A particle image circle determination and comprehensive occlusion identification method and system
Patent Information
- Application Number
- CN202610988716.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-03
- Publication Date
- 2026-11-03
AI Technical Summary
Existing technologies suffer from low accuracy in determining the roundness of particle images, slow detection of neighboring particles, incomplete occlusion detection, and a lack of integrated solutions, resulting in inaccurate particle quality assessment and low efficiency.
A dual-index fusion circularity detection algorithm combined with quadtree index optimization is adopted. The circularity of particles is determined by the circularity of isocircularity and the circularity of least squares circle fitting. The occlusion state is detected by edge truncation, concavity degree and contact with neighboring particles. An integrated particle image recognition system is constructed.
It improves the accuracy of circle detection, reduces the false positive rate, enhances the accuracy of occlusion detection, reduces processing time, and enables real-time detection and efficient evaluation of large-scale particle images.
Smart Images
Figure CN123033236A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of image processing and computer vision, specifically a method and system for recognizing the roundness and occlusion status of particle images. This invention is particularly suitable for the automatic detection, classification, and quality assessment of particles in images. Background Technology
[0002] In fields such as biomedical research, materials science, and industrial testing, the morphological characteristics analysis of particles is of great significance for quality assessment and scientific research. Among them, the roundness and occlusion state of particles are key indicators for evaluating particle quality, directly affecting product performance and the accuracy of experimental results.
[0003] In existing technologies, the recognition of roundness and occlusion status of particle images has the following main problems: First, the method for judging roundness is too simplistic, typically relying solely on roundness indicators, which can easily lead to misjudgments. Some non-circular particles may perform well on one indicator but poorly on others, resulting in inaccurate judgments and affecting subsequent quality assessments.
[0004] Second, the detection efficiency of neighboring particles is low. Traditional methods use brute-force search to check the contact between each pair of particles. When the number of particles is large, the computational complexity increases quadratically, and the processing speed drops drastically, making it difficult to meet the needs of real-time applications.
[0005] Third, the occlusion detection method is not comprehensive enough. It can only identify occlusion caused by a single factor and cannot accurately identify occlusion caused by multiple factors, resulting in a high false negative rate.
[0006] Fourth, there is a lack of integrated system solutions. Particle roundness recognition and occlusion detection are usually implemented as two separate modules. The lack of a unified system framework makes it difficult to achieve efficient integrated applications and data interoperability.
[0007] Fifth, the algorithm lacks robustness. When processing particle images of different types and qualities, it is easily affected by factors such as image noise, lighting conditions, and particle overlap, resulting in poor adaptability.
[0008] Therefore, there is an urgent need to propose a method and system for recognizing the roundness and occlusion status of particle images that is accurate, fast, robust, and integrated. Summary of the Invention
[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for determining the circularity of particles and identifying comprehensive occlusion in particle images. This solves the problems of low accuracy in determining circularity, slow detection speed of adjacent particles, incomplete occlusion detection, and lack of integrated solutions in the prior art, thereby achieving rapid and accurate identification of particle morphological features and improving the efficiency and accuracy of particle quality assessment.
[0010] To achieve the above objectives, the present invention adopts the following technical solution: A method for determining the type of circle in a particle image and for comprehensive occlusion recognition includes the following steps: S1: Obtain the particle label mask output by the deep learning segmentation model, with each particle corresponding to a unique ID identifier; S2: Preprocess the contour of each particle to extract geometric features such as contour coordinates, area, perimeter, width and height of the bounding rectangle, and aspect ratio; S3: A dual-index fusion algorithm for roundness detection is used to identify roundness. When the roundness value of the isoperimeter inequality of the particle is greater than or equal to the first threshold and the roundness value of the least squares circle fitting is greater than or equal to the second threshold, the particle is determined to be a round particle; otherwise, it is determined to be a non-round particle. S4: Optimize the neighbor particle query process by constructing a quadtree spatial index, and calculate the minimum distance between contours only for particles in the candidate particle set; S5: Based on edge truncation, concavity and contact with neighboring particles, a comprehensive occlusion detection is performed. If a particle is truncated by the image edge, or if the concavity is greater than the third threshold and there is contact with neighboring particles, then the particle is determined to be an occluded particle; otherwise, it is determined to be an unoccluded particle. S6: Output the ID, roundness determination result, occlusion state determination result, and corresponding feature value of each particle.
[0011] Furthermore, the circularity value of the isocircular inequality is calculated based on the area and perimeter of the particle profile; the circularity value of the least squares circle fitting is calculated based on the ratio of the root mean square error of the particle profile point to the optimal fitted circle to the radius of the fitted circle.
[0012] Furthermore, the construction process of the quadtree spatial index is as follows: initialize the root node covering the entire image range, calculate the axis-aligned bounding rectangle of each particle; if the number of particles in the current node exceeds the preset maximum capacity of a single node, then subdivide the node into four quadrant child nodes: upper left, upper right, lower left, and lower right, and redistribute the particles to the corresponding child nodes; for particles whose bounding rectangle crosses the node boundary, insert them into all intersecting child nodes simultaneously, or expand the search of the boundaries of adjacent nodes during the query.
[0013] Furthermore, the process of querying neighboring particles is as follows: define a search radius for the current particle, recursively query from the root node, collect all particles in the nodes that intersect with the search range as candidate particles; only calculate the minimum contour distance between the current particle and the candidate particles.
[0014] Furthermore, the edge truncation determination is based on the existence of any contour point whose distance from the image boundary is less than the preset boundary inner margin; the concavity degree is calculated based on the ratio of the particle contour area to the convex hull area; the adjacent particle contact determination is based on the minimum Euclidean distance between two particle contours being less than a preset contact distance threshold.
[0015] A particle image circle determination and comprehensive occlusion recognition system implementing the above method includes: The image segmentation module is used to obtain the label mask after particle segmentation and extract the contour geometric features of each particle. A circle detection module, connected to the image segmentation module, is used to execute a circle determination algorithm based on dual-index fusion. The quadtree index module, connected to the image segmentation module, is used to construct a quadtree spatial index and query neighboring particles; The occlusion detection module is connected to the image segmentation module, the circle detection module, and the quadtree indexing module, respectively, and is used to perform a comprehensive occlusion determination based on edge truncation, concavity, and proximity contact. The output module is connected to the circular detection module and the occlusion detection module respectively, and is used to output the recognition result of each particle.
[0016] Furthermore, the system also includes: The camera acquisition module is used to acquire particle images in real time and supports dynamic adjustment of parameters such as aperture, shutter speed, and white balance. The calibration and measurement module is connected to the camera acquisition module and is used to perform camera calibration, calculate the millimeter / pixel ratio, and achieve accurate measurement of particle size. The data statistics module, connected to the output module, is used to perform statistical analysis on the recognition results, including histogram drawing, pass rate calculation, and data export.
[0017] Furthermore, the image segmentation module achieves automatic particle segmentation based on the Cellpose deep learning model.
[0018] Furthermore, the circle-like detection module includes an isocircular inequality calculation unit, a least squares circle fitting unit, and a circle-like determination unit.
[0019] Furthermore, the occlusion detection module includes an edge truncation detection unit, a concavity calculation unit, a proximity contact detection unit, and a comprehensive judgment unit.
[0020] Furthermore, the data statistics module supports real-time preview of recognition results and can dynamically adjust the roundness threshold, concavity threshold, and contact distance threshold.
[0021] Compared with the prior art, the present invention has the following advantages: A circularity determination algorithm that integrates the circularity of isocircularity inequality and the circularity of least squares circle fitting is adopted. Compared with a single method, the accuracy of circularity detection is improved by more than 11%, and the false positive rate is significantly reduced. When combined with the comprehensive occlusion detection of concavity degree and contact with adjacent particles, the accuracy of occlusion recognition is improved by more than 12%.
[0022] The neighbor particle query process is optimized by using a quadtree spatial index, reducing the time complexity of neighbor particle detection from O(n^2) to O(n^2). 2 The processing time is reduced to O(n log n), increasing the processing speed by tens of times, making it particularly suitable for real-time applications of large-scale particle detection.
[0023] By integrating three dimensions—edge truncation, concavity, and contact with adjacent particles—occlusion detection can accurately identify occlusion phenomena caused by various factors and has good adaptability to particle images of different types and qualities.
[0024] It adopts an integrated system design, which integrates the entire process of image acquisition, calibration, segmentation, detection, and statistical analysis. It supports real-time preview and dynamic parameter adjustment, and is easy to operate, deploy, and use.
[0025] This method and system are applicable to multiple fields such as biomedical cell detection, particle analysis in materials science, and quality testing of industrial products, and have broad market application prospects. Attached Figure Description
[0026] Figure 1 This is an overall flowchart of the particle image roundness and occlusion state recognition method provided in the embodiments of the present invention; Figure 2 This is a preprocessing flowchart of the particle image roundness and occlusion state recognition method provided in an embodiment of the present invention; Figure 3 The overall algorithm flowchart of the particle image roundness and occlusion state recognition method provided in the embodiments of the present invention is shown below; Figure 4 This is a schematic diagram of the circularity calculation of the isoperimetric inequality provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of least-squares circle fitting provided in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the identification results of spherical and non-spherical particles provided in an embodiment of the present invention. Figure 7This is a schematic diagram of edge truncation detection provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the edge truncation detection results provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the concavity calculation provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of a quadtree neighbor granule query provided in an embodiment of the present invention; Figure 11 This is a schematic diagram illustrating the identification results of obscured and unobscured particles provided in an embodiment of the present invention. Figure 12 This is a schematic diagram of the system real-time preview architecture provided in an embodiment of the present invention; Figure 13 This is a schematic diagram of the system quality detection architecture provided in an embodiment of the present invention. Detailed Implementation
[0027] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0028] This invention addresses the problems of low accuracy in circularity determination, slow detection of neighboring particles, incomplete occlusion detection, and lack of integrated solutions in existing particle image recognition technologies. It proposes a comprehensive occlusion recognition technology based on dual-index fusion of circularity determination and quadtree optimization, which can achieve real-time detection of large-scale particle images while ensuring high accuracy.
[0029] I. Examples of Particle Image Circle Determination and Comprehensive Occlusion Recognition Method Reference Figure 1 and Figure 3 The particle image circle determination and comprehensive occlusion recognition method described in this embodiment specifically includes the following steps: S1: Obtain the label mask after particle segmentation Obtain the particle label mask output by the deep learning segmentation model, with each particle corresponding to a unique ID. This embodiment preferably uses the Cellpose model, which has good generalization ability for particles of different shapes and densities, accurately extracting the boundaries of adhered particles and assigning a unique integer ID to each particle. After inputting the RGB or grayscale particle image to be detected, the Cellpose model outputs a single-channel label mask of the same size as the input image. The value of each pixel in the mask is the ID of the corresponding particle, while the background pixel values are uniformly set to 0.
[0030] S2: Particle contour preprocessing and geometric feature extraction Reference Figure 2 Extract the contour coordinates, area, perimeter, width and height of the bounding rectangle, and aspect ratio of each particle. Based on the label mask, extract the closed contour coordinate sequence (x1, y1), (x2, y2), ..., (x...). n ,y n ), where (x n+1 ,y n+1 The coordinates are set to (x1, y1) to ensure contour closure. Based on the contour coordinate sequence, the following core geometric features are extracted: Outline area A: Calculated using the Shoelace formula, the formula is as follows:
[0031] The perimeter P of the profile is the sum of the Euclidean distances between adjacent profile points, and the formula is:
[0032] Parameters of the bounding rectangle: width = max(x) - min(x), height = max(y) - min(y); Aspect Ratio: .
[0033] S3: Circle-like detection and judgment based on dual-indicator fusion A dual-index fusion algorithm for detecting and classifying round particles is employed. Particles are classified as round particles when the isoperimetric inequality roundness value is greater than or equal to a first threshold and the least-squares circle fitting roundness value is greater than or equal to a second threshold. (Refer to...) Figure 4 and Figure 5 This invention combines two complementary circularity indices—circularity based on isocircularity inequality and circularity based on least squares circle fitting—for comprehensive judgment, effectively solving the problem of high misjudgment rate with a single index. The specific calculation steps are as follows: Isoperimetric inequalities and circularity calculations The circularity value of the isopecoid inequality is calculated based on the area and perimeter of the particle profile. Based on the principle of the isopecoid inequality in plane geometry (where the area of a circle is maximized when the perimeter is fixed), the circularity value of the isopecoid inequality is defined as follows:
[0034] The value range of this indicator is (0,1], when C iso When the value is 1, the graphic is a perfect circle. The closer the value is to 1, the closer the graphic is to a circle.
[0035] Least squares circle fitting circularity calculation The least squares circle fitting circularity value is calculated based on the ratio of the root mean square error of the particle profile points to the radius of the fitted circle. The optimal fitted circle is solved using SVD decomposition to quantify the deviation of the profile from the ideal circle. Calculate the centroid of the contour points: , ; Coordinate translation: , Move the centroid to the origin of the coordinate system; Construct a system of linear equations Ax=b, where A is an n×3 matrix and the i-th row is [2u i ,2v i [,1]; b is an n×1 vector, and the i-th element is ; SVD decomposition solution: A = UΣV T The least squares solution is x = VΣ -1 U T b, obtain parameters a, b, c; Calculate the parameters of the fitted circle: center ,radius ; Calculate the root mean square error: ,in ; Calculate the least squares circularity value: The value range is [0,1]. When the calculation result is less than 0, the value is 0.
[0036] Circle-like determination rules Reference Figure 6 If and only if C iso ≥T iso And C ls ≥T ls If the particle is in a spherical state, it is determined to be a near-spherical particle; otherwise, it is determined to be a non-spherical particle. In this embodiment, the default threshold T is... iso =0.8, T ls =0.85, which can be dynamically adjusted through the system interface according to different application scenarios.
[0037] S4: Quadtree spatial index optimizes nearest-neighbor granular queries A quadtree spatial index is constructed to optimize neighbor-to-neighbor particle queries, calculating the minimum distance between contours only for candidate particles. (See reference...) Figure 10 To address the high time complexity (O(n^2)) of traditional brute-force search for neighboring particles, 2 To address the problem of [missing information], this invention uses a quadtree spatial index to reduce the time complexity to O(n log n). The specific process is as follows: Construction of quadtree Initialize the root node covering the entire image and calculate the axis-aligned bounding rectangle of each particle; if the number of particles in a node exceeds the preset maximum capacity of a single node, then subdivide it into four quadrant child nodes and redistribute the particles.
[0038] Initialize the root node, whose boundary covers the entire image area: Boundary root =[0,W]×[0,H], where W is the image width and H is the image height; Align the circumscribed rectangle (AABB) along the computational axis for each particle: AABB i =[min(x),max(x)]×[min(y),max(y)], and calculate the center of the circumscribed rectangle (c x ,c y ); Recursive particle insertion: If the number of particles in the current node exceeds the preset maximum capacity of a single node (10 in this embodiment), the node is subdivided into four equal child nodes: top left (NW), top right (NE), bottom left (SW), and bottom right (SE), and the particles are redistributed to the corresponding child nodes. For particles whose circumscribed rectangle crosses the node boundary, they are simultaneously inserted into all intersecting child nodes, or the boundary search of adjacent nodes is expanded during the query. Appendix Figure 10 The “MAX_CONTOURS_PER_NODE” marked in the figure is the preset maximum capacity of a single node as described in this embodiment, and is a simplified symbol used to identify this parameter in the attached figure.
[0039] Nearest Particle Query Define a search radius for the current particle, and recursively query and collect candidate particles within nodes intersecting the search range. The formula for calculating the search radius is as follows: Where width and height are the width and height of the bounding rectangle of the particle, d threshold This is the contact distance threshold.
[0040] Define the search radius: , where d threshold The contact distance threshold is set to 3 pixels in this embodiment. Recursively query from the root node. If the boundary rectangle of a node intersects with the search range, collect all particles in that node as candidate particles. Only the minimum contour distance between the current particle and candidate particles is calculated; there is no need to calculate the distance with all particles in the image.
[0041] S5: Multi-dimensional comprehensive occlusion detection A comprehensive occlusion detection method is used, based on edge truncation, indentation degree, and contact with adjacent particles. If a particle is truncated by an edge, or simultaneously satisfies the condition that its indentation degree is greater than a threshold and there is contact with adjacent particles, it is determined to be an occluded particle. (Refer to...) Figure 7 , Figure 8 and Figure 9 This invention comprehensively determines the occlusion state from three dimensions: edge truncation, indentation degree, and contact with adjacent particles, thus solving the problem of high false negative rates in single-dimensional detection. The specific determination rules are as follows: Edge truncation detection If there exists any point (x) on the particle profile i ,y i If any of the following conditions are met, the particle is determined to be truncated by the image edge: or or or
[0042] Where m is the inner margin of the boundary (set to 5 pixels in this embodiment), used to exclude incomplete particles near the image boundary.
[0043] Concavity calculation The degree of concavity reflects the extent of the contour's depression; occluded particles typically produce a noticeable depression. The calculation formula is:
[0044] Where A contour Let A be the area of the outline. hull The convex hull area is the area of the contour. The convex hull is calculated using the Graham scan algorithm, and then the area of the convex hull is calculated using the Shoelace formula. The concavity value ranges from [0,1], where D=0 indicates a completely convex shape, and D>0.3 indicates the presence of significant concavity.
[0045] Proximity particle contact detection Calculate the minimum Euclidean distance between the current particle and the candidate particle profiles:
[0046] Where p i q represents the current particle profile point. j These are candidate particle contour points. If d min <d threshold If so, it is determined that the two particles are in contact.
[0047] Comprehensive occlusion determination Reference Figure 11 The final occlusion determination rule is as follows:
[0048] In this embodiment, the indentation threshold D threshold =0.3.
[0049] S6: Output of recognition results Output the ID, roundness determination result, occlusion status determination result, and corresponding feature value for each particle. Output complete information for each particle, including: particle ID, roundness determination result (round / non-round), occlusion status determination result (occluded / not occluded), and roundness value C according to the isoperimetric inequality. iso Least squares circularity value C ls Indentation value D, contour area, perimeter, circumscribed rectangle size, etc.
[0050] II. Implementation Examples of a Particle Image Circle Recognition and Integrated Occlusion Recognition System Reference Figure 12 and Figure 13 The system described in this embodiment is an integrated particle quality detection system, which integrates the entire process from image acquisition to data statistics. The specific module composition is as follows: Image segmentation module: Used to output particle label masks and extract contour geometric features. It achieves automatic particle segmentation based on a pre-trained Cellpose deep learning model, outputting label masks and extracting the geometric features of each particle.
[0051] Circle Detection Module: Used to execute a circle determination algorithm based on dual-index fusion. It includes a perimeter inequality calculation unit, a least-squares circle fitting unit, and a circle determination unit, which respectively perform the calculation and comprehensive determination of two circleness indices.
[0052] The quadtree index module is used to build a spatial index and query neighboring particles. It is responsible for building the quadtree spatial index and performing neighboring particle queries, providing an efficient set of candidate particles for occlusion detection.
[0053] Occlusion detection module: Used to perform comprehensive occlusion determination based on edge truncation, indentation degree, and proximity contact. It includes an edge truncation detection unit, an indentation degree calculation unit, a proximity contact detection unit, and a comprehensive determination unit, which perform occlusion detection in three dimensions and output the final determination result.
[0054] Output module: Used to output the identification results for each particle. The identification results and statistical data are output to the display interface or local storage device.
[0055] Furthermore, the system also includes: Camera acquisition module: Used for real-time acquisition of particle images and supports dynamic adjustment of aperture, shutter speed, and white balance parameters. It employs a Canon EOS series industrial camera to acquire particle images in real time, supporting dynamic adjustment of parameters such as aperture, shutter speed, white balance, and exposure compensation. Image quality can be optimized according to different particle types and lighting conditions.
[0056] Calibration and Measurement Module: Used to calculate the millimeter / pixel ratio for accurate particle size measurement. Camera calibration is performed using a 9×9 checkerboard calibration plate, calculating the image's millimeter / pixel ratio (mm / pixel), converting pixel dimensions into actual physical dimensions, and enabling precise measurement of parameters such as particle diameter and area.
[0057] The data statistics module is used to generate histograms, calculate pass rates, and export data from the recognition results. It performs statistical analysis on the recognition results, including generating histograms of roundness distribution, calculating occlusion rates, and performing pass rate calculations. It supports exporting statistical data to Excel or CSV formats. Real-time preview of the recognition results is also supported, with each particle's ID, roundness status, and occlusion status labeled on the image. All threshold parameters can be dynamically adjusted through a graphical interface.
[0058] III. Experimental Verification To verify the technical effectiveness of this invention, a standard image dataset containing 895 particles was used for testing. The experimental results are as follows: Accuracy of circle detection: The accuracy of the single isocircularity inequality method is 85%, the accuracy of the single least squares circle fitting method is 80%, and the accuracy of the dual index fusion method of this invention reaches 96%, which is more than 11% higher than the accuracy of the single isocircularity inequality method.
[0059] Occlusion detection accuracy: The accuracy of the single indentation method is 82%, while the accuracy of the comprehensive occlusion detection method of this invention reaches 94%, which is an improvement of 12%.
[0060] Processing speed: The traditional brute-force search for neighboring particles takes about 96 seconds, while the quadtree optimization method of this invention takes about 2 seconds, which is 98% faster and fully meets the requirements of real-time detection.
[0061] IV. Optional Embodiments The scope of protection of this invention is not limited to the above embodiments. The following modifications and variations can be made without departing from the core idea of this invention: Deep learning segmentation models can be replaced by other models suitable for granular segmentation, such as U-Net and Mask R-CNN; The Jarvis step algorithm can be used to replace the Graham scan algorithm for convex hull calculation; Least squares solutions can be obtained by using QR decomposition, normal equation method, etc., instead of SVD decomposition; The maximum load capacity of a single node in a quadtree can be adjusted between 5 and 20 based on the particle density.
[0062] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.
Claims
1. A method for determining the circularity of particles in images and for comprehensive occlusion recognition, characterized in that, Includes the following steps: S1: Obtain the particle label mask output by the deep learning segmentation model, with each particle corresponding to a unique ID; S2: Extract the outline coordinates, area, perimeter, width and height of the bounding rectangle, and aspect ratio of each particle. S3: A circular particle detection and determination algorithm with dual index fusion is adopted. When the circularity value of the isoperimeter inequality of the particle is greater than or equal to the first threshold and the circularity value of the least squares circle fitting is greater than or equal to the second threshold, the particle is determined to be a circular particle. S4: Optimize neighbor particle queries by constructing a quadtree spatial index, and calculate the minimum distance between contours only for candidate particles; S5: Based on edge truncation, concavity and contact with neighboring particles, a comprehensive occlusion detection is performed. If a particle is truncated by the image edge, or if the concavity is greater than the third threshold and there is contact with neighboring particles, then the particle is determined to be an occluded particle. S6: Output the ID, roundness determination result, occlusion state determination result, and corresponding feature value of each particle.
2. The identification method according to claim 1, characterized in that, In step S3: the circularity value of the isocircular inequality is calculated based on the area and perimeter of the particle profile; the circularity value of the least squares circle fitting is calculated based on the ratio of the root mean square error of the particle profile point to the optimal fitted circle to the radius of the fitted circle.
3. The identification method according to claim 2, characterized in that: The formula for calculating the circularity value of the isoperimetric inequality is as follows: Where A is the particle outline area and P is the particle outline perimeter; the formula for calculating the least squares circle fitting circle value is: RMS error denoted as the root mean square error from the contour point to the fitted circle, and r as the radius of the fitted circle.
4. The identification method according to claim 1, characterized in that, The quadtree construction and query process in step S4 is as follows: Initialize the root node covering the entire image, and calculate the axis-aligned bounding rectangle of each particle; if the number of particles in a node exceeds the preset maximum capacity of a single node, it is subdivided into four quadrant child nodes and the particles are redistributed; for particles whose bounding rectangle crosses the node boundary, they are simultaneously inserted into all intersecting child nodes, or the boundary of adjacent nodes is expanded during the query; define the search radius for the current particle, recursively query and collect candidate particles in nodes that intersect with the search range.
5. The identification method according to claim 4, characterized in that, The formula for calculating the search radius is as follows: Where width and height are the width and height of the bounding rectangle of the particle, d threshold This is the contact distance threshold.
6. A system for determining the circularity of particles in images and for comprehensive occlusion recognition, characterized in that, include: The image segmentation module is used to output particle label masks and extract contour geometric features; The circle detection module is used to execute the circle determination algorithm based on the fusion of two indicators; The quadtree index module is used to build a spatial index and query neighboring particles; the occlusion detection module is used to perform a comprehensive occlusion determination based on edge truncation, concavity, and proximity contact. The output module is used to output the identification results for each particle.
7. The identification system according to claim 6, characterized in that, Also includes: The camera acquisition module is used to acquire grain images in real time and supports dynamic adjustment of aperture, shutter speed, and white balance parameters. The calibration and measurement module is used to calculate the millimeter / pixel ratio to achieve accurate particle size measurement; the data statistics module is used to draw histograms of the recognition results, calculate the pass rate, and export the data.
8. The identification system according to claim 6, characterized in that, The image segmentation module achieves automatic particle segmentation based on the Cellpose deep learning model.
9. The identification system according to claim 6, characterized in that, The circle-like detection module includes an isocircular inequality calculation unit, a least squares circle fitting unit, and a circle-like determination unit.
10. The identification system according to claim 6, characterized in that, The occlusion detection module includes an edge truncation detection unit, a concavity calculation unit, a proximity contact detection unit, and a comprehensive judgment unit.