A production image processing method for mixed fertilizer

Abstract

The present invention discloses a method for processing images of mixed fertilizer production, belonging to the field of image monitoring technology. The method comprises the following steps: S1, capturing images of finished products on a conveyor belt and constructing a uniform value for the finished product images; S2, determining local results for each corner point based on the uniform value of the finished product images; S3, determining suspected clumping regions based on the local results for each corner point in the finished product images, screening the suspected clumping regions, and determining the final clumping regions. The present invention quantifies the uniformity of the entire image by constructing a uniform value for the image, rather than relying on a fixed threshold. The uniform value is based on global statistics (such as the extreme values ​​of RGB values ​​in an eight-neighborhood), suppressing the effects of local noise or sensor fluctuations. Furthermore, the present invention only performs local analysis on corner points, significantly reducing the amount of computation while retaining key clump edges. Neighborhood connectivity analysis of corner points is used to eliminate isolated noise and preserve true clumps.

Description

Technical Field

[0001] The invention belongs to the technical field of image monitoring, and in particular relates to a production image processing method for mixed fertilizer. Background Art

[0002] In fertilizer production, the clumping problem of mixed fertilizers directly impacts product quality and performance. This phenomenon is often caused by factors such as raw material hygroscopicity, fluctuations in production process parameters, or insufficient cooling. Once formed, fertilizer particles can clump together, leading to poor feeding and uneven dosing during application, and even causing production accidents such as equipment blockage. Therefore, real-time clumping detection of finished products on the conveyor belt is a critical step in ensuring product quality during mixed fertilizer production.

[0003] Traditional agglomeration detection methods mainly rely on manual sampling inspection or simple mechanical screening, which has the following limitations: 1. Subjectivity and lag: Manual inspection is inefficient and difficult to cover the entire process. The results are affected by personnel experience and cannot achieve real-time feedback; 2. Destructive testing: Mechanical screening requires sampling and offline analysis, which may damage product integrity and cannot reflect the real-time status of the conveyor belt during dynamic production; 3. Poor environmental adaptability: Fertilizer production sites have problems such as dust, vibration and uneven lighting. Traditional visual inspection methods are easily interfered with, resulting in false detection or missed detection.

[0004] With the development of machine vision technology, automated detection based on image processing has become a research hotspot. However, existing technologies still face the following challenges: 1. Difficulty in extracting agglomeration features: The surface texture of mixed fertilizer particles is complex, and the grayscale and texture differences between agglomerated areas and normal particles may change dynamically due to fluctuations in the raw material ratio. Traditional global threshold segmentation or fixed template matching methods lack robustness. 2. The conflict between computational efficiency and accuracy: High-precision algorithms (such as deep learning) require large amounts of annotated data and have poor real-time performance, while lightweight algorithms are susceptible to noise interference, making it difficult to balance detection speed and accuracy. 3. Weak adaptability to dynamic scenarios: Scenarios such as conveyor belt vibration, lighting changes, and overlapping particles make traditional corner detection or edge analysis methods prone to generating false targets. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a production image processing method for mixed fertilizer.

[0006] The technical solution of the present invention is: a production image processing method of mixed fertilizer comprises the following steps:

[0007] S1, collect the finished product image on the conveyor belt and construct a uniform value for the finished product image;

[0008] S2. Determine the local results of each corner point based on the average value of the finished image;

[0009] S3. Determine suspected agglomeration areas based on the local results of each corner point in the finished image, and screen the suspected agglomeration areas to determine the final agglomeration areas.

[0010] Furthermore, S1 includes the following sub-steps:

[0011] S11, collecting images of finished products on the conveyor belt;

[0012] S12, generating a two-dimensional matrix for each pixel position of the finished image;

[0013] S13. Generate a uniform value of the finished image according to the two-dimensional matrix of all pixel positions.

[0014] The beneficial effect of the above further solution is that: in the present invention, by constructing a two-dimensional matrix of pixel positions, the local neighborhood information (L of the 8 neighborhoods) is converted into max , L min The algorithm encodes the pixel values ​​(and the current pixel value) into matrix elements, converting local texture or edge information into matrix form. By extracting the eigenvalues ​​of the two-dimensional matrix and calculating the covariance, local structural information is converted into a global uniformity indicator. As a global uniformity value, the covariance can comprehensively assess the uniformity of the entire image, rather than relying on local thresholds, and can adapt to dynamic changes in the surface texture of the mixed fertilizer (such as fluctuations in the raw material ratio).

[0015] Furthermore, in S12, the expression of the two-dimensional matrix A of pixel positions is: , , , , Where a 11 Represents the first value of a two-dimensional matrix, a 22 Represents the second value of the two-dimensional matrix, a 12 Represents the third value of the two-dimensional matrix, a 21 Represents the fourth value of the two-dimensional matrix, L max Indicates the maximum RGB value of the 8-neighborhood pixel position, L min Represents the minimum RGB value of the 8-neighborhood of the pixel position, and l represents the RGB value of the pixel position.

[0016] Furthermore, S13 includes the following sub-steps:

[0017] S131, taking the first eigenvalue of the two-dimensional matrix corresponding to all pixel positions as the first eigenvalue;

[0018] S132. Taking the second eigenvalue of the two-dimensional matrix corresponding to all pixel positions as the second eigenvalue;

[0019] S133 , calculating the covariance between the first characteristic variable and the second characteristic variable as the average value of the finished image.

[0020] The beneficial effect of the above further solution is that in the present invention, the traditional method may misjudge the natural bumps and depressions of the particles as agglomerates, while covariance analysis can effectively distinguish between true agglomerates and surface texture changes by quantifying structural consistency.

[0021] Furthermore, S2 includes the following sub-steps:

[0022] S21, determining the weight coefficient of the pixel position using the average value of the finished image;

[0023] S22, taking the difference between 1 and the weight coefficient as the autocorrelation coefficient of the pixel position;

[0024] S23. The product of the RGB value of each corner point in the finished image and the autocorrelation coefficient is used as the local result of the corner point.

[0025] The beneficial effects of this further solution are as follows: In this invention, weight coefficients are determined by using the global mean value (covariance), enabling the algorithm to adaptively focus on non-uniform areas (possibly clumps). Local results are calculated only for corner points (salient points in the image, such as edge intersections), reducing computational effort while preserving key structural information. Corner points are typically located at the edges of clumps or where texture abruptly changes. Local result analysis allows for precise location of clump boundaries. Structural consistency (autocorrelation coefficient) and color information are combined to avoid misclassifying areas with unusual color but uniform structure as clumps.

[0026] Furthermore, in S21, the calculation formula of the weight coefficient k0 is: ; Where C represents the uniform value, e represents the exponent, and G represents the gradient amplitude at the pixel position.

[0027] The beneficial effect of the above further solution is that: in the present invention, k0 is dynamically determined by the ratio of C to G, adapting to different production batches (such as raw material ratios, humidity changes) and lighting conditions.

[0028] Furthermore, S3 includes the following sub-steps:

[0029] S31, taking the corner points whose local results are smaller than the mean of all local results as suspected agglomeration areas;

[0030] S32: In the suspected agglomeration area, if there is a pixel position in the four neighborhoods of the corner point that belongs to the suspected agglomeration area, then the corner point is determined to be an agglomeration point; otherwise, the corner point is a normal area;

[0031] S33. All agglomeration points are taken as the final agglomeration area.

[0032] The above-mentioned further solution has the following beneficial effects: In this invention, connectivity checks are performed on four neighborhoods (up, down, left, and right), requiring that clumping regions be spatially continuous. Isolated noise points are excluded because they lack other suspected clumping points in their neighborhoods, significantly reducing false positives. Discrete clumping points are merged into continuous regions without requiring a preset number of neighborhoods or distance thresholds. This approach, based entirely on data-driven connectivity judgment, accommodates clumping of varying sizes.

[0033] The present invention has the following advantages: it quantifies the uniformity of the entire image by constructing a uniform value, rather than relying on a fixed threshold. This uniform value is based on global statistics (such as the extreme values ​​of the RGB values ​​in an eight-neighborhood), suppressing the effects of local noise or sensor fluctuations. Furthermore, the present invention performs local analysis only on corner points, significantly reducing computational complexity while preserving key agglomeration edges. Neighborhood connectivity analysis of corner points eliminates isolated noise and preserves true agglomeration. Accurately locating agglomeration areas provides data support for adjusting process parameters (such as cooling time and raw material ratio), thereby reducing agglomeration. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Flowchart of the image processing method for the production of mixed fertilizer. DETAILED DESCRIPTION

[0035] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0036] like Figure 1 As shown, the present invention provides a production image processing method for mixed fertilizer, comprising the following steps:

[0037] S1, collect the finished product image on the conveyor belt and construct a uniform value for the finished product image;

[0038] S2. Determine the local results of each corner point based on the average value of the finished image;

[0039] S3. Determine suspected agglomeration areas based on the local results of each corner point in the finished image, and screen the suspected agglomeration areas to determine the final agglomeration areas.

[0040] In this embodiment of the present invention, S1 includes the following sub-steps:

[0041] S11, collecting images of finished products on the conveyor belt;

[0042] S12, generating a two-dimensional matrix for each pixel position of the finished image;

[0043] S13. Generate a uniform value of the finished image according to the two-dimensional matrix of all pixel positions.

[0044] In the present invention, by constructing a two-dimensional matrix of pixel positions, the local neighborhood information (L of the 8 neighborhoods) ismax , L min The algorithm encodes the pixel values ​​(and the current pixel value) into matrix elements, converting local texture or edge information into matrix form. By extracting the eigenvalues ​​of the two-dimensional matrix and calculating the covariance, local structural information is converted into a global uniformity indicator. As a global uniformity value, the covariance can comprehensively assess the uniformity of the entire image, rather than relying on local thresholds, and can adapt to dynamic changes in the surface texture of the mixed fertilizer (such as fluctuations in the raw material ratio).

[0045] In the embodiment of the present invention, in S12, the expression of the two-dimensional matrix A of pixel positions is: , , , , Where a 11 Represents the first value of a two-dimensional matrix, a 22 Represents the second value of the two-dimensional matrix, a 12 Represents the third value of the two-dimensional matrix, a 21 Represents the fourth value of the two-dimensional matrix, L max Indicates the maximum RGB value of the 8-neighborhood pixel position, L min Represents the minimum RGB value of the 8-neighborhood of the pixel position, and l represents the RGB value of the pixel position.

[0046] In this embodiment of the present invention, S13 includes the following sub-steps:

[0047] S131, taking the first eigenvalue of the two-dimensional matrix corresponding to all pixel positions as the first eigenvalue;

[0048] S132. Taking the second eigenvalue of the two-dimensional matrix corresponding to all pixel positions as the second eigenvalue;

[0049] S133 , calculating the covariance between the first characteristic variable and the second characteristic variable as the average value of the finished image.

[0050] In the present invention, traditional methods may misjudge the natural bumps and depressions of particles as agglomerates, while covariance analysis effectively distinguishes true agglomerates from surface texture changes by quantifying structural consistency.

[0051] In this embodiment of the present invention, S2 includes the following sub-steps:

[0052] S21, determining the weight coefficient of the pixel position using the average value of the finished image;

[0053] S22, taking the difference between 1 and the weight coefficient as the autocorrelation coefficient of the pixel position;

[0054] S23. The product of the RGB value of each corner point in the finished image and the autocorrelation coefficient is used as the local result of the corner point.

[0055] In this invention, weighting coefficients are determined by the global mean (covariance), enabling the algorithm to adaptively focus on non-uniform areas (possibly clumps). Local results are calculated only for corner points (salient points in the image, such as edge intersections), reducing computational effort while preserving key structural information. Corners are typically located at the edges of clumps or where texture changes suddenly occur. Local result analysis allows for precise location of clump boundaries. Structural consistency (autocorrelation coefficient) and color information are combined to avoid misclassifying areas with unusual color but uniform structure as clumps.

[0056] In the embodiment of the present invention, in S21, the calculation formula of the weight coefficient k0 is: ; Where C represents the uniform value, e represents the exponent, and G represents the gradient amplitude at the pixel position.

[0057] The beneficial effect of the above further solution is that: in the present invention, k0 is dynamically determined by the ratio of C to G, adapting to different production batches (such as raw material ratios, humidity changes) and lighting conditions.

[0058] In this embodiment of the present invention, S3 includes the following sub-steps:

[0059] S31, taking the corner points whose local results are smaller than the mean of all local results as suspected agglomeration areas;

[0060] S32: In the suspected agglomeration area, if there is a pixel position in the four neighborhoods of the corner point that belongs to the suspected agglomeration area, then the corner point is determined to be an agglomeration point; otherwise, the corner point is a normal area;

[0061] S33. All agglomeration points are taken as the final agglomeration area.

[0062] In this method, connectivity checks are performed across four neighborhoods (up, down, left, and right), requiring that clustered regions be spatially continuous. Isolated noise points are excluded if they lack other suspected clustered points in their neighborhoods, significantly reducing false positives. Discrete clustered points are merged into continuous regions without requiring a preset number of neighbors or distance thresholds. This method, based entirely on data-driven connectivity judgment, accommodates clusters of varying sizes.

[0063] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A production image processing method for mixed fertilizer, characterized in that: The following steps are involved: S1, collect the finished product image on the conveyor belt and construct a uniform value for the finished product image; S2. Determine the local results of each corner point based on the average value of the finished image; S3. Determine suspected agglomeration areas based on the local results of each corner point in the finished image, and screen the suspected agglomeration areas to determine the final agglomeration areas; The S1 includes the following sub-steps: S11, collecting images of finished products on the conveyor belt; S12, generating a two-dimensional matrix for each pixel position of the finished image; S13, generating a uniform value of the finished image according to the two-dimensional matrix of all pixel positions; In S12, the expression of the two-dimensional matrix A of pixel positions is: , , , , Where a 11 Represents the first value of a two-dimensional matrix, a 22 Represents the second value of the two-dimensional matrix, a 12 Represents the third value of the two-dimensional matrix, a 21 Represents the fourth value of the two-dimensional matrix, L max Indicates the maximum RGB value of the 8-neighborhood pixel position, L min Represents the minimum RGB value of the 8-neighborhood of the pixel position, and l represents the RGB value of the pixel position; The S13 includes the following sub-steps: S131, taking the first eigenvalue of the two-dimensional matrix corresponding to all pixel positions as the first eigenvalue; S132. Taking the second eigenvalue of the two-dimensional matrix corresponding to all pixel positions as the second eigenvalue; S133, calculating the covariance between the first characteristic variable and the second characteristic variable as the mean value of the finished image; The S3 includes the following sub-steps: S31, taking the corner points whose local results are smaller than the mean of all local results as suspected agglomeration areas; S32: In the suspected agglomeration area, if there is a pixel position in the four neighborhoods of the corner point that belongs to the suspected agglomeration area, then the corner point is determined to be an agglomeration point; otherwise, the corner point is a normal area; S33. All agglomeration points are taken as the final agglomeration area.

2. The production image processing method of mixed fertilizer according to claim 1, characterized in that, The S2 includes the following sub-steps: S21, determining the weight coefficient of the pixel position using the average value of the finished image; S22, taking the difference between 1 and the weight coefficient as the autocorrelation coefficient of the pixel position; S23. The product of the RGB value of each corner point in the finished image and the autocorrelation coefficient is used as the local result of the corner point.

3. The production image processing method of mixed fertilizer according to claim 2, characterized in that, In S21, the calculation formula of the weight coefficient k0 is: ; Where C represents the uniform value, e represents the exponent, and G represents the gradient amplitude at the pixel position.

Images