordinates of the foreground pixel set can be used as candidate reference region information for subsequent reference unit formation and matching processing.

[0030] Correspondingly, when candidate reference region information is represented in the form of region labels, the candidate reference region information can be automatically generated from the gold wire microscopic image. Specifically, after acquiring the gold wire microscopic image, contrast enhancement and noise suppression processing can be performed on the gold wire microscopic image, and the brightness distribution and morphological structure of the local area can be analyzed to identify candidate local bright areas and generate an initial candidate binary mask. Then, the initial candidate binary mask is filtered by combining gray-level statistical features and morphological features to obtain the candidate reference region information.

[0031] Optionally, as an implementation, the above analysis may specifically include the following steps: First, on the gold wire micrograph after contrast enhancement and noise suppression (i.e., the enhanced image), locally bright areas are extracted based on pixel brightness or local contrast in the enhanced image to generate an initial candidate binary mask. Then, connected component labeling is performed on the initial candidate mask to obtain at least one connected component, and the gray-level statistical features and morphological features of each connected component are calculated. The gray-level statistical features characterize the brightness difference between the connected component and the surrounding background, while the morphological features characterize the geometric appearance of the connected component.

[0032] Furthermore, each connected component is screened based on its grayscale statistical and morphological characteristics to remove components with insignificant brightness features or geometric shapes that do not conform to the characteristics of solder joints / end regions. Components that simultaneously satisfy both brightness significance and morphological constraints are retained as candidate reference regions. The morphological constraints can be set based on the differences between the main body of the gold wire instance, which typically exhibits a slender linear structure, and solder joints or end regions, which tend to have a blocky or near-circular structure. For example, the area of ​​the restricted region may be within a preset range, and the aspect ratio of the circumscribed rectangle may be lower than a preset threshold.

[0033] Step S102: Image enhancement and edge-preserving denoising processing are performed on the gold wire microscopic image to obtain an enhanced image. The enhanced image is then binarized and structurally optimized to obtain a binary result.

[0034] In optional embodiments of this application, image enhancement and edge-preserving denoising processing are performed on the gold wire micrograph to obtain an enhanced image, and binarization and structure optimization processing are performed on the enhanced image to obtain a binary result, including: Edge-preserving smoothing was applied to the gold wire micrograph to obtain an enhanced image; The enhanced image is processed using an adaptive thresholding method based on local region statistical features to obtain the binary result to be optimized. The binary result to be optimized is then combined with morphological optimization to obtain the final binary result.

[0035] Optionally, after acquiring the gold wire microscopic image, image enhancement and edge-preserving denoising processing can be performed on the gold wire microscopic image to obtain an enhanced image. This enhancement and edge-preserving denoising processing can be achieved by edge-preserving smoothing, which can suppress background noise and reflection interference while preserving the slender edge structure of the gold wire instance as much as possible.

[0036] Furthermore, the enhanced image can be binarized and structurally optimized to obtain a binary result (hereinafter referred to as binary result B) representing the morphology of the gold wire body. The binarization process can be implemented using an adaptive thresholding method based on local region statistical features, and the structural optimization process can be combined with morphological processing to reduce isolated noise and enhance the connectivity of the gold wire body structure. Optionally, the specific algorithm and parameter settings involved in obtaining the binary result can be set according to actual conditions, and this application does not limit them.

[0037] Step S103: Based on the binary results, perform linear geometric feature extraction to obtain a set of candidate line segments, and perform spatial analysis on the candidate reference area information to obtain a set of reference points.

[0038] Optionally, after obtaining the binary results, a set of candidate line segments for determining the gold line example can be obtained based on the linear geometric features of the binary results. Then, spatial analysis can be performed on the obtained candidate reference area information to obtain a set of reference points.

[0039] In optional embodiments of this application, the linear geometric feature extraction process is one of Hough line segment detection and skeleton-tracking-based linear segment construction. Based on the binary results, linear geometric feature extraction is performed to obtain a candidate line segment set, including: Linear geometric feature extraction is performed based on the binary results to obtain each initial candidate line segment; Each initial candidate line segment is filtered based on preset filtering conditions, and a candidate line segment set is formed based on the filtered initial candidate line segments; wherein, the preset filtering conditions include at least one of the minimum line segment length threshold and the maximum break gap threshold.

[0040] Optionally, linear geometric feature extraction can be either Hough line segment detection or skeleton-tracking-based linear segment construction. When Hough line segment detection is used for linear geometric feature extraction, the initial candidate line segments can be directly detected; while when skeleton-tracking-based linear segment construction is used, the binary results can be processed. Two-dimensional image thinning or skeletonization operations are performed on the foreground region to obtain a skeleton pixel set with a width of one pixel. Then, the endpoint pixels or bifurcation pixels of the skeleton pixels are selected as the starting points in the skeleton pixel set, and adjacent skeleton pixels are tracked step by step along the local direction to organize the skeleton pixel sequence with consistent direction and spatial continuity into linear segments. Then, the pixel coordinates of each linear segment are fitted with a straight line or piecewise to obtain a set of candidate line segments.

[0041] Furthermore, in order to reduce the impact of the background structure on the construction of subsequent gold line instances, each initial candidate line segment can be screened according to preset filtering conditions. These preset filtering conditions include at least one of the minimum line segment length threshold and the maximum break gap threshold. This can suppress excessively short line segments, excessively fragmented line segments, and noisy line segments from entering the candidate line segment set.

[0042] In an optional embodiment of this application, a candidate line segment set is formed based on the filtered initial candidate line segments, including: For each filtered initial candidate line segment, the direction angle of the filtered initial candidate line segment is determined based on the endpoint coordinates of the filtered initial candidate line segment; Obtain a preset direction angle range, and perform direction filtering on each filtered initial candidate line segment based on the direction angle range and the direction angle of each filtered initial candidate line segment to obtain a set of candidate line segments.

[0043] Optionally, to further reduce the impact of the background structure on subsequent instance construction, this application can perform direction filtering on each of the initial candidate line segments obtained after filtering when forming a candidate line segment set based on the filtered initial candidate line segments. Specifically, for each of the filtered initial candidate line segments... Let the coordinates of its two endpoints be respectively and At this point, the direction angle of the initial candidate line segment can be calculated based on the coordinates of the two endpoints. Then Normalization to (when At that time, then order )Inside.

[0044] Furthermore, a preset directional angle range is defined. As the range of interference directions, direction filtering is performed, when the direction angle of the initial candidate line segment... When a line segment falls within this directional angle range, the initial candidate line segment is identified as an interference line segment that does not conform to the directional constraints and is discarded; otherwise, the initial candidate line segment is retained and added to the candidate line segment set. Specifically, when the background structure in the image mainly presents a near-circular shape... When the direction of distribution is , one can take As a preset direction angle range.

[0045] At the same time, the candidate reference region information is spatially merged to obtain a reference point set including one or more reference points. This reference point is used to represent the starting position of the gold line instance. At this point, the regional information is abstracted into point-level anchor points, which reduces the complexity of subsequent matching.

[0046] In an optional embodiment of this application, spatial analysis is performed on the candidate reference region information to obtain a set of reference points, including: When the candidate reference region information is represented in either a binary mask or pixel coordinates, spatial aggregation processing is performed on the foreground pixels corresponding to the candidate reference region information to form at least one aggregation cluster; for each aggregation cluster, the geometric center of the pixel coordinates within the aggregation cluster is calculated, and the geometric center or the pixel with the smallest distance from the geometric center is taken as the reference point of the aggregation cluster; a reference point set is formed based on the reference point of each aggregation cluster. When candidate reference region information is represented in the form of region labels, connected component labeling is performed on the binary mask corresponding to the candidate reference region to obtain at least one region component, and each region component corresponds to a set of pixels; spatial merging is performed on the set of pixels corresponding to each region component to form at least one merged region; for each merged region, the centroid position of the merged region is determined, and the centroid position or the pixel with the smallest distance from the centroid position is taken as the reference point of the merged region; a set of reference points is formed based on the reference points of each merged region.

[0047] Optionally, the method for spatial analysis of candidate reference region information differs depending on the representation format. Specifically, when the candidate reference region is given in the form of a binary mask or a set of pixel coordinates, the set of foreground pixel coordinates corresponding to the candidate reference region can be denoted as... ,in Furthermore, since the same physical solder joint / starting area usually corresponds to a group of adjacent or nearby pixels, the foreground pixel coordinate set can be used in this case. Spatial aggregation is performed to form at least one cluster, at which point pixels that are close to each other can be grouped together. One implementation method is density-based clustering. Perform aggregation and set the neighborhood radius. With minimum sample size , the distance does not exceed And the number of sample points in the neighborhood reaches The pixels are grouped into the same cluster.

[0048] Furthermore, below each cluster, the geometric center of the pixel coordinates within that cluster can be calculated using the following formula, and the geometric center, or the pixel with the smallest distance from the geometric center, can be used as the reference point for the cluster: in, The geometric center of the pixel coordinates within the cluster is used as a reference point. For aggregated clusters, These are the coordinates of the foreground pixels.

[0049] Accordingly, after determining the reference point for each cluster, the reference points of each cluster form a reference point set. Furthermore, isolated pixels that fail to belong to any cluster can be considered noise and ignored.

[0050] Optionally, when candidate reference region information is represented in the form of region labels, instead of using clustering algorithms to aggregate the pixel sets of candidate reference regions, a merging method based on connected component components and spatial proximity is used to obtain reference points. Specifically, the binary mask corresponding to the candidate reference region can be labeled with connected components, for example, using 4-neighborhood or 8-neighborhood connectivity, to obtain at least one mutually unconnected region component, and each region component corresponds to a pixel set.

[0051] Furthermore, spatial merging is performed based on the pixel set corresponding to each region component. When multiple region components satisfy adjacency or distance constraints in space, they are merged into the same merged region. This adjacency or distance constraint can include any one of the following: the minimum squared distance between region boundary pixels is less than a preset threshold, the squared distance between region centroids is less than a preset threshold, and the minimum spacing between circumscribed rectangles is less than a preset threshold. Correspondingly, for each merged region, the centroid position of its pixel set can be used as a reference point. Alternatively, the pixel closest to the centroid can be selected as the reference point to obtain the reference point set. .

[0052] Step S104: Perform geometric constraint matching and association processing on the candidate line segment set based on the reference point set to obtain the target line segment corresponding to each reference point in the reference point set, and determine the associated position of each target line segment.

[0053] Optionally, after constructing the candidate line segment set and the reference point set, geometric matching and association processing can be performed on the candidate line segment set based on each reference point to obtain the target line segment corresponding to each reference point, and determine the association position corresponding to each target line segment. This association position is used to represent the unique association position after the reference point is matched with the target line segment.

[0054] In optional embodiments of this application, geometric constraint matching and association processing are performed on the candidate line segment set based on the reference point set to obtain the target line segment corresponding to each reference point in the reference point set, including: For any reference point in the set of reference points and any candidate line segment in the set of candidate line segments, construct the projection relationship from the reference point to the candidate line segment to obtain the projection parameters and the initial projection position; when the initial projection position is outside the spatial range of the candidate line segment, constrain and correct the initial projection position according to the projection parameters to obtain the corrected projection position. Using the squared distance from the reference point to the corrected projection position as the matching cost, we traverse all candidate segments in the candidate segment set and select the candidate segment with the smallest matching cost as the target segment corresponding to the reference point.

[0055] Optionally, as shown in Figure 3, after completing the candidate line segment set... With reference point set After construction, for any reference point With any candidate line segment Construct the projection relationship from the reference point to the line segment to obtain the projection parameters. and initial projection position The projection parameters can be expressed as: in, This represents the vector length, i.e., the magnitude. Represents the square of the length. and Candidate line segments The two endpoints, As a reference point, These are the projection parameters.

[0056] Furthermore, it can be based on projection parameters. Get reference point Online segment Initial projection position on : Among them, when At that time, the initial projection position Located at the endpoint of the line segment Within the defined line segment range; when or At that time, the initial projection position If the projection falls on the extension of the line connecting the endpoints, the projection position can be considered to be outside the spatial range of the line segment, which is a case of projection exceeding the boundary.

[0057] Optionally, for cases where the projection exceeds the boundary, the initial projection position is constrained and corrected according to the projection parameters to be within the effective range of the line segment, thus obtaining the corrected projection position. One approach is to apply the following formula to the projection parameters. Truncate the projection to obtain the truncated projection parameters. : Furthermore, The corrected projection position is obtained. At this point, the out-of-bounds projection will be constrained to the nearest line segment endpoint.

[0058] Correspondingly, an equivalent implementation for constraining and correcting the initial projection position is to directly compare the reference point. To candidate line segment The squared distance between the two endpoints is obtained respectively. and Furthermore, the endpoint with the smaller distance is used as the corrected projection position. Or in candidate line segments From the corresponding set of linear segment pixels, select the reference point. The effective pixel with the smallest square distance is used as the corrected projection position. It is understood that the above projection and constraint correction methods are merely illustrative examples, and this invention does not limit their specific mathematical forms.

[0059] Furthermore, at the corrected projection position Then, calculate the reference point using the following formula. To the corrected projection position square distance As a matching cost: Then, based on this matching cost, all candidate segments are traversed, and the segment that makes the matching cost is selected. The smallest line segment is used as the reference point. Matching target line segment .

[0060] It is understandable that the method for determining the matching cost described above is merely an example. Other possible matching costs include the smaller squared distance from the reference point to the endpoints of the candidate line segment, or the squared distance component along the projection direction. Regardless of the form of the matching cost used, the reference point can be determined based on the criterion of "traversing the candidate line segments and selecting the one with the smallest matching cost". Corresponding target line segment .

[0061] In an optional embodiment of this application, determining the associated position corresponding to each target line segment includes: For each target line segment, the corresponding corrected projection position on the target line segment is determined as the associated position of the target line segment.

[0062] Optionally, the target line segment can be... The corresponding corrected projection position Defined as reference point And the associated position of the target line segment, denoted as Among them, the projection position Used to indicate a reference point The projection result to any candidate line segment, associated position Used to indicate a reference point Matched the target line segment The only associated position after that.

[0063] Step S105: Based on the geometric relationship between the associated positions of each target line segment, determine and output a structured result set, which includes the start and end positions of each gold line instance.

[0064] Optionally, based on the geometric relationships between the associated positions of each target line segment, the start and end positions of each gold line instance can be determined, and a structured result set containing multiple gold line instances can be output, such as... Figure 4 As shown, each gold wire instance includes at least an instance identifier, a start position, an end position, and the geometric parameters of the line segment corresponding to the instance. This structured result set can be used for subsequent open circuit detection, length measurement, direction analysis, or other quality control processes. A visualization example on a real sample is shown below. Figure 2 As shown in (c) in the figure.

[0065] In optional embodiments of this application, a structured result set is determined and output based on the geometric relationship between each target line segment and its corresponding associated position, including: For each target line segment, the reference point or associated position corresponding to the target line segment is used as the starting position of the gold line instance corresponding to the target line segment; Compare the squared distances from the associated position of the target line segment to the two endpoints of the target line segment, and determine the endpoint with the largest squared distance as the termination position of the gold line instance corresponding to the target line segment. The two endpoints of the target line segment are the start endpoint and the end endpoint of the target line segment. A structured result set is constructed based on the start and end positions of the gold line instance corresponding to each target line segment.

[0066] Optionally, when determining the reference point target line segment and its associated locations Next, it is necessary to further determine the gold line instance, that is, the extension direction and termination position of the target line segment. The starting position of the gold line instance can be determined by a reference point. Or its associated position on the target line segment In this embodiment of the application, to ensure the geometric consistency between the gold wire instance and the target line segment, it is preferable to use associated positions. As the starting position.

[0067] For the termination position of the gold line instance, it can be based on the target line segment. The two endpoints and Calculate the associated positions separately Squared distances to the two endpoints: in, For associated location To the endpoint The square distance, For associated location To the endpoint The squared distance.

[0068] Furthermore, the side containing the endpoint with the larger distance is considered the side furthest from the reference point, and the endpoint with the largest squared distance is taken as the termination position of the gold line instance. That is, when Time to take For the termination position, when Time to take The termination position is indicated by the given information. It is understood that the method for determining the termination position described above can be equivalently adjusted based on the integrity or breakage of the line segment, and this application does not impose any limitations on this.

[0069] In summary, such as Figure 5 As shown, the implementation process of the method provided in this application may specifically include data input (specifically including gold wire microscopic image + candidate reference area information), enhancement and preservation of edge noise, binarization and structure optimization, candidate line segment extraction and screening, reference area spatial analysis and reference point generation, geometric matching and association (specifically including projection + boundary correction + distance evaluation), direction and termination point determination, and output of structured result set, etc.

[0070] In this application, a set of reference points and a set of candidate line segments can be calculated based on the acquired microscopic image of gold wire and candidate reference region information. Then, geometric constraint matching and association processing are performed on the candidate line segment set based on the set of reference points to obtain a structured result set including the start and end positions of each gold wire instance. In this process, there is no need to rely on the conditions of model training. The gold wire microscopic image is constructed at the instance level based solely on image processing and geometric constraints. Gold wire instance results with geometric semantics and start and end information can be output. Furthermore, it is no longer limited by the influence of imaging noise, illumination changes, and detection parameters. Even in cases of gold wire breakage, intersection, or complex backgrounds, gold wire instance results can still be output. This can meet the application requirements for instance-level analysis in semiconductor packaging inspection scenarios and is more suitable for industrial applications such as semiconductor packaging inspection.

[0071] Optionally, the system implementing the method provided in this embodiment can be deployed in an automated optical inspection device or an offline inspection workstation in a semiconductor packaging production line. The system may consist of a processor, a memory, and a data acquisition interface connected to the imaging device. The processor executes program instructions stored in the memory to complete each processing step of gold wire instance-level segmentation and geometry construction. The system can also be deployed in an industrial control computer, edge computing device, or server environment, and receive images and auxiliary information via a network, outputting structured results to meet the engineering requirements of online inspection or centralized analysis on the production line. Optionally, the system specifically includes an image acquisition module, a preprocessing module, a reference point generation module, a candidate line segment extraction module, an instance construction module, and a result output module.

[0072] Regarding data input, the system's image acquisition module receives gold-line microscopic images and candidate reference region information used to assist in instance construction. The gold-line microscopic images can be input in matrix form; the candidate reference region information can be input in the form of binary masks, region marker maps, or pixel coordinate sets, and must be in the same image coordinate system as the gold-line microscopic images. For candidate reference region information, the system supports input using masks of the same resolution as the image, as well as input in the form of coordinate sets or vector annotations, and performs coordinate alignment and format conversion internally. The source of candidate reference region information is unrestricted; it can be obtained through manual annotation, rule extraction, device-side output, or external model output.

[0073] As shown in Figure 6, in terms of system processing flow, the preprocessing module performs image enhancement and edge-preserving denoising on the input image, and performs binarization and structure optimization based on the processing results, outputting a binary result. The candidate line segment extraction module extracts linear geometric features based on the binary result, forming a candidate line segment set including candidate line segments, and can filter candidate line segments according to geometric attributes such as line segment direction and line segment length to reduce the influence of background interference lines on subsequent instance construction. The reference point generation module spatially merges the candidate reference region information to generate a reference point set, where the reference points are used to represent the local reference position where the gold line starts. The instance construction module uses the reference point set as a basis to perform matching and association processing based on geometric projection and squared distance cost on the candidate line segment set, determines the target line segment corresponding to the reference point, and further determines the extension direction and termination position of the gold line instance, thereby constructing a gold line instance set including the start position and termination position. The result output module outputs the constructed gold line instance set in the form of a structured result set, i.e., outputs a structured result set.

[0074] Regarding the output format, the structured result set can be output using a table structure or a serializable data structure. For example, each gold wire instance can be used as a record unit, including at least an instance identifier, start position, end position, and geometric parameter information of the line segment corresponding to the target line segment; the geometric parameter information may include the endpoint coordinates, direction angle, and line segment length of the target line segment. The system can output the results to a file, database, or host computer software interface, and can further link the output results with the production line quality judgment module for use in open circuit detection, length measurement, direction analysis, or other quality control processing. The above output method is only an example; the system can adopt other equivalent data output and interaction methods according to engineering needs.

[0075] As can be seen from this embodiment, the system of the present invention can adapt to different deployment environments and input / output interface forms, has good engineering integrability and scalability, and is suitable for industrial application scenarios such as semiconductor packaging and testing.

[0076] This application provides a device for segmenting and geometric construction of gold wire micrographs. The device may include: an image acquisition module 701, a preprocessing module 702, a reference point generation module 703, an instance construction module 704, and a result output module 705. The relationships between the modules in this device are as follows: Figure 7 As shown, where, The image acquisition module is used to acquire the gold wire microscopic image of the chip to be processed and the candidate reference region information. The candidate reference region information is used to indicate the location information of the starting end of the gold wire instance. The candidate reference region and the gold wire microscopic image are in the same image coordinate system. The preprocessing module is used to perform image enhancement and edge-preserving denoising on the gold wire microscopic image to obtain an enhanced image, and then perform binarization and structure optimization on the enhanced image to obtain a binary result; The reference point generation module is used to perform linear geometric feature extraction based on the binary results to obtain a set of candidate line segments, and to perform spatial analysis on the candidate reference area information to obtain a set of reference points. The instance construction module is used to perform geometric constraint matching and association processing on the candidate line segment set based on the reference point set, to obtain the target line segment corresponding to each reference point in the reference point set, and to determine the associated position of each target line segment; based on the geometric relationship between each target line segment and its corresponding associated position, a structured result set is determined. The results output module is used to output a structured result set, which includes the start and end positions of each gold line instance.

[0077] Optionally, the preprocessing module, when performing image enhancement and edge-preserving denoising on the gold wire micrograph to obtain an enhanced image, and then performing binarization and structure optimization on the enhanced image to obtain a binary result, is specifically used for: Edge-preserving smoothing was applied to the gold wire micrograph to obtain an enhanced image; The enhanced image is processed using an adaptive thresholding method based on local region statistical features to obtain the binary result to be optimized. The binary result to be optimized is then combined with morphological optimization to obtain the final binary result.

[0078] Optionally, the linear geometric feature extraction processing is one of Hough line segment detection processing and skeleton-tracking-based linear segment construction. The device also includes a candidate line segment extraction module, which, when performing linear geometric feature extraction processing based on the binary results to obtain a candidate line segment set, is specifically used for: Linear geometric feature extraction is performed based on the binary results to obtain each initial candidate line segment; Each initial candidate line segment is filtered based on preset filtering conditions, and a candidate line segment set is formed based on the filtered initial candidate line segments; wherein, the preset filtering conditions include at least one of the minimum line segment length threshold and the maximum break gap threshold.

[0079] Optionally, when forming a candidate line segment set based on the filtered initial candidate line segments, the candidate line segment extraction module is specifically used for: For each filtered initial candidate line segment, the direction angle of the filtered initial candidate line segment is determined based on the endpoint coordinates of the filtered initial candidate line segment; Obtain a preset direction angle range, and perform direction filtering on each filtered initial candidate line segment based on the direction angle range and the direction angle of each filtered initial candidate line segment to obtain a set of candidate line segments.

[0080] Optionally, the candidate reference region information can be represented in any of the following forms: binary mask, region marker, or pixel coordinates. The candidate reference region information is obtained in the following ways: When the candidate reference region information is represented in either a binary mask or pixel coordinates, the binary mask image corresponding to the gold wire microscopic image is obtained, and the coordinate set corresponding to the non-zero pixels in the binary mask image is extracted as the foreground pixel set to obtain the candidate reference region information. When the candidate reference region information is represented in the form of region labeling, the enhanced image corresponding to the gold line microscopic image is obtained, and the local bright region is extracted based on the pixel brightness or local contrast in the enhanced image to generate the initial candidate binary mask. Connected component labeling is performed on the initial candidate binary mask to obtain at least one connected component, and the gray-level statistical features and morphological features of each connected component are calculated. Each connected component is filtered based on its gray-scale statistical and morphological characteristics, and the filtered connected components are used as candidate reference region information.

[0081] Optionally, when the reference point generation module performs spatial analysis on the candidate reference region information to obtain the reference point set, it is specifically used for: When the candidate reference region information is represented in either a binary mask or pixel coordinates, spatial aggregation processing is performed on the foreground pixels corresponding to the candidate reference region information to form at least one aggregation cluster; for each aggregation cluster, the geometric center of the pixel coordinates within the aggregation cluster is calculated, and the geometric center or the pixel with the smallest distance from the geometric center is taken as the reference point of the aggregation cluster; a reference point set is formed based on the reference point of each aggregation cluster. When candidate reference region information is represented in the form of region labels, connected component labeling is performed on the binary mask corresponding to the candidate reference region to obtain at least one region component, and each region component corresponds to a set of pixels; spatial merging is performed on the set of pixels corresponding to each region component to form at least one merged region; for each merged region, the centroid position of the merged region is determined, and the centroid position or the pixel with the smallest distance from the centroid position is taken as the reference point of the merged region; a set of reference points is formed based on the reference points of each merged region.

[0082] Optionally, when the instance construction module performs geometric constraint matching and association processing on the candidate line segment set based on the reference point set to obtain the target line segment corresponding to each reference point in the reference point set, it is specifically used for: For any reference point in the set of reference points and any candidate line segment in the set of candidate line segments, construct the projection relationship from the reference point to the candidate line segment to obtain the projection parameters and the initial projection position; when the initial projection position is outside the spatial range of the candidate line segment, constrain and correct the initial projection position according to the projection parameters to obtain the corrected projection position. Using the squared distance from the reference point to the corrected projection position as the matching cost, we traverse all candidate segments in the candidate segment set and select the candidate segment with the smallest matching cost as the target segment corresponding to the reference point.

[0083] Optionally, the instance building module is specifically used to determine the associated position corresponding to each target line segment when: For each target line segment, the corresponding corrected projection position on the target line segment is determined as the associated position of the target line segment.

[0084] Optionally, when the instance building module determines and outputs a structured result set based on the geometric relationship between each target line segment and its corresponding associated position, it is specifically used for: For each target line segment, the reference point corresponding to the target line segment is used as the starting position of the gold line instance corresponding to the target line segment; Compare the squared distances from the associated position of the target line segment to the two endpoints of the target line segment, and determine the endpoint with the largest squared distance as the termination position of the gold line instance corresponding to the target line segment. The two endpoints of the target line segment are the start endpoint and the end endpoint of the target line segment. A structured result set is constructed based on the start and end positions of the gold line instance corresponding to each target line segment.

[0085] The gold wire micrograph segmentation and geometry construction apparatus of this embodiment can execute the gold wire micrograph segmentation and geometry construction method shown in the embodiment of this application. The implementation principle is similar and will not be described again here.

[0086] This application provides an electronic device, which includes a processor and a memory configured to store machine-readable instructions that, when executed by the processor, cause the processor to perform a method for segmenting and geometrically constructing a gold wire micrograph.

[0087] This application provides an electronic device, such as... Figure 8 As shown, Figure 8 The illustrated electronic device includes a processor 2001 and a memory 2003. The processor 2001 and the memory 2003 are connected, for example, via a bus 2002. Optionally, the electronic device 2000 may further include a transceiver 2004. It should be noted that in practical applications, the transceiver 2004 is not limited to one type, and the structure of this electronic device 2000 does not constitute a limitation on the embodiments of this application.

[0088] Processor 2001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 2001 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0089] Bus 2002 may include a pathway for transmitting information between the aforementioned components. Bus 2002 may be a PCI bus or an EISA bus, etc. Bus 2002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0090] The memory 2003 may be ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0091] The memory 2003 stores the application code that executes the scheme of this application, and its execution is controlled by the processor 2001. The processor 2001 executes the application code stored in the memory 2003 to implement... Figure 7 The illustrated embodiment provides the operation of a segmentation and geometry construction device for gold wire micrographs.

[0092] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0093] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for segmentation and geometric construction of gold wire micrographs, characterized in that, include: Acquire a microscopic image of the gold wires of the chip to be processed and candidate reference region information. The candidate reference region information is used to indicate the location information of the starting end of the gold wire instance. The candidate reference region and the microscopic image of the gold wires are in the same image coordinate system. The gold wire microscopic image is subjected to image enhancement and edge-preserving denoising processing to obtain an enhanced image, and the enhanced image is subjected to binarization and structure optimization processing to obtain a binary result; Based on the binary results, linear geometric feature extraction is performed to obtain a candidate line segment set, and spatial analysis is performed on the candidate reference region information to obtain a reference point set. Geometric constraint matching and association processing are performed on the candidate line segment set based on the reference point set to obtain the target line segment corresponding to each reference point in the reference point set, and the associated position corresponding to each target line segment is determined. Based on the geometric relationship between the associated positions of each target line segment, a structured result set is determined and output, which includes the start and end positions of each gold line instance.

2. The method according to claim 1, characterized in that, The process of performing image enhancement and edge-preserving denoising on the gold wire micrograph to obtain an enhanced image, and then performing binarization and structure optimization on the enhanced image to obtain a binary result, includes: The enhanced image is obtained by applying edge-preserving smoothing to the gold wire micrograph. The enhanced image is processed using an adaptive thresholding method based on local region statistical features to obtain a binary result to be optimized. The binary result to be optimized is then combined with morphological optimization to obtain the final binary result.

3. The method according to claim 1, characterized in that, The linear geometric feature extraction process is one of Hough line segment detection and skeleton-based linear segment construction. The linear geometric feature extraction process based on the binary results yields a candidate line segment set, including: Based on the binary results, linear geometric feature extraction is performed to obtain each initial candidate line segment; Each of the initial candidate line segments is filtered based on preset filtering conditions, and a candidate line segment set is formed based on the filtered initial candidate line segments; wherein, the preset filtering conditions include at least one of a minimum line segment length threshold and a maximum break gap threshold.

4. The method according to claim 3, characterized in that, The process of forming a candidate line segment set based on the filtered initial candidate line segments includes: For each of the filtered initial candidate line segments, the direction angle of the filtered initial candidate line segment is determined based on the endpoint coordinates of the filtered initial candidate line segment; Obtain a preset direction angle range, and perform direction filtering on each of the filtered initial candidate line segments based on the direction angle range and the direction angle of each of the filtered initial candidate line segments to obtain the candidate line segment set.

5. The method according to claim 1, characterized in that, The candidate reference region information is represented in any one of the following forms: binary mask, region marker, or pixel coordinates. The candidate reference region information is obtained through the following methods: When the candidate reference region information is represented in either a binary mask or pixel coordinates, the binary mask image corresponding to the gold wire microscopic image is obtained, and the coordinate set corresponding to the non-zero pixels in the binary mask image is extracted as the foreground pixel set to obtain the candidate reference region information. When the candidate reference region information is represented in the form of region labeling, the enhanced image corresponding to the gold wire microscopic image is obtained, and the local bright region is extracted based on the pixel brightness or local contrast in the enhanced image to generate an initial candidate binary mask. Connected component labeling is performed on the initial candidate binary mask to obtain at least one connected component, and the gray-level statistical features and morphological features of each connected component are calculated to obtain the gray-level statistical features and morphological features of each connected component. Each connected component is filtered based on its gray-scale statistical features and morphological features, and the filtered connected components are used as candidate reference region information.

6. The method according to claim 5, characterized in that, The spatial analysis of the candidate reference region information to obtain a set of reference points includes: When the candidate reference region information is represented in either a binary mask or pixel coordinates, spatial aggregation processing is performed on the foreground pixels corresponding to the candidate reference region information to form at least one aggregation cluster; for each aggregation cluster, the geometric center of the pixel coordinates within the aggregation cluster is calculated, and the geometric center or the pixel with the smallest distance from the geometric center is taken as the reference point of the aggregation cluster; the reference point set is formed based on the reference point of each aggregation cluster. When the candidate reference region information is represented in the form of region labels, the binary mask corresponding to the candidate reference region is labeled with connected components to obtain at least one region component, and each region component corresponds to a set of pixels; spatial merging is performed on the set of pixels corresponding to each region component to form at least one merged region; for each merged region, the centroid position of the merged region is determined, and the centroid position or the pixel with the smallest distance from the centroid position is used as the reference point of the merged region; the reference point set is formed based on the reference point of each merged region.

7. The method according to claim 1, characterized in that, The step of performing geometric constraint matching and association processing on the candidate line segment set based on the reference point set to obtain the target line segment corresponding to each reference point in the reference point set includes: For any reference point in the set of reference points and any candidate line segment in the set of candidate line segments, construct the projection relationship from the reference point to the candidate line segment to obtain projection parameters and initial projection position; when the initial projection position is outside the spatial range of the candidate line segment, constrain and correct the initial projection position according to the projection parameters to obtain the corrected projection position. Using the squared distance from the reference point to the corrected projection position as the matching cost, all candidate line segments in the candidate line segment set are traversed, and the candidate line segment with the smallest matching cost is selected as the target line segment corresponding to the reference point.

8. The method according to claim 7, characterized in that, Determining the associated position corresponding to each target line segment includes: For each target line segment, the corresponding corrected projection position on the target line segment is determined as the associated position of the target line segment.

9. The method according to claim 8, characterized in that, The process of determining and outputting a structured result set based on the geometric relationships between the associated positions of each target line segment includes: For each target line segment, the reference point corresponding to the target line segment is taken as the starting position of the gold line instance corresponding to the target line segment; Compare the squared distances from the associated position of the target line segment to the two endpoints of the target line segment, and determine the endpoint with the largest squared distance as the termination position of the gold line instance corresponding to the target line segment. The two endpoints of the target line segment are the start endpoint and the end endpoint of the target line segment. The structured result set is constructed based on the start and end positions of the gold wire instance corresponding to each target line segment.

10. A device for segmenting and geometrically constructing gold wire micrographs, characterized in that, include: The image acquisition module is used to acquire a microscopic image of the gold wire of the chip to be processed and candidate reference region information. The candidate reference region information is used to indicate the location information of the starting end of the gold wire instance. The candidate reference region and the microscopic image of the gold wire are in the same image coordinate system. The preprocessing module is used to perform image enhancement and edge-preserving denoising on the gold wire micrograph to obtain an enhanced image, and to perform binarization and structure optimization on the enhanced image to obtain a binary result. The reference point generation module is used to perform linear geometric feature extraction processing based on the binary results to obtain a set of candidate line segments, and to perform spatial analysis on the candidate reference area information to obtain a set of reference points. The instance construction module is used to perform geometric constraint matching and association processing on the candidate line segment set according to the reference point set, to obtain the target line segment corresponding to each reference point in the reference point set, and to determine the associated position corresponding to each target line segment; and to determine the structured result set based on the geometric relationship between each target line segment and its corresponding associated position. The results output module is used to output a structured result set, which includes the start and end positions of each gold wire instance.

Images