N x N dot matrix circular target detection and missing completion method and system based on deep learning

Abstract

The application belongs to the technical field of computer vision and image processing, and specifically discloses an N*N dot matrix circular target detection and missing completion method and system based on deep learning, which comprises the following steps: pre-processing an input image and performing initial circular detection by using a pre-trained YOLO model; filtering the detection results based on geometric constraint conditions and eliminating misrecognized points; performing row and column clustering and structure grouping on normal points; preliminarily completing some missing points according to known dot matrix information; extrapolating and completing missing rows or columns; and finally outputting the coordinates and radius of all circle centers. The application combines deep learning and geometric rule constraints, can effectively deal with noise interference, bubble interference and dot matrix missing problems in a complex image background, significantly improves the accuracy, integrity and robustness of circular dot matrix detection, and is suitable for dot matrix automatic recognition and analysis in multiple detection scenes such as biochips and microfluidic detection.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and image processing technology, and is particularly applicable to high-precision detection of regular circular dot arrays and automatic completion of missing circles in scenarios such as image chips, microfluidics, and biological arrays. Specifically, it relates to a method and system for N×N dot array circular target detection and missing completion based on deep learning. Background Technology

[0002] In the field of molecular diagnostics, dot-matrix-based multiplex PCR detection methods are often used to simultaneously detect multiple target genes, such as multi-pathogen screening for infectious diseases and multi-indicator detection for genetic diseases. Before data analysis, it is necessary to identify the circular dot-matrix regions. Traditional dot-matrix detection methods rely on manual annotation or simple threshold segmentation, which has the following drawbacks:

[0003] Traditional dot matrix detection methods are mostly based on handcrafted features from image processing, such as edge detection and Hough transform. These methods have high requirements for image quality, perform poorly in complex scenes, and are difficult to handle situations where dots are missing.

[0004] With the development of deep learning technology, deep learning-based object detection methods have been applied in dot matrix detection. However, existing methods often suffer from incomplete detection and insufficient ability to handle missing dots. For example, although YOLO can locate targets, it only outputs bounding boxes, which makes it difficult to handle the problem of "missing circles" in the dot matrix and cannot guarantee geometric regularity. Relying solely on confidence filtering makes it difficult to distinguish noise from low-quality targets (such as blurry dots). Without considering the topological structure of the dot matrix (such as row and column spacing and symmetry), it cannot accurately identify outliers.

[0005] Furthermore, existing post-processing techniques for filling missing circles mostly rely on NMS or simple clustering, which cannot fill missing circles based on prior information such as row and column spacing and array dimensions. On the other hand, based on simple interpolation (such as linear fitting), since the point matrix is ​​not very regular and does not consider the local geometric features of the point matrix, it is impossible to confirm whether the filling point is the missing circle.

[0006] Therefore, there is an urgent need for a method that can quickly and accurately detect and complete an N×N dot matrix to meet the needs of practical applications. Summary of the Invention

[0007] The technical problem to be solved by this invention is to design a method and system for detecting and filling missing points in an N×N dot matrix circular target based on deep learning, so as to solve the problems of incomplete detection of N×N dot matrix in complex scenes and poor ability to handle missing points in existing methods. This invention enables high-precision detection of the dot matrix, noise removal and automatic filling of missing points in complex environments, thereby improving detection efficiency and robustness.

[0008] To address the aforementioned technical problems, this invention provides a deep learning-based method for detecting and completing missing objects in an N×N dot matrix circular target, where N is a positive integer and N≤6. The method specifically includes the following steps:

[0009] Step S1: Perform initial circle search based on the pre-trained YOLO model to obtain initial detection point information.

[0010] Step S2: Filter based on the geometric constraints of the dot matrix to remove circles that are incorrectly identified due to background noise or experimental bubbles.

[0011] Step S3: Remove the outliers marked as incorrectly identified in step S2 from the model output.

[0012] Step S4: Perform structural grouping on the remaining normal points, and determine the specific row and column position of each point in the N×N matrix based on the row and column clustering results.

[0013] Step S5: After determining the theoretical row and column combinations of the N×N lattice, identify the missing lattice points and complete them.

[0014] Step S6: Determine if there are any missing rows or columns:

[0015] If it exists, proceed to step S7;

[0016] If it does not exist, proceed to step S8.

[0017] Step S7: Perform extrapolation and completion: First, confirm the direction of extrapolation, that is, whether the missing data is an entire row or an entire column; then, perform extrapolation search and completion based on the direction of the missing data.

[0018] Step S8: Output the center coordinates and radius of all identified and completed circular dot matrices.

[0019] Furthermore, in this invention, step S1 specifically includes the following steps:

[0020] Step S11: Perform Contrast Limited Adaptive Histogram Equalization (CLAHE) on the input image to enhance image contrast, improve image quality, and highlight the difference between the dot matrix target and the background.

[0021] Step S12: Load the pre-trained YOLO model and input the pre-processed image into the model for detection.

[0022] Step S13: Based on confidence threshold and area threshold The detection box is filtered to remove noise or small bubbles, retaining the valid dot matrix detection results, and obtaining the bounding box, center coordinates, radius, and confidence information of the dot matrix.

[0023] Furthermore, in step S13 of this invention, the confidence threshold... The value is Area threshold The value is Filter out detection boxes with low confidence scores below 0.75 and areas smaller than 50 pixels.

[0024] Furthermore, in step S2 of this invention, the lattice geometric constraints include:

[0025] Constraint (1): Clustering algorithms based on distance thresholds will cluster nodes with distances less than the threshold. The coordinates are grouped into the same category. Threshold clustering is performed on the coordinates of the detected center points in both row and column directions to obtain n row lines and n column lines of an n×n array, as well as the average row coordinates and average column coordinates of the identified points in each row and column after clustering.

[0026] Constraint (2): For multiple circles that are close to each other and meet the constraint (1), the most suitable point matrix is ​​selected based on their features in the image. Closer means that the distance between the centers of the circles is less than 1.5 times the radius.

[0027] Constraint (3): Mark circles that do not satisfy constraints (1) and (2) as incorrectly identified outliers.

[0028] In this invention, a further specific feature selection method is as follows: the average gray value of the circular region composed of the center and radius of the identified circle is compared with the average gray value of the square with the center as the center and the side length is greater than the diameter of the circular dot matrix to select the dot matrix that meets the requirements. For the characteristic that the dot matrix should be brighter than the background pixels, the larger the difference, the higher the probability of it being a circle. This rule is used to remove the circles that are repeatedly identified.

[0029] Furthermore, in step S5 of this invention, the method for initially completing the missing point matrix is ​​as follows: based on the average row coordinates, average column coordinates, and radius parameters of the corresponding known point matrix, calculate the average row coordinates of the other known point matrices corresponding to the location of the missing point matrix. and average column coordinates and average radius and with Centered on a circle with a radius of 3 to 5 times the radius The search is performed within the area, and the most likely location of the missing points is calculated by combining image features to complete the initial completion of the missing points.

[0030] Furthermore, the search rule in this invention is as follows: after determining the search area, all points in the search area are traversed, and these points are used as the center of a circle. The average radius of all identified circles is used as the radius of the search circle. The point positions that meet the requirements are filtered out by comparing the difference between the average gray value of the circular area and the average gray value of the rectangle centered on the circle and with a side length greater than the diameter of the search circular point array. The larger the difference, the higher the probability of it being a circle. The point with the largest difference in the area is selected, and this point is the most likely center position of the circular point array.

[0031] Furthermore, in step S7 of this invention, the specific search rule is as follows:

[0032] If the missing row / column is in the middle, the average of the already solved center coordinates in the row / column direction is directly calculated. The average coordinates are the search center of the missing circle. Then, the precise circle search and completion are performed according to the method in step S5.

[0033] If the missing rows / columns are on both sides, then the search needs to be performed on both sides of the already solved rows / columns using the method in step S5, and the set of points with the largest difference in gray values ​​should be selected as the completion points.

[0034] This invention also provides a deep learning-based N×N dot matrix circular target detection and missing object completion system, which employs the aforementioned deep learning-based N×N dot matrix circular target detection and missing object completion method, and includes the following modules:

[0035] Image acquisition module: Used to acquire input images containing regular circular dot matrices.

[0036] Preprocessing module: Used to perform block histogram equalization on the input image to enhance image contrast.

[0037] Initial Detection Module: This module loads the pre-trained YOLO model, performs initial circle detection on the pre-processed image, and outputs the initial detection results, including the center coordinates, radius, and confidence score.

[0038] Geometric constraint filtering module: used to cluster and filter the initial detection results based on the set distance threshold and grayscale difference features, in order to remove incorrectly identified outliers.

[0039] Structure grouping module: used to perform row and column direction clustering on the filtered normal points to determine the position of each point in the N×N point matrix.

[0040] Missing point completion module: Based on the row and column coordinates and radius of the known point matrix, it searches or extrapolates at the estimated missing positions to confirm the missing points and complete them.

[0041] The result output module is used to integrate all the identified and completed dot matrix information, output the coordinates and radius of each circle's center, and generate complete N×N dot matrix data.

[0042] Furthermore, in the initial detection module, the training process of the pre-trained YOLO model is as follows: First, a large number of raster data images are collected, and the raster regions in the images are manually labeled using a labeling tool; then, the raster regions in the labeled images are enhanced using methods such as rotation, translation, and scaling, which allows for the manual construction of more datasets and improves the robustness of the model; finally, the training set and validation set are divided, and the model is trained based on the divided and labeled images and labels, ultimately obtaining the trained circle-finding model.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] This invention presents a deep learning-based N×N dot matrix circular target detection and missing point completion method. It employs a YOLO model for initial detection, combined with image preprocessing and filtering mechanisms, enabling rapid and accurate detection of N×N dots in an image, thus improving detection efficiency and accuracy. Through cluster analysis and outlier removal, this invention can filter out effective dots, reducing the impact of noise and false detections on the results. For missing dots, this invention completes the detection based on the features of known dots, improving the completeness of dot matrix detection and ensuring a complete N×N dot matrix is ​​obtained even when some dots are missing. This invention exhibits strong robustness, adapting to different image qualities and complex scenes, and has broad application prospects. Attached Figure Description

[0045] The specific embodiments of the present invention will be further explained below with reference to the accompanying drawings.

[0046] Figure 1 This is a flowchart of the deep learning-based N×N dot matrix circular target detection and missing object completion method of the present invention.

[0047] Figure 2 This is a schematic diagram of the initial circular dot matrix search result based on the pre-trained YOLO model in Example 1.

[0048] Figure 3 This is a schematic diagram of the initial dot matrix completion effect in Example 1.

[0049] Figure 4 This is a schematic diagram of the dot matrix extrapolation effect in Example 1.

[0050] Figure 5 This is a system block diagram of the deep learning-based N×N dot matrix circular target detection and missing object completion system of the present invention. Detailed Implementation

[0051] Example 1

[0052] Combination Figure 1 This embodiment provides a deep learning-based method for detecting and completing missing objects in an N×N dot matrix circular target, where N is a positive integer and N≤6. In this embodiment, N=3 is specifically chosen, taking a 3×3 dot matrix as an example. The specific steps include:

[0053] Step S1: Perform initial circle search based on the pre-trained YOLO model to obtain initial detection point information.

[0054] In this embodiment, the main process of model training is as follows: First, a large number of dot matrix data images are organized, and the dot matrix regions in the images are manually labeled using a labeling tool; then, the labeled images are enhanced, mainly the dot matrix regions, by methods such as rotation, translation, and scaling. This allows for the manual construction of more datasets and improves the robustness of the model; finally, the training set and validation set are divided, and the model is trained based on the divided and labeled images and labels to obtain the trained circle-finding model.

[0055] In this embodiment, step S1 preferably includes the following steps:

[0056] Step S11: Perform Contrast Limited Adaptive Histogram Equalization (CLAHE) on the input image to enhance image contrast, improve image quality, and highlight the difference between the dot matrix target and the background.

[0057] Step S12: Load the pre-trained YOLO model and input the pre-processed image into the model for detection.

[0058] Step S13: Based on confidence threshold and area threshold The detection box is filtered to remove noise or small bubbles, retaining the valid dot matrix detection results, and obtaining the bounding box, center coordinates, radius, and confidence information of the dot matrix.

[0059] The results of finding the initial circular dot matrix based on the pre-trained YOLO model are as follows: Figure 2 As shown.

[0060] In this embodiment, preferably, in step S13, the confidence threshold is... The value is Area threshold The value is Filter out detection boxes with low confidence scores below 0.75 and areas smaller than 50 pixels (which are likely noise or small bubbles, not normal dot size).

[0061] Step S2: Filter based on the geometric constraints of the dot matrix to remove circles that are incorrectly identified due to background noise or experimental bubbles.

[0062] In this embodiment, preferably, the lattice geometric constraints in step S2 include:

[0063] Constraint (1): Clustering algorithms based on distance thresholds will cluster nodes with distances less than the threshold. The coordinates are grouped into the same category. Specifically, in this embodiment, the threshold... The coordinates of the detected center points are set to 1.5 to 2.5 times the average radius. Threshold clustering is performed on the coordinates of the detected center points in both row and column directions to obtain n row lines and n column lines of an n×n array, as well as the average row coordinates and average column coordinates of the identified points in each row and column after clustering.

[0064] Constraint (2): For multiple circles that are close to each other and meet the constraint (1), the most suitable point matrix is ​​selected based on their features in the image. Closer means that the distance between the centers of the circles is less than 1.5 times the radius.

[0065] In this embodiment, the preferred feature selection method is as follows: the average gray value of the circular region composed of the center and radius of the identified circle is compared with the average gray value of the square with the center as the center and the side length is greater than the diameter of the circular dot matrix to select the dot matrix that meets the requirements. For the characteristic that the dot matrix should be brighter than the background pixels, the larger the difference, the higher the probability of it being a circle. This rule is used to remove the circles that are repeatedly identified.

[0066] Constraint (3): Mark circles that do not satisfy constraints (1) and (2) as incorrectly identified outliers.

[0067] Step S3: Remove the outliers marked as incorrectly identified in step S2 from the model output.

[0068] Step S4: Perform structural grouping on the remaining normal points, and determine the specific row and column position of each point in the N×N matrix based on the row and column clustering results.

[0069] Step S5: After determining the theoretical row and column combinations of the N×N lattice, identify the missing lattice points and complete them.

[0070] In this embodiment, preferably, in step S5, preliminary completion is performed for a small number of missing points (not all of them occupy the same row or column; there are other known points in the same row and column of the missing points). The method for preliminary completion of the missing points is as follows: based on the average row coordinates, average column coordinates, and radius parameters of the corresponding known points, the average row coordinates of the other known points corresponding to the location of the missing points are calculated. and average column coordinates and average radius and with Centered on a circle with a radius of 3 to 5 times the radius The search is performed within the area, and the most likely location of the missing points is calculated by combining image features to complete the initial completion of the missing points.

[0071] In this embodiment, preferably, the search rule is as follows: after determining the search area, all points in the search area are traversed, and these points are used as the center of a circle. The average radius of all identified circles is used as the radius of the search circle. The point positions that meet the requirements are filtered by comparing the difference between the average gray value of the circular area and the average gray value of the rectangle centered on the circle and with a side length greater than the diameter of the search circular point array. The larger the difference, the higher the probability of it being a circle. The point with the largest difference in the area is selected, and this point is the most likely center position of the circular point array.

[0072] The completion effect of step S5 is as follows: Figure 3 As shown, since the size of the dot matrix is ​​basically the same, the radius of the completed circle uses the average radius of all the identified dots. Yellow represents the completed circle, and red represents the initially identified circle.

[0073] Step S6: Determine whether there are any missing rows or columns after the initial missing dot matrix search and completion in Step S5.

[0074] If it exists, proceed to step S7;

[0075] If it does not exist, proceed to step S8.

[0076] Step S7: Perform extrapolation and completion: First, confirm the direction of extrapolation, that is, whether the missing data is an entire row or an entire column; then, perform extrapolation search and completion based on the direction of the missing data.

[0077] In this embodiment, preferably, the specific search rule in step S7 is as follows:

[0078] If the missing row / column is in the middle, the average of the already solved center coordinates in the row / column direction is directly calculated. The average coordinates are the search center of the missing circle. Then, the precise circle search and completion are performed according to the method in step S5.

[0079] If the missing rows / columns are on both sides, then the search needs to be performed on both sides of the already solved rows / columns using the method in step S5, and the set of points with the largest difference in gray values ​​should be selected as the completion points.

[0080] In this embodiment, specifically taking the behavior of missing both sides of a 3×3 dot matrix as an example, combined with... Figure 4As shown in the figure, the red circles represent the initially identified circles. During the search, the area above the first row is used as the search area. The search rules in step S5 are executed to search for three circles, and the sum of the differences in the average grayscale values ​​of the circles and their circumscribed rectangles is output. Then, the area below the second row is used as the search area, and the same operation is performed, resulting in another set of sums of the differences in the average grayscale values ​​of the circles and their circumscribed rectangles. Comparing the two sets, the larger sum indicates that the three circles in the row that need to be filled are more likely to be the three circles in the row that need to be filled. In this embodiment, the sum of the differences in the average grayscale values ​​of the three circles in the area below the second row is larger, meaning that the 3*3 dot matrix is ​​missing the third row. It is then filled in, as shown below. Figure 4 As shown, yellow represents the extrapolated circle, and red represents the initially identified circle.

[0081] Step S8: Output the center coordinates and radius of all identified and completed circular dot matrices. Organize all the filtered model outputs of circular dot matrices, as well as the results of subsequent insertions and extrapolations, outputting the coordinates and radius of each circle's center to achieve precise positioning of the circular dot matrices.

[0082] Example 2

[0083] Combination Figure 5 As shown, the N×N dot matrix circular target detection and missing object completion system based on deep learning in this embodiment adopts the N×N dot matrix circular target detection and missing object completion method based on deep learning in Embodiment 1, and includes the following modules:

[0084] Image acquisition module: Used to acquire input images containing regular circular dot matrices.

[0085] Preprocessing module: Used to perform block histogram equalization on the input image to enhance image contrast.

[0086] Initial Detection Module: This module loads the pre-trained YOLO model, performs initial circle detection on the pre-processed image, and outputs the initial detection results, including the center coordinates, radius, and confidence score.

[0087] Geometric constraint filtering module: used to cluster and filter the initial detection results based on the set distance threshold and grayscale difference features, in order to remove incorrectly identified outliers.

[0088] Structure grouping module: used to perform row and column direction clustering on the filtered normal points to determine the position of each point in the N×N point matrix.

[0089] Missing point completion module: Based on the row and column coordinates and radius of the known point matrix, it searches or extrapolates at the estimated missing positions to confirm the missing points and complete them.

[0090] The result output module is used to integrate all the identified and completed dot matrix information, output the coordinates and radius of each circle's center, and generate complete N×N dot matrix data.

[0091] In this embodiment, preferably, the training process of the pre-trained YOLO model in the initial detection module is as follows: First, a large number of bitmap data images are collected, and the bitmap regions in the images are manually labeled using a labeling tool; then, the bitmap regions in the labeled images are enhanced using methods such as rotation, translation, and scaling, which allows for the manual construction of more datasets and improves the robustness of the model; finally, the training set and validation set are divided, and the model is trained based on the divided and labeled images and labels to obtain the trained circle-finding model.

[0092] Many specific details have been set forth in the foregoing description to provide a thorough understanding of the present invention. However, the above description is merely a preferred embodiment of the present invention, and the present invention can be implemented in many other ways different from those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed above. Furthermore, any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention, or modify them into equivalent embodiments, using the methods and techniques disclosed above, without departing from the scope of the present invention. Any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention, without departing from the content of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A method for detecting and completing missing objects in an N×N dot matrix based on deep learning, characterized in that: N is a positive integer and N≤6, specifically including the following steps: Step S1: Perform initial circle search based on the pre-trained YOLO model to obtain initial detection point information; Step S2: Filter based on the geometric constraints of the dot matrix to remove circles that are incorrectly identified due to background noise or experimental bubbles; Step S3: Remove the outliers marked as incorrectly identified in step S2 from the model output; Step S4: Perform structural grouping on the remaining normal points, and determine the specific row and column position of each point in the N×N matrix based on the row and column clustering results; Step S5: After determining the theoretical row and column combinations of the N×N lattice, initially identify and complete the missing lattice points; Step S6: Determine if there are any missing rows or columns: If it exists, proceed to step S7; If it does not exist, proceed to step S8; Step S7: Perform extrapolation and completion: First, confirm the direction of extrapolation, i.e., whether the missing data is an entire row or an entire column; then, perform extrapolation search and completion based on the direction of the missing data. Step S8: Output the center coordinates and radius of all identified and completed circular dot matrices; In step S5, the method for initially completing the missing point matrix is ​​as follows: based on the average row coordinates, average column coordinates, and radius parameters of the corresponding known point matrix, calculate the average row coordinates of the other known point matrices corresponding to the location of the missing point matrix. and average column coordinates and average radius and with Centered on a circle with a radius of 3 to 5 times the radius The search is conducted within the defined area, and the most likely point location is calculated by combining image features to complete the initial completion of the missing point array. The search rules are as follows: after determining the search area, all points in the search area are traversed, and these points are used as the center of a circle. The average radius of all identified circles is used as the radius of the search circle. The point location that meets the requirements is selected by comparing the difference between the average gray value of the circular area and the average gray value of the rectangle centered at the center of the circle and with a side length greater than the diameter of the search circular point array. The point with the largest difference in the area is the most likely center location of the circular point array. In step S7, the extrapolation search rule is as follows: If the missing row / column is in the middle, then the average of the already solved center coordinates in the row / column direction is taken. The average coordinates are the search center of the missing circle. Then, the precise circle search and completion are performed according to the method in step S5. If the missing rows / columns are on both sides, then the search needs to be performed on both sides of the already solved rows / columns using the method in step S5, and the set of points with the largest difference in gray values ​​should be selected as the completion points.

2. The method for detecting and completing missing objects in an N×N dot matrix based on deep learning according to claim 1, characterized in that: Step S1 specifically includes the following steps: Step S11: Perform block histogram equalization on the input image to enhance image contrast; Step S12: Load the pre-trained YOLO model and input the pre-processed image into the model for detection; Step S13: Based on confidence threshold and area threshold The detection boxes are filtered to retain valid dot matrix detection results, and the bounding boxes, center coordinates, radii, and confidence information of the dots are obtained.

3. The method for detecting and completing missing objects in an N×N dot matrix based on deep learning according to claim 2, characterized in that: In step S13, the confidence threshold The value is Area threshold The value is Filter out detection boxes with low confidence scores below 0.75 and areas smaller than 50 pixels.

4. The method for detecting and completing missing objects in an N×N dot matrix based on deep learning according to claim 1, characterized in that: In step S2, the lattice geometric constraints include: Constraint (1): Clustering algorithms based on distance thresholds will cluster nodes with distances less than the threshold. The coordinates are grouped into the same category. Threshold clustering is performed on the coordinates of the detected center of the circle in both row and column directions to obtain n row lines and n column lines of the n×n array, as well as the average row coordinates and average column coordinates of the identified points in each row and column after clustering. Constraint (2): For multiple circles that are close to each other and meet the constraint (1), the most suitable point matrix is ​​selected based on their features in the image. Closer means that the distance between the centers of the circles is less than 1.5 times the radius. Constraint (3): Mark circles that do not satisfy constraints (1) and (2) as incorrectly identified outliers.

5. The method for detecting and completing missing objects in an N×N dot matrix based on deep learning according to claim 4, characterized in that: The specific feature selection method is as follows: the average gray value of the circular region composed of the center and radius of the identified circle is compared with the average gray value of the square with the center as the center and the side length is greater than the diameter of the circular dot matrix to select the dot matrix that meets the requirements. The larger the difference, the higher the probability of it being a circle. This rule is used to remove the circles that are repeatedly identified.

6. A deep learning-based N×N dot matrix circular target detection and missing object completion system, characterized in that: The method for detecting and completing missing objects in an N×N dot matrix based on deep learning as described in any one of claims 1-5 includes the following modules: Image acquisition module: used to acquire input images containing regular circular dot matrices; Preprocessing module: Used to perform block histogram equalization on the input image to enhance image contrast; Initial Detection Module: This module loads the pre-trained YOLO model, performs initial circle detection on the pre-processed image, and outputs initial detection results including the center coordinates, radius, and confidence score. Geometric constraint filtering module: used to cluster and filter the initial detection results based on the set distance threshold and gray-level difference features to remove misidentified outliers; The structure grouping module is used to perform row and column clustering on the filtered normal points to determine the position of each point in the N×N matrix. Missing point completion module: It is used to search or extrapolate at the estimated missing position based on the row and column coordinates and radius of the known point matrix to confirm the missing point matrix and complete it. The result output module is used to integrate all the identified and completed dot matrix information, output the coordinates and radius of each circle's center, and generate complete N×N dot matrix data.

7. The deep learning-based N×N dot matrix circular target detection and missing object completion system according to claim 6, characterized in that: In the initial detection module, the training process of the pre-trained YOLO model is as follows: First, a large number of dot matrix data images are collected, and the dot matrix regions in the images are manually labeled using a labeling tool; then, the dot matrix regions in the labeled images are enhanced; finally, the training set and the validation set are divided, and the model is trained based on the divided and labeled images and labels, and finally the trained circle-finding model is obtained.

Images