A method for detecting the particle size of compound fertilizer granules

Abstract

The application discloses a composite fertilizer particle size detection method, relates to the technical field of composite fertilizer detection, and can more comprehensively reflect the size, shape and uniformity characteristics of particles by extracting the morphological complexity coefficient of particles and the relative standard deviation of particle size distribution, thereby providing multi-dimensional data support for composite fertilizer quality evaluation. On this basis, a particle scoring model is constructed by combining deep learning technology, the qualified evaluation index Hgzs is obtained in the input model, and a detection instruction is issued according to a preset threshold value, so that high-precision particle quality discrimination is realized, and manual intervention and human errors are effectively reduced. In addition, the method designs a secondary response detection evaluation mechanism, further analysis is carried out based on the results of the primary detection, and the detection rigor and the stability of the quality control of the composite fertilizer production are ensured.

Description

Technical Field

[0001] This invention relates to the field of compound fertilizer testing technology, specifically a method for detecting the particle size of compound fertilizer particles. Background Technology

[0002] Particle size testing of compound fertilizer granules falls under the field of particle characterization and measurement, specifically involving the analysis of the three-dimensional morphology, particle size distribution, and complexity of the particles. In the production and quality control of compound fertilizers, the particle size uniformity and morphological structure directly affect the fertilizer's solubility, release rate, and absorption efficiency of its active ingredients. Unlike general granular products, compound fertilizer granules have multiple components and complex structures. Therefore, particle size testing requires comprehensive consideration of factors such as particle size, shape, and distribution to ensure that the product meets the standards for different application scenarios.

[0003] Existing methods for particle size detection in compound fertilizer mainly rely on two-dimensional image analysis and traditional morphological description parameters. These methods have limitations when dealing with the complexity of particle morphology and combinations of multiple particle size characteristics. Due to the irregular shape of compound fertilizer particles, traditional two-dimensional analysis struggles to accurately obtain their true surface area and volume. Furthermore, the shape of compound fertilizer particles is typically irregular, potentially exhibiting spherical, elliptical, plate-like, or irregular polygonal forms. This diversity of shapes poses a challenge to detection technology because most particle size analysis equipment (such as laser particle size analyzers) assumes that the particles are regular spheres, resulting in an "equivalent diameter." For irregular particles, the equivalent diameter fails to accurately reflect their true size and surface area, especially when the particle shape is complex, leading to significant deviations in the detection results. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for detecting the particle size of compound fertilizer granules, thus solving the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting the particle size of compound fertilizer granules, comprising the following steps:

[0006] S1. Randomly capture the compound fertilizer particles to be detected, and capture two-dimensional projection images of the compound fertilizer particles to be detected from multiple angles through image acquisition equipment. After stereo reconstruction, a three-dimensional surface model is generated.

[0007] S2. Based on the three-dimensional surface model and combined with the two-dimensional projection image of the corresponding compound fertilizer particles to be detected, extract the relevant structural data set of the corresponding compound fertilizer particles to be detected. Based on the relevant structural data set, analyze the structural complexity of the compound fertilizer particles to be detected to obtain the morphological complexity coefficient Xfxs. Based on several groups of compound fertilizer particles to be detected, analyze the particle size state between the compound fertilizer particles to be detected to obtain the relative standard deviation of particle size distribution Xbc.

[0008] S3. Construct a particle scoring model using deep learning technology, and input the relative standard deviation of particle size distribution Xbc and the morphological complexity coefficient Xfxs into the particle scoring model to fit and output the pass assessment index Hgzs. Based on the comparison result between the pass assessment index Hgzs and the preset assessment threshold Q, issue the corresponding detection command.

[0009] S4. After receiving the detection instruction in S3, execute the secondary response detection and evaluation mechanism, formulate response conditions to generate secondary detection results, and adjust the process based on the secondary detection results.

[0010] Preferably, step S1 specifically includes:

[0011] S11. Deploy multi-view image acquisition equipment in advance, and randomly select several groups of compound fertilizer particles to be tested. After taking pictures of the compound fertilizer particles to be tested from multiple angles, obtain two-dimensional projection images of each compound fertilizer particle to be tested. After classification, obtain a two-dimensional set of each compound fertilizer particle to be tested. At the same time, according to the set position and equipment calibration of each image acquisition device during capture, obtain the intrinsic and extrinsic parameters of each image acquisition device. The intrinsic parameters include the focal length and optical center position of the image acquisition device, and the extrinsic parameters include the position and angle between the image acquisition devices.

[0012] Preferably, step S1 further includes:

[0013] S12. Based on the two-dimensional set of each compound fertilizer particle to be detected, find the corresponding feature points p between the two-dimensional projected images within the two-dimensional set. j To perform feature point p j The registration process is performed, and geometric relationships are used to identify the relative points from multiple viewpoints, i.e., the feature points p corresponding to the multiple viewpoints. j To obtain the corresponding feature points p of each compound fertilizer particle to be tested. j The location in three-dimensional space, and marked as three-dimensional point P. i Specifically, it is calculated and obtained as follows:

[0014]

[0015] In the formula, j = 1, 2, 3, ..., N; N represents the number of viewpoints; p j π represents the two-dimensional coordinates of the feature points extracted from the j-th image acquisition device viewpoint of the corresponding compound fertilizer granule to be detected; j δ is the projection matrix; j (P i ) is the projection function, which projects the three-dimensional point P. i Projected onto the two-dimensional projection image plane of the j-th image acquisition device; ||p j -δj (P i Let P be the 3D point from the perspective of the j-th image acquisition device. i The projection of the feature point p in the image and its two-dimensional coordinates j The error between the two points is the Euclidean distance between them. () represents finding the 3D point P that minimizes the total error. i Where i represents the sequence number of the three-dimensional point within the corresponding compound fertilizer particle to be detected;

[0016] S13. Obtain the 3D point P based on S12. i The method involves obtaining several sets of three-dimensional points P on the corresponding compound fertilizer granules to be tested. i And by combining several sets of three-dimensional points P i Connect the links to generate the three-dimensional surface of the corresponding compound fertilizer particles to be tested;

[0017] S14. Based on the obtained three-dimensional surface of the corresponding compound fertilizer particles to be tested, and combined with DEM software, perform three-dimensional simulation on each compound fertilizer particle to be tested to construct a three-dimensional surface model.

[0018] Preferably, step S2 specifically includes:

[0019] S21. Based on the three-dimensional surface model and combined with the two-dimensional projection image of the corresponding compound fertilizer particle to be detected, extract the relevant structural data set of the corresponding compound fertilizer particle to be detected. The relevant structural data set includes the projection perimeter, projection area Tmj, convex hull area Tbmj, major axis length Czc, minor axis length Dzc, particle size value Kjz, and volume of the corresponding two-dimensional projection image of each compound fertilizer particle to be detected.

[0020] S22. Based on the aforementioned relevant structural data set, analyze the complexity of the structure of each compound fertilizer particle to be tested, and after linear normalization, obtain the morphological complexity coefficient Xfxs of each compound fertilizer particle to be tested, specifically obtained in the following manner:

[0021]

[0022] In the formula, Xyz represents the shape factor, and Tyz represents the convexity factor. The shape regularity is represented by Pxz, the eccentricity is represented by α and β, both of which are weight values, and C is the first correction constant. The specific values ​​of α and β are set by the user according to the situation.

[0023] S23. Based on the morphological complexity coefficients Xfxs of each compound fertilizer particle to be tested obtained in S22, the average morphological complexity coefficient is obtained by summing them. The specific method is as follows: Where k = 1, 2, 3, ..., n; n represents the number of compound fertilizer granules randomly selected from S11 to be tested, Xfxs k Let be the morphological complexity coefficient of the k-th compound fertilizer particle.

[0024] Preferably, step S2 further includes:

[0025] S221, The shape factor Xyz is obtained by the following formula:

[0026]

[0027] In the formula, This represents the average projected perimeter of the corresponding compound fertilizer granule to be tested. denoted as the average projected area of ​​the corresponding compound fertilizer particle to be tested; the shape factor Xyz is used to represent the shape complexity of the corresponding compound fertilizer particle to be tested, and π represents pi.

[0028] S222, The convexity factor Tyz is obtained by the following formula:

[0029]

[0030] In the formula, Tmj represents the projected area; Tbmj represents the convex hull area, which is the smallest convex region surrounding the compound fertilizer granules.

[0031] S223, the eccentricity Pxz is obtained through the following formula:

[0032]

[0033] In the formula, Dzc represents the short axis length of the compound fertilizer particle to be tested, and Czc represents the long axis length of the compound fertilizer particle to be tested.

[0034] The phenomenon indicates that the closer the eccentricity Dzc is to 1, the flatter or longer the compound fertilizer particles are, and the greater the deviation from the spherical shape. If the eccentricity of the compound fertilizer particles is large, it may affect the bulk density and flowability of the particles, resulting in uneven flow during transportation and fertilization.

[0035] Preferably, step S2 further includes:

[0036] S24. Based on the random sampling of several groups of compound fertilizer particles to be tested in S11, analyze the particle size distribution among the compound fertilizer particles to be tested to obtain the relative standard deviation Xbc of the particle size distribution. The relative standard deviation Xbc of the particle size distribution is obtained in the following way:

[0037]

[0038] In the formula, n represents the number of compound fertilizer granules randomly selected from S11 to be tested, Kjz avg The average particle size of the compound fertilizer granules to be tested is represented by Kjz. k This represents the particle size value of the k-th compound fertilizer particle to be tested.

[0039] Preferably, step S3 specifically includes:

[0040] S31. A basic model is initially constructed using deep learning technology. Relevant constructed data sets are input into the basic model for training and testing. The trained basic model is then used as a recognition model. Feature information within the recognition model is acquired, and this acquired feature information is used to train and test the recognition model. The trained recognition model is then used as a particle scoring model. After training and linear normalization, the qualified assessment index Hgzs of all randomly selected compound fertilizer particles to be tested from S11 is fitted and output. Specifically, this is obtained through the following methods:

[0041]

[0042] In the formula, Xbc represents the average morphological complexity coefficient, G represents the relative standard deviation of particle size distribution, Tcc represents the volume difference value, and a, b, and c are all weight values. The specific values ​​of a, b, and c are set by the user according to the situation.

[0043] Preferably, step S3 further includes:

[0044] S32. A pre-set evaluation threshold Q is used to compare and analyze the evaluation threshold Q with the qualified evaluation index Hgzs to determine whether the randomly selected groups of compound fertilizer granules to be tested in the current step S11 are in a qualified state. The specific judgment content is as follows:

[0045] If the qualified assessment index Hgzs ≤ the assessment threshold Q, then it will be determined that the several groups of compound fertilizer granules randomly selected in the current S11 step are in a qualified state, and the first detection command will be triggered.

[0046] If the qualified assessment index Hgzs > the assessment threshold Q, it will be determined that several groups of compound fertilizer granules randomly selected in the current S11 step are not in a qualified state, and the second detection command will be triggered.

[0047] Preferably, step S4 specifically includes:

[0048] S41. Upon receiving the detection instruction in S3, the secondary response detection and evaluation mechanism is executed. According to the type of detection instruction, corresponding response conditions are formulated. If the response conditions are met, it indicates that the production of this batch of compound fertilizer granules is in an unqualified state, and a secondary conclusion instruction is triggered. At this time, the current production batch will be stopped, and the production operators will be notified to intervene manually.

[0049] Preferably, step S4 further includes:

[0050] S42. If a detection command No. 1 is received, define response conditions No. 1, as follows:

[0051]

[0052] In the formula, Rjz1 is the first response condition, representing the number of times the secondary conclusion instruction is triggered during X random sampling processes; x = 1, 2, 3, ..., X, where X represents the number of times several groups of compound fertilizer granules to be tested are randomly sampled; Hgzs x Let represent the pass evaluation index of several groups of compound fertilizer granules to be tested randomly selected in the xth time, U represent the preset minimum number of triggers, and I() is an indicator function, which means 0 if the condition is met and 1 otherwise.

[0053] S43. If the second detection command is received, formulate the second response condition, the specific formulation of which is as follows:

[0054]

[0055] In the formula, Rjz2 is the second response condition, representing the number of times the secondary conclusion instruction is triggered during X random sampling processes; x = 1, 2, 3, ..., X, where X represents the number of times several groups of compound fertilizer granules to be tested are randomly sampled; Hgzs x Let represent the pass / fail evaluation index of several groups of compound fertilizer granules to be tested randomly selected in the xth time, U represent the preset minimum number of triggers, and I() is an indicator function, which means that it is 1 if the condition is met, and 0 otherwise.

[0056] This invention provides a method for detecting the particle size of compound fertilizer granules, which has the following beneficial effects:

[0057] (1) This method generates a three-dimensional surface model of the particles through multi-angle image acquisition, which can more accurately describe the three-dimensional morphology and structural features of the particles, thereby obtaining more reliable results in morphological complexity analysis. This three-dimensional reconstruction technology has more advantages than traditional two-dimensional detection and can effectively solve the problems of complex shape and low detection accuracy of compound fertilizer particles. By extracting the morphological complexity coefficient and the relative standard deviation of particle size distribution, this method can more comprehensively reflect the size, shape and uniformity characteristics of the particles, providing multi-dimensional data support for compound fertilizer quality assessment. On this basis, a particle scoring model is constructed by combining deep learning technology. The input is used to obtain the qualification assessment index Hgzs, and a detection command is issued according to the preset threshold. This not only achieves high-precision particle quality discrimination, but also effectively reduces manual intervention and human error. In addition, this method designs a secondary response detection and evaluation mechanism, which further analyzes the results of the initial detection to ensure the rigor of the detection and the stability of the quality control of compound fertilizer production. By adjusting the production process through the feedback of this secondary detection, the manufacturing process of compound fertilizer can be optimized in a timely manner, improving the uniformity and consistency of the product, so that the compound fertilizer particles have a better distribution effect and a longer-lasting fertilizer effect during application.

[0058] (2) This method effectively quantifies the shape complexity of particles by introducing a shape factor, making particle characterization more refined. The calculation of eccentricity can further reflect the elongation or flatness of particles, helping to predict the flowability and bulk density of particles during transportation and fertilization, thereby ensuring the uniformity of compound fertilizer application. Based on obtaining particle morphological characteristics, the method also quantifies the uniformity of particle size by calculating the relative standard deviation of particle size distribution. By statistically analyzing the particle size data of randomly selected compound fertilizer particles, the relative standard deviation can be obtained, which can effectively measure the concentration of particle size distribution. The smaller the relative standard deviation, the more uniform the particle size, and the more consistent the release characteristics of the particles during application, improving the persistence and uniformity of fertilizer effect. In addition, standard deviation analysis can provide data support for further optimization of production processes, ensuring the quality stability of different batches of products.

[0059] (3) This method constructs an efficient particle scoring model using deep learning technology, achieving accurate evaluation and automated control of compound fertilizer particle quality. First, by training and testing the basic model and extracting feature information, the model is further optimized and trained into a recognition model, ultimately forming a particle scoring model. This allows the evaluation process to better capture the characteristics of particle morphology and particle size distribution. This multi-level model training method combines the self-learning and recognition capabilities of deep learning, ensuring the accuracy and reliability of the evaluation. Through linear normalization processing after model training, the data of compound fertilizer particles from different batches can be effectively standardized, thereby fitting and outputting the qualified evaluation index Hgzs, achieving high-precision prediction and evaluation of product quality. By comparing the size of the qualified evaluation index with the preset evaluation threshold, the system can automatically determine the qualified status of the current compound fertilizer particles, simplifying the manual operation process. When the particle is in a qualified state, the system automatically determines that the particle is in a qualified state and triggers the first detection command; when it is not in a qualified state, the second detection command is triggered to start further evaluation. This intelligent detection mechanism can promptly identify and screen out particles that do not meet the standards, providing data support for real-time control of the production process. Overall, this method further improves the automation and testing efficiency of compound fertilizer quality control, effectively reduces the risk of human error, ensures the consistency and stability of product quality, and helps to meet the quality standards for different fertilization needs.

[0060] (4) By introducing a secondary response detection and evaluation mechanism, the quality control level in the production process has been further improved. When a potential quality problem is detected, the system formulates corresponding response conditions based on different types of detection instructions to conduct a more in-depth evaluation of the production process. The beneficial effect of this mechanism is that it can prevent unqualified batches of granules from entering the market through multiple random samplings and strict judgment standards. If the detection result meets the preset response conditions, the system will automatically trigger a secondary conclusion instruction, stop the current production batch, and notify the production operators to intervene manually, effectively avoiding the accumulation and expansion of quality problems. Through the dynamic setting of different types of response conditions, the quality control in the production process is more flexible and adaptable. In addition, this detection mechanism can provide production personnel with reliable early warning information, which helps to promptly discover and resolve potential quality hazards in the production process, ensure the stability and consistency of compound fertilizer products, maximize production efficiency, and reduce economic losses caused by rework and quality problems. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the process for detecting the particle size of compound fertilizer particles according to the present invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1

[0064] Please see Figure 1 This invention provides a method for detecting the particle size of compound fertilizer granules, comprising the following steps:

[0065] S1. Randomly capture the compound fertilizer particles to be detected, and capture two-dimensional projection images of the compound fertilizer particles to be detected from multiple angles through image acquisition equipment. After stereo reconstruction, a three-dimensional surface model is generated.

[0066] S2. Based on the three-dimensional surface model and combined with the two-dimensional projection image of the corresponding compound fertilizer particles to be detected, extract the relevant structural data set of the corresponding compound fertilizer particles to be detected. Based on the relevant structural data set, analyze the structural complexity of the compound fertilizer particles to be detected to obtain the morphological complexity coefficient Xfxs. Based on several groups of compound fertilizer particles to be detected, analyze the particle size state between the compound fertilizer particles to be detected to obtain the relative standard deviation of particle size distribution Xbc.

[0067] S3. Construct a particle scoring model using deep learning technology, and input the relative standard deviation of particle size distribution Xbc and the morphological complexity coefficient Xfxs into the particle scoring model to fit and output the pass assessment index Hgzs. Based on the comparison result between the pass assessment index Hgzs and the preset assessment threshold Q, issue the corresponding detection command.

[0068] S4. After receiving the detection instruction in S3, execute the secondary response detection and evaluation mechanism, formulate response conditions to generate secondary detection results, and adjust the process based on the secondary detection results.

[0069] In this embodiment, the method employs multi-angle image acquisition and stereo reconstruction technology to efficiently acquire the true three-dimensional surface model of the compound fertilizer granules to be tested. This overcomes the shortcomings of traditional two-dimensional image detection methods in measuring the surface area and volume of irregular particles. Based on the constructed data set extracted from the three-dimensional surface model, it can not only accurately calculate the morphological complexity coefficient Xfxs, reflecting the geometric complexity of the particles, but also further obtain the relative standard deviation Xbc of the particle size distribution to evaluate the uniformity of particle size. In this way, by comprehensively analyzing the particle size differences between different particles, a comprehensive and refined measurement of the particle size of compound fertilizer granules is further realized. In addition, the particle scoring model constructed with deep learning technology can quickly output the qualification assessment index Hgzs and compare it with the preset assessment threshold Q. If the detection result is unqualified, a secondary response detection and evaluation mechanism will be triggered to further verify the particle quality and formulate strategies to optimize and adjust the process conditions based on the results, ensuring that the particle quality continuously meets the standards. The proposed method moves the particle size detection of compound fertilizer granules from traditional simple particle screening to intelligent, automated, and high-precision particle size detection, further improving the quality control level and detection efficiency of compound fertilizer production.

[0070] Example 2

[0071] Please refer to Figure 1 Specifically, the steps in S1 include:

[0072] S11. Deploy multi-view image acquisition equipment in advance, and randomly select several groups of compound fertilizer particles to be tested. After taking pictures of the compound fertilizer particles to be tested from multiple angles, obtain two-dimensional projection images of each compound fertilizer particle to be tested. After classification, obtain a two-dimensional set of each compound fertilizer particle to be tested. At the same time, according to the set position and equipment calibration of each image acquisition device during capture, obtain the intrinsic and extrinsic parameters of each image acquisition device. The intrinsic parameters include the focal length and optical center position of the image acquisition device, and the extrinsic parameters include the position and angle between the image acquisition devices. The specific two-dimensional images provide projection information of the particles at various angles.

[0073] The specific steps in S1 also include:

[0074] S12. Based on the two-dimensional set of each compound fertilizer particle to be detected, find the corresponding feature points p between the two-dimensional projected images within the two-dimensional set. j For example, the corners and edges of the corresponding compound fertilizer particles to be detected, in order to perform feature point p j The registration process is performed, and geometric relationships are used to identify the relative points from multiple viewpoints, i.e., the feature points p corresponding to the multiple viewpoints. j To obtain the corresponding feature points p of each compound fertilizer particle to be tested. j The location in three-dimensional space, and marked as three-dimensional point P. iSpecifically, it is calculated and obtained as follows:

[0075]

[0076] In the formula, j = 1, 2, 3, ..., N; N represents the number of viewpoints; p j Let π be the two-dimensional coordinates of the feature point extracted from the j-th image acquisition device for the corresponding compound fertilizer particle to be detected; that is, the projected position of the corresponding compound fertilizer particle to be detected on the image plane of the j-th image acquisition device. j δ is the projection matrix; j (P i ) is the projection function, which projects the three-dimensional point P. i Projected onto the two-dimensional projection image plane of the j-th image acquisition device; ||p j -δ j (P i Let P be the 3D point from the perspective of the j-th image acquisition device. i The projection of the feature point p in the image and its two-dimensional coordinates j The error between the two points is the Euclidean distance between them. () represents finding the 3D point P that minimizes the total error. i Where i represents the sequence number of the three-dimensional point within the corresponding compound fertilizer particle to be detected;

[0077] This formula utilizes feature points p within a multi-view two-dimensional projection image. j and its projection matrix δ j The three-dimensional point P of the compound fertilizer granules to be tested was calculated. i The goal of this study is to minimize the projection error in each viewpoint, thereby finding a spatial point P in three-dimensional space that relatively conforms to the viewpoints of all two-dimensional projected images. i ;

[0078] S13. Obtain the 3D point P based on S12. i The method involves obtaining several sets of three-dimensional points P on the corresponding compound fertilizer granules to be tested. i And by combining several sets of three-dimensional points P i Connect the links to generate the three-dimensional surface of the corresponding compound fertilizer particles to be tested;

[0079] S14. Based on the obtained 3D surface data of the corresponding compound fertilizer granules to be tested, and using DEM software such as EDEM or LIGGGHTS, perform 3D simulation on each granule to construct a 3D surface model. Through 3D simulation, you can pre-evaluate the performance of compound fertilizer granules in a virtual environment, optimize production processes, and avoid design flaws or problems that may occur in actual production. Simulation can help analyze the shape, size, flowability, and other characteristics of compound fertilizer granules, optimize production line design, and ensure the consistency and stability of compound fertilizer product quality. Specifically, 3D reconstruction helps to further accurately calculate complex geometric features such as the volume, surface area, and convex hull volume of the granules.

[0080] In this embodiment, the method effectively acquires the three-dimensional surface model of the compound fertilizer granules to be detected through multi-view image acquisition and stereo reconstruction technology, achieving high-precision three-dimensional morphological reconstruction. First, in S11, a multi-view image acquisition device is deployed to acquire two-dimensional projection images of the granules. Combining intrinsic and extrinsic parameters, the capture of granule feature points at different angles is more accurate. Second, in S12, multi-view registration and error minimization calculations of feature points are performed to calculate the three-dimensional point coordinates of the compound fertilizer granules, ensuring high accuracy and stability of the three-dimensional reconstruction. In S13, a complete three-dimensional surface is generated based on the three-dimensional points, further improving the reconstruction effect and enabling a comprehensive and accurate restoration of the true morphology of the granules to be detected. In S14, three-dimensional simulation is performed using the three-dimensional surface combined with DEM software. This allows for the evaluation of granule performance in a virtual environment, enabling early optimization of the production process and further avoiding potential problems in production. The simulation process helps to comprehensively analyze the shape, size, flowability, and other characteristics of the granules, making the production process design more scientific and reasonable, and effectively improving the quality consistency and stability of the compound fertilizer granules. Meanwhile, 3D reconstruction provides strong support for accurately calculating the complex geometric features of particles, such as volume, surface area, and convex hull volume, offering reliable data for further improvements in production and quality control. Therefore, this method not only improves the accuracy and efficiency of compound fertilizer particle size detection but also minimizes the number of process adjustments in actual production, achieving efficient and precise management of the compound fertilizer production process.

[0081] Example 3

[0082] Please refer to Figure 1 Specifically, the steps in S2 include:

[0083] S21. Based on the three-dimensional surface model and combined with the two-dimensional projection image of the corresponding compound fertilizer particle to be detected, extract the relevant structural data set of the corresponding compound fertilizer particle to be detected. The relevant structural data set includes the projection perimeter, projection area Tmj, convex hull area Tbmj, major axis length Czc, minor axis length Dzc, particle size value Kjz, and volume of the corresponding two-dimensional projection image of each compound fertilizer particle to be detected.

[0084] S22. Based on the aforementioned relevant structural data set, analyze the complexity of the structure of each compound fertilizer particle to be tested, and after linear normalization, obtain the morphological complexity coefficient Xfxs of each compound fertilizer particle to be tested, specifically obtained in the following manner:

[0085]

[0086] In the formula, Xyz represents the shape factor, and Tyz represents the convexity factor. The shape regularity is represented by Pxz, which reflects the degree of regularity of the shape. Pxz represents the eccentricity. α and β are both weight values. C represents the first correction constant. The specific values ​​of α and β are set by the user according to the situation.

[0087] S23. Based on the morphological complexity coefficients Xfxs of each compound fertilizer particle to be tested obtained in S22, the average morphological complexity coefficient is obtained by summing them. The specific method is as follows: Where k = 1, 2, 3, ..., n; n represents the number of compound fertilizer granules randomly selected from S11 to be tested, Xfxs k Let be the morphological complexity coefficient of the k-th compound fertilizer particle.

[0088] A larger shape factor indicates a more complex and irregular particle shape. Particles with complex shapes may have a higher surface area, affecting flowability and dissolution behavior.

[0089] In this embodiment, the method extracts the shape and size features of each compound fertilizer particle from multiple aspects through a data set constructed based on a three-dimensional surface model and a two-dimensional projection image, such as the projected perimeter, projected area, convex hull area, major and minor axis lengths, particle size, and volume. These feature parameters can describe the appearance and structural characteristics of the particles in more detail and accurately. By integrating these parameters into a constructed data set and combining it with linear normalization processing, a more comprehensive and standardized morphological complexity coefficient is generated. This coefficient accurately reflects the morphological complexity of the compound fertilizer particles, which helps to achieve fine control of particle quality. Furthermore, by calculating the average morphological complexity coefficient, the present invention can effectively evaluate the overall morphological consistency of randomly selected compound fertilizer particles, making it easier to monitor and control quality fluctuations during the production process.

[0090] Example 4

[0091] Please refer to Figure 1 Specifically, the S2 steps also include:

[0092] S221, The shape factor Xyz is obtained by the following formula:

[0093]

[0094] In the formula, This represents the average projected perimeter of the corresponding compound fertilizer granule to be detected, which is calculated from the edge detection results. The average projected area of ​​the corresponding compound fertilizer particle to be detected is represented by the number of pixels inside the edge; the shape factor Xyz is used to represent the shape complexity of the corresponding compound fertilizer particle to be detected; π refers to pi, which has a value of approximately 3.14159.

[0095] S222, The convexity factor Tyz is obtained by the following formula:

[0096]

[0097] In the formula, Tmj represents the projected area; Tbmj represents the convex hull area, which is the smallest convex region surrounding the compound fertilizer particle; the convexity factor Tyz reflects the ratio of the actual projected area of ​​the compound fertilizer particle to its smallest enclosing convex polygon. The closer the value is to 1, the more ideal the convexity of the compound fertilizer particle.

[0098] The convex hull area Tbmj mentioned above is the area of ​​the smallest convex region of the particle. It is obtained by first extracting the edge contour of the particle using an edge detection algorithm (such as Canny edge detection), and then generating the smallest convex hull containing all contour points using Gift Wrapping or GrahamScan algorithms. The area is calculated based on the geometric contour of the convex hull, and this area can be the pixel count.

[0099] S223, the eccentricity Pxz is obtained through the following formula:

[0100]

[0101] In the formula, Dzc represents the short axis length of the compound fertilizer particle to be tested, and Czc represents the long axis length of the compound fertilizer particle to be tested.

[0102] The minor axis length Dzc of the compound fertilizer granules to be tested refers to the diameter of the granule shape in the maximum direction, that is, the longest distance from one end of the granule to the other end. The major axis length is suitable for irregular or elliptical granules and can better describe the size characteristics of the granules.

[0103] The major axis length Czc of the compound fertilizer granules to be tested refers to the shortest diameter in the direction perpendicular to the major axis. It is used to supplement the geometric characteristics of the granules. The combination of the minor axis length and the major axis length can more accurately describe the aspect ratio or flatness of the granules and help to judge the morphological deviation of the granules.

[0104] The phenomenon indicates that the closer the eccentricity Dzc is to 1, the flatter or longer the compound fertilizer particles are, and the greater the deviation from the spherical shape. If the eccentricity of the compound fertilizer particles is large, it may affect the bulk density and flowability of the particles, resulting in uneven flow during transportation and fertilization.

[0105] S2 also includes the following specific steps:

[0106] S24. Based on the random sampling of several groups of compound fertilizer particles to be tested in S11, analyze the particle size distribution among the compound fertilizer particles to be tested to obtain the relative standard deviation Xbc of the particle size distribution. The relative standard deviation Xbc of the particle size distribution is obtained in the following way:

[0107]

[0108] In the formula, n represents the number of compound fertilizer granules randomly selected from S11 to be tested, Kjz avg The average particle size of the compound fertilizer granules to be tested is represented by Kjz. k This represents the particle size value of the kth compound fertilizer particle to be tested;

[0109] The particle size values ​​Kjz of each of the above-mentioned compound fertilizer particles to be tested can be obtained by monitoring with a laser particle size analyzer.

[0110] The smaller the relative standard deviation Xbc value of the particle size distribution, the more concentrated the distribution and the more uniform the particle size of the compound fertilizer.

[0111] In this embodiment, the method effectively characterizes the geometric morphological features of compound fertilizer granules by quantifying indicators such as shape factor, convexity factor, and eccentricity in the detection of morphological complexity. The shape factor reflects the complexity of the granule shape based on its projected perimeter and area; the convexity factor measures how close the granule is to an ideal convex shape; and the eccentricity evaluates the degree of flatness or elongation of the granule through the ratio of its major and minor axes. Compared with traditional detection methods, these image analysis-based parameters are more accurate and comprehensive, reflecting the morphological characteristics of the granules in greater detail and helping to determine whether the granule quality meets standards. Furthermore, this method calculates the relative standard deviation of the particle size distribution based on the particle size value of each granule through random sampling, accurately measuring the uniformity of particle size in the sample. The smaller the relative standard deviation, the more concentrated the particle size distribution and the more consistent the particle size, which is beneficial for achieving uniform distribution and flowability of granules during production, transportation, and fertilization. This detection method is fundamental to further improving particle consistency, thus making the quality control of compound fertilizer granules more precise and efficient.

[0112] Example 5

[0113] Please refer to Figure 1 Specifically, the S3 steps include:

[0114] S31. A basic model is initially constructed using deep learning technology. Relevant constructed data sets are input into the basic model for training and testing. The trained basic model is then used as a recognition model. Feature information within the recognition model is acquired, and this acquired feature information is used to train and test the recognition model. The trained recognition model is then used as a particle scoring model. After training and linear normalization, the qualified assessment index Hgzs of all randomly selected compound fertilizer particles to be tested from S11 is fitted and output. Specifically, this is obtained through the following methods:

[0115]

[0116] In the formula, Xbc represents the average morphological complexity coefficient, G represents the relative standard deviation of particle size distribution, Tcc represents the volume difference value, and a, b, and c are all weight values. The specific values ​​of a, b, and c are set by the user according to the situation.

[0117] The specific steps in S3 also include:

[0118] S32. A pre-set evaluation threshold Q is used to compare and analyze the evaluation threshold Q with the qualified evaluation index Hgzs to determine whether the randomly selected groups of compound fertilizer granules to be tested in the current step S11 are in a qualified state. The specific judgment content is as follows:

[0119] If the qualified assessment index Hgzs ≤ the assessment threshold Q, then it will be determined that the several groups of compound fertilizer granules randomly selected in the current S11 step are in a qualified state, and the first detection command will be triggered.

[0120] If the qualified assessment index Hgzs > the assessment threshold Q, it will be determined that several groups of compound fertilizer granules randomly selected in the current S11 step are not in a qualified state, and the second detection command will be triggered.

[0121] In this embodiment, the method utilizes a particle scoring model constructed using deep learning technology to efficiently and accurately assess the quality of compound fertilizer particles. First, the constructed data set of compound fertilizer particles is input into a basic model for initial training and testing. The resulting basic model serves as a recognition model, used to further acquire key feature information of the particles. Based on this, the recognition model is trained and tested again, ultimately constructing a scoring model suitable for particle quality evaluation. This scoring model obtains the qualification assessment index (Hgzs) of randomly selected compound fertilizer particles, and combines it with linear normalization to ensure high accuracy and consistency of the assessment index. Under a pre-set assessment threshold, the method automatically determines whether the particles meet quality standards by comparing the qualification assessment index. This automated judgment process further improves detection efficiency, enabling compound fertilizer production to quickly obtain quality feedback and promptly identify and handle substandard products. Through this method, compound fertilizer production lines can more consistently output high-quality products, reducing the impact of human intervention on detection results, optimizing production processes, effectively ensuring the consistency and stability of compound fertilizer products, and thus enhancing the product's market competitiveness.

[0122] Example 6

[0123] Please refer to Figure 1 Specifically, the S4 steps include:

[0124] S41. Upon receiving the detection instruction in S3, the secondary response detection and evaluation mechanism is executed. According to the type of detection instruction, corresponding response conditions are formulated. If the response conditions are met, it indicates that the production of this batch of compound fertilizer granules is in an unqualified state, and a secondary conclusion instruction is triggered. At this time, the current production batch will be stopped, and the production operators will be notified to intervene manually.

[0125] The specific steps in S4 also include:

[0126] S42. If a detection command No. 1 is received, define response conditions No. 1, as follows:

[0127]

[0128] In the formula, Rjz1 is the first response condition, representing the number of times the secondary conclusion instruction is triggered during X random sampling processes; x = 1, 2, 3, ..., X, where X represents the number of times several groups of compound fertilizer granules to be tested are randomly sampled; Hgzs x Let represent the pass evaluation index of several groups of compound fertilizer granules to be tested randomly selected in the xth time, U represent the preset minimum number of triggers, and I() is an indicator function, which means 0 if the condition is met and 1 otherwise.

[0129] S43. If the second detection command is received, formulate the second response condition, the specific formulation of which is as follows:

[0130]

[0131] In the formula, Rjz2 is the second response condition, representing the number of times the secondary conclusion instruction is triggered during X random sampling processes; x = 1, 2, 3, ..., X, where X represents the number of times several groups of compound fertilizer granules to be tested are randomly sampled; Hgzs x Let represent the pass / fail evaluation index of several groups of compound fertilizer granules to be tested randomly selected in the xth time, U represent the preset minimum number of triggers, and I() is an indicator function, which means that it is 1 if the condition is met, and 0 otherwise.

[0132] In this embodiment, the method achieves dynamic quality control and real-time early warning response for production batches through a secondary response detection and evaluation mechanism. Specifically, when the system receives a detection command after the initial detection, it formulates different response conditions according to the command type and further analyzes the detection results. By designing response conditions under detection commands one and two, the method can accurately determine whether the production batch meets quality requirements under different scenarios, and immediately issue a secondary conclusion command in the case of non-compliance, thereby promptly stopping the current production batch and notifying operators for manual intervention. This mechanism ensures the early detection and handling of potential quality problems, minimizing the risk of non-conforming products entering the market. In this mechanism, the reliability of the evaluation is ensured by setting response condition formulas and trigger minimum thresholds during multiple random sampling detection processes. Through the judgment mechanism of the indicator function, the system can automatically determine the detection status when the trigger conditions are met or not, thereby achieving fully automated monitoring. Overall, this method further improves the accuracy and flexibility of quality control in the production process of compound fertilizer granules, reduces the workload of manual inspection, optimizes production efficiency and product quality stability, and enhances the scientificity and controllability of production management.

[0133] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the particle size of compound fertilizer particles, characterized by: The method comprises the following steps, S1, randomly capturing the compound fertilizer particles to be detected, and capturing the two-dimensional projection images of the compound fertilizer particles to be detected from multiple angles by an image acquisition device, and generating a three-dimensional surface model after stereoscopic reconstruction; S2, based on the three-dimensional surface model, and in combination with the two-dimensional projection images of the corresponding compound fertilizer particles to be detected, extracting the relevant structure data set of the corresponding compound fertilizer particles to be detected, analyzing the structure complexity of the compound fertilizer particles to be detected based on the relevant structure data set, obtaining the morphological complexity coefficient Xfxs, and analyzing the particle size state among the compound fertilizer particles to be detected according to a plurality of groups of compound fertilizer particles to be detected, and obtaining the relative standard deviation Xbc of the particle size distribution; S3, constructing a particle scoring model using deep learning technology, inputting the relative standard deviation Xbc of the particle size distribution and the morphological complexity coefficient Xfxs into the particle scoring model, fitting the qualified evaluation index Hgzs, and issuing a corresponding detection instruction according to the comparison result of the qualified evaluation index Hgzs and the preset evaluation threshold Q; S4, after receiving the detection instruction in S3, a secondary response detection evaluation mechanism is executed, a response condition is formulated, a secondary detection result is generated, and the process is adjusted based on the secondary detection result; The specific steps of S1 include: S11, deploying multiple-view image acquisition devices in advance, and randomly extracting a plurality of groups of compound fertilizer particles to be detected, obtaining the two-dimensional projection images of each compound fertilizer particle to be detected by photographing the compound fertilizer particles to be detected from multiple angles, and obtaining the two-dimensional set of each compound fertilizer particle to be detected after classification, and obtaining the intrinsic and extrinsic parameters of each image acquisition device according to the set position and device calibration when each image acquisition device captures, wherein the intrinsic parameters include the focal length and optical center position of the image acquisition device, and the extrinsic parameters include the position and angle between the image acquisition devices; The specific steps of S3 include: S32, the evaluation threshold Q is preset, and the evaluation threshold Q is compared and analyzed with the qualified evaluation index Hgzs to determine whether the plurality of groups of compound fertilizer particles to be detected randomly extracted in the current S11 step are in a qualified state, and the specific determination content is as follows: If the qualified evaluation index Hgzs is less than or equal to the evaluation threshold Q, it is determined that the plurality of groups of compound fertilizer particles to be detected randomly extracted in the current S11 step are in a qualified state, and a first detection instruction is triggered; If the qualified evaluation index Hgzs is greater than the evaluation threshold Q, it is determined that the plurality of groups of compound fertilizer particles to be detected randomly extracted in the current S11 step are not in a qualified state, and a second detection instruction is triggered; The specific steps of S4 include: S41, after receiving the detection instruction in S3, a secondary response detection evaluation mechanism is executed, and according to the type of the detection instruction, a corresponding response condition is formulated, if the response condition is met, it indicates that the production of the batch of compound fertilizer particles is in an unqualified state, and a secondary conclusion instruction is triggered, at which time the current production batch is stopped and the production operator is notified for manual intervention; The specific steps of S4 further include: S42, if the first detection instruction is received, a first response condition is formulated, and the specific formulation content is as follows: ; In the formula, is a first response condition, indicating the number of times of triggering a second conclusion instruction in X random extraction processes; x = 1, 2, 3,..., X, X representing the number of times of randomly extracting a plurality of groups of composite fertilizer particles, is an evaluation index of the plurality of groups of composite fertilizer particles for the xth random extraction, is a preset minimum triggering number, and I() is an indicator function, indicating 0 if the condition is met, and 1 otherwise. S43, if the second detection instruction is received, a second response condition is formulated, and specific formulation contents are as follows: ; In the formula, is the second response condition, indicating the number of times of triggering the second conclusion instruction in X random extraction processes; x = 1, 2, 3,..., X, X represents the number of times of randomly extracting a plurality of groups of composite fertilizer particles, is the qualified evaluation index of the plurality of groups of composite fertilizer particles randomly extracted for the xth time, is the preset minimum number of triggering times, and I() is an indicator function, indicating 1 if the condition is met, and 0 otherwise.

2. The method according to claim 1, wherein: S1The specific steps further include: S12. Based on the two-dimensional set of each compound fertilizer particle to be detected, find the corresponding feature points between the two-dimensional projection images within the two-dimensional set. To perform feature point analysis The registration process is performed, and geometric relationships are used to identify the relative points from multiple perspectives, i.e., the feature points corresponding to each perspective. To obtain the corresponding feature points of each compound fertilizer particle to be tested. The location in three-dimensional space, and marked as a three-dimensional point. Specifically, it is calculated and obtained as follows: ； In the formula, j = 1, 2, 3,..., N; N represents the number of view angles; is the two-dimensional coordinate of the feature point extracted for the corresponding to-be-detected compound fertilizer particle under the jth image acquisition device view angle; is a projection matrix; is a projection function, which projects the three-dimensional point onto the two-dimensional projection image plane of the jth image acquisition device; is the projection of the three-dimensional point in the image under the jth image acquisition device view angle; is the error between the two-dimensional coordinate of the feature point and the projection of the three-dimensional point ; wherein, i represents the serial number of the three-dimensional point in the corresponding to-be-detected compound fertilizer particle; and the error is the Euclidean distance between the two points. S13. Obtaining 3D points based on S12 The method involves obtaining several sets of three-dimensional points on the corresponding compound fertilizer particles to be tested. And by combining several sets of three-dimensional points Connect the links to generate the three-dimensional surface of the corresponding compound fertilizer particles to be tested; S14, according to the three-dimensional surface of the corresponding to-be-detected compound fertilizer particles obtained, the three-dimensional simulation of each to-be-detected compound fertilizer particle is performed in combination with the DEM software, so as to construct a three-dimensional surface model.

3. The method according to claim 1, wherein: S2The specific steps include: S21, based on the three-dimensional surface model, and in combination with the two-dimensional projection image of the corresponding compound fertilizer particle to be detected, relevant structure data sets of the corresponding compound fertilizer particle to be detected are extracted, wherein the relevant structure data sets include the projection perimeter, the projection area, the convex hull area, the long axis length, the short axis length, the particle size value, and the volume of the corresponding two-dimensional projection image of each compound fertilizer particle to be detected ​​​​​ S22, according to the related structure data set, the complexity of the structure of each to-be-detected compound fertilizer particle is analyzed, and after linear normalization processing, the shape complexity coefficient Xfxs of each to-be-detected compound fertilizer particle is obtained, and the shape complexity coefficient Xfxs is obtained in the following manner: ; In the formula, is expressed as a shape factor, is expressed as a convexity factor, is expressed as a shape regularity, is expressed as an eccentricity, and are weight values, is expressed as a first correction constant, wherein, and the specific values are set by the user according to the situation; S23、According to the morphological complexity coefficient Xfxs of each to-be-detected compound fertilizer particle obtained in S22, an average morphological complexity coefficient is obtained after summation The specific method is: Wherein, k=1, 2, 3,..., n; n represents the number of randomly selected to-be-detected compound fertilizer particles in S11, Xfk represents the morphological complexity coefficient of the kth compound fertilizer particle.

4. The method according to claim 3, wherein: S2The specific steps further include: S221. the form factor is obtained by the following equation: ; wherein is expressed as the average projected perimeter of the respective compound fertilizer particle to be detected; is expressed as the average projected area of the respective compound fertilizer particle to be detected; shape factor for expressing the shape complexity of the respective compound fertilizer particle to be detected, is expressed as the circumference ratio; S222、 the convexity factor is obtained by the following equation: ; wherein expressed as projected area; expressed as convex hull area, the convex hull being the smallest convex region that encompasses the compound fertilizer particle; S223, eccentricity Obtained by the following formula: ; In the formula, represents the short axis length of the compound fertilizer particle to be detected, represents the long axis length of the compound fertilizer particle to be detected.

5. The method according to claim 3, wherein: S2The specific steps further include: S24, according to the to-be-detected compound fertilizer particles randomly extracted in S11, the particle size state between each to-be-detected compound fertilizer particle is analyzed, so as to obtain the relative standard deviation Xbc of the particle size distribution, and the relative standard deviation Xbc of the particle size distribution is obtained in the following manner: ; In the formula, n represents the number of the composite fertilizer particles to be detected randomly selected in S11, represents the average particle size value of the composite fertilizer particles to be detected, represents the particle size value of the kth composite fertilizer particle to be detected.

6. The method according to claim 5, wherein: S3The specific steps include: S31, a basic model is initially constructed by using a deep learning technology, and related structure data sets are input into the basic model for training and testing, the basic model after training is taken as an identification model, feature information in the identification model is obtained, the obtained feature information is used for training and testing of the identification model, the identification model after training is taken as a particle scoring model, after training and linear normalization processing, the qualified evaluation index Hgzs of all to-be-detected compound fertilizer particles randomly extracted from S11 is fitted and output, and the qualified evaluation index Hgzs is obtained in the following manner: ; In the formula, is expressed as an average morphological complexity coefficient, is expressed as a relative standard deviation of the particle size distribution, is expressed as a second correction coefficient, is expressed as a volume difference value, , and are weight values, wherein, , and are specific numerical values which are set by the user according to the situation.

Images