A method and system for detecting the granularity of fertilizer based on image processing

Abstract

The present application relates to the technical field of image processing, more particularly, the present application relates to a kind of based on image processing's fertilizer granularity detection method and system, comprising: identifying cluster region in image, the cluster region is the connected domain of area greater than preset area threshold and convex hull area ratio less than preset non-convexity threshold, extract the edge profile of the cluster region, and based on the multiscale curvature difference response of the edge profile obtains true concave point set;Cutting obstruction potential energy value is constructed inside the cluster region, and the cutting obstruction potential energy value is positively correlated with the distance value of pixel point to cluster region background boundary.The present application combines topological affinity and global maximum weight matching algorithm with non-intersection constraint, effectively solves the problem that path passes through granular entity and segmentation line intersection in complex ring-shaped stacked structure, significantly improves the accuracy of online detection.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for detecting fertilizer particle size based on image processing. Background Technology

[0002] In modern chemical production, online monitoring of fertilizer particle size distribution is a crucial step in ensuring product quality. Currently, analyzing images of particles on a conveyor belt is the mainstream non-contact detection method. However, in real-world high-density online detection scenarios, particles are prone to random adhesion and stacking, forming clusters of varying shapes. This makes accurately separating individual particles from the images and statistically analyzing particle size information a significant technical challenge.

[0003] Existing technologies typically employ segmentation algorithms based on concave point geometric features to handle adhered particles. The main idea of ​​this type of method is to identify concave points (i.e., depressions) on the contour of the adhered region, using them as potential starting and ending points for segmentation. Subsequently, shortest path search or geometric matching algorithms are used to connect the pairs of concave points, thereby decomposing the adhered region into independent objects.

[0004] However, the above methods have obvious limitations when applied to fertilizer particle detection: on the one hand, fertilizer particle surfaces are usually rough and have irregular textures, and concave point detection based on traditional local curvature calculation is easily affected by high-frequency texture noise, resulting in the identification of a large number of false concave points and causing oversegmentation problems; on the other hand, when faced with complex ring-shaped stacked structures composed of three or more particles, simple concave point pairing and path planning often lack effective constraints on the physical morphology of the particles and the overall topology, which can easily produce physically infeasible segmentation results such as segmentation paths passing through the particle entity and segmentation lines intersecting each other, which seriously affects the accuracy of detection. Summary of the Invention

[0005] This invention provides a fertilizer particle size detection method and system based on image processing, aiming to solve the problem of cutting decision errors under multi-particle cluster stacking in related technologies.

[0006] In a first aspect, the present invention provides a fertilizer particle size detection method based on image processing, comprising: identifying clustered regions in an image, wherein the clustered regions are connected components with an area greater than a preset area threshold and a convex hull area ratio less than a preset non-convexity threshold; extracting the edge contours of the clustered regions; and obtaining a set of true concave points based on the multi-scale curvature difference response of the edge contours; constructing a cutting obstacle potential energy value within the clustered regions, wherein the cutting obstacle potential energy value is positively correlated with the distance from a pixel to the background boundary of the clustered regions; and between any pair of concave points in the set of true concave points, The minimum potential energy geodesic path is obtained by planning based on the cutting obstacle potential energy value; a topological weighted graph is constructed, with the true concave points as graph nodes and the topological affinity of the concave point pairs as the weight of the corresponding graph edges; the topological affinity is proportional to the number of local particles connected by the concave point pairs and inversely proportional to the minimum cumulative obstacle of the minimum potential energy geodesic path; maximum weight matching with topological non-crossing constraints is performed on the topological weighted graph to obtain the optimal cutting edge set; the clustered region is segmented based on the optimal cutting edge set, and the particle size parameters of the segmented independent particles are analyzed. Multi-scale curvature difference response effectively suppresses texture noise and accurately locates the true concave points; simultaneously, a cutting obstacle potential energy field based on distance transformation and the minimum potential energy geodesic path are introduced to ensure that the cutting line travels along the weak neck of the particle adhesion, avoiding cutting off the particle body; combined with the maximum weight matching algorithm with topological non-crossing constraints, the segmentation problem of complex ring-shaped stacked structures composed of three or more particles can be solved from a global perspective, significantly improving the segmentation accuracy and particle size detection precision in high-density adhesion scenarios.

[0007] Furthermore, the method for obtaining the multi-scale curvature difference response of the edge contour includes: calculating the local curvature of each pixel on the edge contour to obtain a first curvature value at a first scale and a second curvature value at a second scale, wherein the second scale is greater than the first scale; and using the difference between the first curvature value and the second curvature value as the multi-scale curvature difference response. This method utilizes a differential response mechanism to enhance the realistic geometric concave features while smoothing the curvature fluctuations caused by micro-textures, thereby significantly reducing the false detection rate of false concave points caused by irregular textures on the fertilizer surface without destroying the overall morphology of the particles.

[0008] Furthermore, this includes: constructing a statistical distribution of the multi-scale curvature difference response of all pixels on the edge contour; the set of true concave points includes pixels with response values ​​less than an adaptive threshold determined according to the statistical distribution. Compared to using a fixed threshold to filter concave points, using an adaptive threshold determined based on the statistical distribution can dynamically adjust the filtering criteria according to the basic texture noise level and dispersion of the current batch of images. This makes the algorithm more robust and adaptable to different lighting conditions and surface characteristics of different batches of fertilizer.

[0009] Furthermore, the method for constructing the cutting obstacle potential energy value includes: performing an Euclidean distance transformation on the interior of the clustered region to obtain the distance value from each pixel to the background boundary of the clustered region; the cutting obstacle potential energy value is a power function of the distance value. By constructing a potential energy field related to the distance to the background boundary through the Euclidean distance transformation and defining the cutting obstacle potential energy as a power function of the distance, a high potential energy (i.e., high obstacle) is formed in the particle center region, while a low potential energy is formed in the adhesion neck region. This physical constraint mechanism forces the subsequently planned cutting path to automatically tend to avoid the thick parts of the particles, thereby ensuring the rationality of the segmentation line.

[0010] Furthermore, the planning method for the minimum potential energy geodesic path includes: solving the Eikonal equation using the fast travel method to generate an arrival time field starting from the first concave point in the concave point pair; and backtracking from the second concave point in the concave point pair along the negative gradient direction of the arrival time field to obtain the minimum potential energy geodesic path. Using the fast travel method to solve the Eikonal equation to generate the arrival time field and backtracking along the negative gradient direction can mathematically rigorously solve for the globally optimal minimum potential energy geodesic path. Compared to heuristic search or simple straight-line connections, this method generates a smoother and more natural path that strictly adheres to the physical constraints of the potential energy field. It is not only computationally efficient but also produces a path quality that better matches the geometric characteristics of natural particle adhesion.

[0011] Furthermore, the minimum cumulative obstacle of the minimum potential energy geodesic path is obtained by integrating the potential energy value of the cutting obstacle potential energy field along the minimum potential energy geodesic path. By integrating the potential energy value along the minimum potential energy geodesic path, the minimum cumulative obstacle is obtained, quantifying the cost of each potential cutting path. This provides a reliable physical basis for subsequent topology matching, enabling the algorithm to quantitatively distinguish between high-cost erroneous paths that forcibly cut through particle entities and low-cost correct paths that separate along adhesion points.

[0012] Furthermore, the topological non-intersection constraint includes prohibiting pixel-level intersections between the currently matched cutting path and the determined cutting path. This ensures that the generated segmentation scheme is topologically valid when dealing with complex multi-particle stacks, avoiding logical conflicts and further improving the accuracy of the segmentation results.

[0013] Furthermore, before identifying clustered regions in the image, the process includes: acquiring an image containing multiple particles and binarizing it to obtain a binary image; performing connected component analysis on the binary image to obtain multiple connected components. Through dual constraints of area and convex hull area ratio, the system quickly filters out the clustered, adherent regions that truly require segmentation from complex images, automatically filtering out single particles and noise interference. This not only reduces the computational load of subsequent complex calculations but also improves the overall detection speed of the system.

[0014] Furthermore, the first scale is 3 to 5 pixels, and the second scale is 15 to 20 pixels.

[0015] In a second aspect, the present invention also provides a fertilizer particle size detection system based on image processing, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the fertilizer particle size detection method based on image processing described above.

[0016] Beneficial effects: Multi-scale curvature difference response effectively eliminates pseudo-concave interference caused by rough texture of particle surface; a cutting obstacle potential energy field based on distance transformation is constructed to plan the minimum potential energy geodesic path that conforms to physical morphological constraints; combined with topological affinity and global maximum weight matching algorithm with non-crossing constraints, the problem of paths passing through particle entities and intersecting dividing lines in complex ring stacked structures is effectively solved, significantly improving the accuracy of online detection. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating a particle size detection flowchart according to an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram illustrating the comparison of concave dot screening according to an embodiment of the present invention;

[0019] Figure 3 This is a schematic diagram illustrating a global decision comparison flowchart according to an embodiment of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] like Figure 1 As shown, S101: Acquire images and perform preprocessing, and identify clustered regions.

[0022] In one embodiment, a multi-directional illumination image sequence of the target scene is first acquired, and a high-quality grayscale image is obtained after multi-source fusion. The target scene is an online detection scene of high-density solid particles. Gaussian smoothing is performed on the image to remove high-frequency noise, and then binarization is performed using the OTSU adaptive thresholding method to obtain a primary particle distribution map.

[0023] Connectivity analysis is performed on the primary particle distribution map to calculate the area and convex hull area of ​​each connected region. To quickly locate and centrally process resources from massive amounts of data, this embodiment employs a filtering mechanism based on size and shape constraints. Specifically, the convex hull area ratio of each connected region is defined as the ratio of the area of ​​that connected region to the area of ​​its convex hull.

[0024] During the screening process, when a connected component simultaneously meets the following two conditions, the system marks it as a clustered region: the first condition is that its area is greater than a set threshold. The set threshold is slightly larger than the area of ​​the largest single particle; the second condition is that the convex hull area ratio is less than a non-convexity threshold. In this embodiment, The preferred setting is a high value close to 1, specifically 0.95.

[0025] As can be seen from the convex hull area ratio, when the connected region is an ideal convex polygon, such as an independent circular or elliptical particle, the area of ​​the connected region is infinitely close to the area of ​​the convex hull, and the convex hull area ratio is approximately equal to 1. When the convex hull area ratio is less than 0.95, it indicates that there is a significant depression at the edge of the connected region, that is, the area of ​​the connected region is significantly smaller than the area of ​​the convex hull. This is usually caused by gaps or neck structures formed by the adhesion of two or more particles. Therefore, by setting a non-convexity threshold, single particles (highly convex) and adherent clusters (non-convex) can be effectively distinguished, avoiding unnecessary segmentation calculations for single particles, thereby significantly improving detection efficiency.

[0026] S102: Multi-scale curvature analysis and adaptive screening of true concave points.

[0027] After identifying the clustered regions, it is necessary to extract their edge contours and identify true concave points for segmentation. In this embodiment, multi-scale curvature difference semantic scoring is used to evaluate the concave point intensity for surface textures and pseudo-concave points present in the high-resolution acquired images. Any pixel on the contour is selected. Local curvature is calculated at both small and large scales. The small scale is preferably... Pixels, large scale preferred Pixel.

[0028] The semantic score for concave points is constructed, and its calculation formula is as follows: In the formula, Represents pixels The semantic score of the concave point; Represents pixels On a small scale The local curvature under the surface is mainly used to capture high-frequency information such as texture; Represents pixels Large scale The local curvature is mainly used to capture low-frequency information such as structure.

[0029] From this formula, we can see that the numerator of the formula... This reflects the change in curvature from the micro-texture scale to the macro-structural scale. For pseudo-concave points, they will be smoothed out under large-scale smoothing, leading to… The difference is small, but for true concave points, i.e., where particles adhere, their geometry remains persistent across different scales. To maintain stability, the difference is kept relatively large. The denominator is then used for normalization to obtain a relative rate of change that is independent of the absolute size of the concave point.

[0030] To accommodate different noise levels in different batches of images, an adaptive statistical model is used to determine the screening threshold. A semantic score distribution of all concave points is constructed, and its mean is calculated. with standard deviation The mean represents the basic texture noise level, and the standard deviation represents the dispersion of the concave point semantic score distribution. Then, an adaptive threshold is determined. The calculation formula is: In the formula, For statistical confidence range coefficients, the preferred empirical value is... .

[0031] As can be seen from the formula, the threshold It dynamically follows the changes in the underlying texture noise level and dispersion of the current batch of images. When the concave point is identified as a true concave point, a set of true concave points is formed. It's important to note that, based on the curvature direction definition, the curvature at the concave point is typically negative or a minimum value. If curvature is defined as positive, the sign is opposite. In this embodiment, a smaller curvature value indicates a deeper concave point; therefore, concave points below a threshold are considered true concave points. Using this method, true concave points can be robustly separated from the background texture, ensuring the accuracy of subsequent topology analysis input.

[0032] Step S103: Calculate the cutting obstacle potential energy value and obtain the geodesic path with minimum potential energy.

[0033] Based on the set of true concave points, a cutting obstacle potential energy value is constructed to plan the optimal cutting path. First, an Euclidean distance transformation is performed on the interior of the clustered region to obtain a distance map. , representing the distance from each pixel within the clustered region to the nearest background boundary. The cutting obstacle potential energy value of each pixel within the clustered region is calculated, expressed in pixels. For example, the calculation formula is as follows: In the formula, Represents pixels The cutting resistance potential energy value; For pixels The corresponding Euclidean distance transformation value; For extremely small positive numbers, such as Used to prevent The time value is unstable; As an exponential factor, The value of is 2.

[0034] Based on the cutting obstacle potential energy value of each pixel, the Eikonal equation is constructed for any pair of concave points in the set of true concave points: The arrival time field is generated by solving this equation using the Fast Marching Method (FMM). Recorded from true concave point (Starting point) The minimum time required to travel along the path of least resistance. From a true concave point Starting from the (end point), along the negative gradient direction of the time field. Backtracking to obtain the geodesic path with minimum potential energy. The formula for calculating the minimum potential energy geodesic path is: ; . Represents the velocity vector, defining the path direction for backtracking along the fastest descent direction (negative gradient) in the time field.

[0035] Finally, by integrating the hindering potential energy value along the geodesic path of minimum potential energy, the minimum cumulative physical obstacle is obtained, and its calculation formula is as follows: In the formula, The minimum cumulative resistance represents the minimum potential energy geodesic path, and the resistance required to cut this point is evaluated. This indicates reading the geodesic path with minimum potential energy. Upper The cutting resistance potential energy value at the coordinate point; then... Normalization is performed to serve as the physical basis for subsequent decisions.

[0036] S104: Construct topological affinity and constrained global maximum weight matching.

[0037] To address the topological ring decomposition problem in complex clustered stacks (three or more particles), this embodiment integrates minimum geodesic energy with topological importance. First, the topological fraction of any pair of concave points is calculated, using the concave point pair... For example, the calculation formula is: In the formula; Indicates a pair of concave points topological fractions Indicates a concave point and In the topological map of a clustered region, the local degree of the connection, i.e. the total number of connected particles, indicates that the higher the value, the more critical the cutting line is to breaking down complex ring structures.

[0038] Next, using the concave point to... For example, the topological affinity of this concave point pair is calculated using the following formula: In the formula, Indicates a pair of concave points The topological affinity, which is the weight subsequently used as a graph edge; Indicates a pair of concave points Topological fractions; The minimum cumulative obstacle after normalization; For balance coefficient, Preferred The value of is 2.

[0039] As can be seen from the formula, the denominator It serves as a physical constraint. If a path passes through a granular entity, its... Approaching 1. When When the value is 2, the denominator becomes 3, causing the topological affinity to decrease significantly to its original value. Thus, it is suppressed in maximum weight matching. If a path travels along the adherent neck, its As the denominator approaches 0, the affinity is primarily determined by the topological score, and the algorithm prioritizes critical paths that effectively deconstruct topological loops. Therefore, this formula achieves a dynamic balance between prioritizing critical topological paths and prohibiting physically infeasible paths.

[0040] Finally, treating concave points as graph nodes and topological affinity as the weights of graph edges, a topologically weighted graph is constructed. Introducing a topological non-crossing constraint: In each pairing iteration, the system checks the current cutting path to be matched. Is it related to any path in the established set of cutting paths? There is pixel-level intersection between the two cutting paths. and If any pixels are shared, then the matching pair It was deemed illegal and excluded.

[0041] The constrained maximum weight matching problem is solved using the improved Blossom algorithm, yielding the following results: : In the formula, Let represent the optimal set of cutting edges, and let represent the final set of maximum weighted matching edges with no intersections. This represents the sum of affinity of all cutting edges in the matching set, and maximizing this value is the solution objective; This represents the optimal choice operator. This set... This is the final concave point pairing scheme, representing the cutting pairs that should be executed under the premise of global optimization and correct topology.

[0042] S105: Perform segmentation and granularity statistics.

[0043] Based on the determined final set of cutting edges Along the path of the minimum potential energy geodesic Perform pixel-level segmentation to decouple the clustered region into independent sub-connected regions. Perform ellipse fitting on each decoupled independent sub-connected region (i.e., independent particle), calculate and output standardized indices such as major axis, minor axis, and equivalent particle size, and generate the final particle size statistics report.

[0044] like Figure 2 As shown in the figure, the left image illustrates how traditional local geometry methods, when processing particle boundaries, easily confuse pseudo-concave points with true concave points due to high-frequency textures or noise. The different markings in the figure highlight the shortcomings of traditional methods at the input stage, which can easily lead to incorrect cutting. The right image describes how the present invention, through multi-scale curvature analysis combined with an adaptive statistical model, successfully eliminates pseudo-concave point interference, accurately identifies and retains only the set of true concave points representing genuine adhesion, ensuring that subsequent decision-making processes are based on highly reliable input.

[0045] like Figure 3 As shown in the figure, the left side illustrates how traditional local pairing decisions, lacking global topological constraints, lead to topological errors or insufficient cutting in complex clustered stacks of three or more particles. This is illustrated by the red dashed line, revealing the fundamental flaw in decision-making failure when dealing with complex structures. The yellow dashed line in the figure has the same meaning as the red dashed line, both representing erroneous pairing decisions caused by the lack of global constraints in existing technologies. The right side illustrates how this invention, by constructing a topological affinity function A and solving for the maximum weighted matching with topological non-intersecting constraints, successfully obtains a globally optimal set of cutting edges G with correct topology and no intersections. This is illustrated by the green solid line, demonstrating that this method can effectively solve the decision-making error problem in complex clustered stacks.

[0046] The present invention also provides a fertilizer particle size detection system based on image processing. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements the fertilizer particle size detection method based on image processing according to the first aspect of the present invention.

[0047] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0048] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0049] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for detecting the granularity of fertilizer based on image processing, characterized by, include: Identify clustered regions in an image. Clustered regions are connected components whose area is greater than a preset area threshold and whose convex hull area ratio is less than a preset non-convex threshold. Extract the edge contours of the clustered regions and obtain a set of true concave points based on the multi-scale curvature difference response of the edge contours. Concave point strength is evaluated using multi-scale curvature difference semantic scoring; any pixel on the contour is selected. Local curvature was calculated at both small and large scales. Constructing concave point semantic scores: ; For pixels Concave semantic score; For pixels On a small scale Local curvature below; For pixels Large scale Local curvature below; Construct the semantic score distribution of all pits, and calculate the mean value thereof and the standard deviation The mean value represents the basic texture noise level, and the standard deviation represents the dispersion degree of the pit semantic score distribution; Determine the adaptive threshold : ; For statistical confidence range coefficients; When the concave point is determined as a true concave point, a true concave point set is formed; Construct the cutting resistance potential energy value within the clustered region. ; For pixels The cutting resistance potential energy value; For pixels The corresponding Euclidean distance transformation value; It is a very small positive number. As an exponential factor, the cutting obstacle potential energy value is positively correlated with the distance from the pixel to the background boundary of the clustered region; between any pair of concave points in the set of true concave points, a minimum potential energy geodesic path is planned based on the cutting obstacle potential energy value. ; ; It is the velocity vector; Construct a topologically weighted graph, with true concave points as graph nodes and the topological affinity of concave point pairs as the weights of the corresponding graph edges; the topological affinity is directly proportional to the number of local particles connected by the concave point pairs and inversely proportional to the minimum cumulative resistance of the minimum potential energy geodesic path. The optimal cutting edge set is obtained by performing maximum weight matching with topological non-intersection constraints on the topological weighted graph. The topological non-intersection constraints include: prohibiting the current matching cutting path from having pixel-level intersection with the determined cutting path; the clustered region is segmented based on the optimal cutting edge set, and the granularity parameters of the segmented independent particles are analyzed. Topological score for pairs of concave points: ; Topological score for pairs of concave points ; Local degree of concave points and connected in a cluster region topological map; Topological affinity of concave point pairs: ; Topological affinity of concave point pairs ; is the normalized minimum cumulative hindrance; is the balance coefficient.

2. The fertilizer particle size detection method based on image processing according to claim 1, characterized in that, The method for obtaining the multi-scale curvature difference response of the edge contour includes: The local curvature of each pixel on the edge contour is calculated to obtain a first curvature value at a first scale and a second curvature value at a second scale, wherein the second scale is greater than the first scale; the difference between the first curvature value and the second curvature value is used as the multi-scale curvature differential response.

3. The fertilizer particle size detection method based on image processing according to claim 2, characterized in that, include: Construct the statistical distribution of the multi-scale curvature difference response of all pixels on the edge contour; The set of true concave points includes pixels whose response values ​​are less than an adaptive threshold determined based on statistical distribution.

4. The image processing-based fertilizer granule size detection method of claim 1, wherein, The method for constructing the cutting resistance potential energy value includes: Perform an Euclidean distance transformation on the interior of the clustered region to obtain the distance value from each pixel to the background boundary of the clustered region; the cutting obstacle potential energy value is a power function of the distance value.

5. The image processing-based fertilizer granule size detection method of claim 1, wherein, The method for planning the minimum potential energy geodesic path includes: The Eikonal equation is solved using the fast travel method to generate the arrival time field starting from the first concave point in the concave point pair; the minimum potential energy geodesic path is obtained by tracing back from the second concave point in the concave point pair along the negative gradient direction of the arrival time field.

6. The image processing-based fertilizer granule size detection method of claim 1, wherein, The minimum cumulative resistance of the minimum potential energy geodesic path is obtained by integrating the potential energy value of the cutting resistance potential energy field along the minimum potential energy geodesic path. 7.The image processing-based fertilizer granule size detection method according to claim 1, wherein, Before identifying clustered regions in an image, the following steps are also included: An image containing multiple particles is acquired and binarized to obtain a binary image. Connectivity analysis is then performed on the binary image to obtain multiple connected components. 8.The image processing-based fertilizer granule size detection method according to claim 2, characterized in that, The first scale is 3 to 5 pixels, and the second scale is 15 to 20 pixels.

9. An image processing based fertilizer granule size detection system comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the image processing-based fertilizer particle size detection method as described in any one of claims 1-8.

Images