Illite and pine needle oil load uniformity detection method based on image recognition

Abstract

The invention relates to the field of image detection, in particular to an illite and pine needle oil load uniformity detection method based on image recognition, which comprises the following steps: acquiring a composite powder microscopic image under a lateral illumination condition to obtain an effective detection grayscale image and a seed point set; carrying out joint evaluation on local gradient and curvature response of the effective detection grayscale image to obtain an artifact suppression factor; performing joint evaluation on the consistency of the region growth direction and the local flux characteristic to obtain a flux compensation factor; carrying out fusion modulation on the gray difference judgment cost through an artifact inhibition factor and a flux compensation factor, obtaining a final growth judgment cost, and generating a load region binary mask; a load uniformity index is obtained by carrying out gridding load rate statistics and dispersion calculation on a binary mask of a load area, so that the problem of load uniformity quantization misalignment caused by shadow error growth leakage and coexistence of early stop and missing detection on an oil film weak edge is solved.

Description

Technical Field

[0001] This invention relates to the field of image detection technology, and in particular to a method for detecting the uniformity of illite and pine needle oil loading based on image recognition. Background Technology

[0002] In the field of fine processing and functionalized composite modification of inorganic non-metallic materials, loading volatile active ingredients such as pine needle oil onto the surface of submicron-sized illite powder can impart functions such as antibacterial properties, slow release, or odor regulation to the material while maintaining the stability of the mineral matrix. This is an important process route for preparing functional composite powders. Industrially, spray drying is often used to achieve the composite curing of active ingredients and illite powder. The composite powder formed after spray drying usually exists in the form of spherical or near-spherical aggregates. The product performance and batch stability are highly dependent on the coverage, wetting distribution, and spatial uniformity of pine needle oil on the surface of the aggregates. Therefore, it is necessary to conduct online or quasi-online detection of the coverage and distribution uniformity of the pine needle oil loading area on the production line to support closed-loop adjustment of process parameters and avoid functional degradation and decreased batch consistency due to uneven loading.

[0003] Current online detection methods often employ industrial microscopes to acquire macro images of spray-dried composite powders under fixed lighting conditions. Image segmentation algorithms then distinguish between the loaded region formed by pine needle oil infiltration and the unloaded region formed by the dried substrate. Load coverage is then statistically analyzed using a binary mask, and uniformity is evaluated using gridded statistics or spatial dispersion indices. Among commonly used segmentation algorithms, gradient-based seed region growth algorithms are widely used for extracting infiltration coverage areas because they can simulate the continuous penetration and diffusion behavior of oil on a solid surface in a "dark to light, inside to outside" growth process. These algorithms also feature strong constraints on regional connectivity, simple implementation, and ease of engineering deployment. Typically, these algorithms use pixels with the minimum grayscale value as seed points, and the grayscale difference between neighboring pixels and the seed point or the current growth region as the decision cost. A preset threshold is used to determine whether to incorporate neighboring pixels into the growth region. When the grayscale difference is less than the threshold, the pixel is considered part of the loaded region and expansion continues outward; when the grayscale difference is greater than the threshold, expansion stops, thus obtaining the segmentation result of the loaded region.

[0004] However, for spray-dried submicron-sized illite and pine needle oil composite powder samples, traditional region growing and segmentation based on a single gray-level difference metric is prone to identification bias under actual macro imaging conditions. This is because submicron-sized illite powder is prone to agglomeration. After spray drying, the surface of these agglomerates contains numerous microscopic stacking gaps and structural undulations. Under side lighting or oblique illumination, these microstructures generate high-frequency geometric self-occlusion shadows. The gray-level values ​​of these shadow areas are low and highly overlap with the gray-level characteristics of the pine needle oil-infiltrated areas. This causes region growing algorithms, when using only gray-level difference as a criterion, to mistakenly incorporate the drying gap shadows into the load area, resulting in growth leakage and the introduction of a large number of pseudo-load pixels. Consequently, this leads to false increases and fluctuations in coverage and uniformity statistics. Meanwhile, pine needle oil is volatile and somewhat translucent. At the wetting edge, it often exhibits a nonlinear gradient decay caused by a gradual change in oil film thickness and a shortening optical path. The grayscale of the edge region gradually lightens, and the contrast with the substrate decreases. Under a fixed threshold constraint, the region growing algorithm is prone to prematurely stopping due to increased grayscale difference before the oil film is fully extracted, resulting in missed detections of weak edge transition zones and region fragmentation. This leads to an underestimation of the load area and distortion of uniformity evaluation. These two types of errors often coexist and mutually constrain each other in the same imaging scene, making it difficult to simultaneously suppress misjudgments of gap shadows and complete the detection of missed weak edges of the oil film by adjusting only a single grayscale difference threshold. This affects the accuracy, stability, and batch comparability of online load uniformity detection results. Summary of the Invention

[0005] In view of this, the present invention aims to propose an image recognition-based method for detecting the uniformity of illite and pine needle oil loading, in order to solve the problem of inaccurate loading uniformity quantification caused by the coexistence of false growth leakage of aggregate gap shadows and early stoppage of weak oil film edges in submicron-level illite and pine needle oil composite powder microscopic images by a single gray-scale difference region growth algorithm.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] A method for detecting the uniformity of illite and pine needle oil loading based on image recognition, the method comprising:

[0008] Step S1: Acquire effective detection grayscale images and seed point sets by acquiring microscopic images of composite powder under side illumination conditions;

[0009] Step S2: Obtain the artifact suppression factor by jointly evaluating the local gradient and curvature response of the effectively detected grayscale image;

[0010] Step S3: Obtain the flux compensation factor by jointly evaluating the consistency of regional growth direction and local flux characteristics;

[0011] Step S4: The gray-scale difference determination cost is fused and modulated by the artifact suppression factor and the flux compensation factor to obtain the final growth determination cost and generate a binary mask for the load region.

[0012] Step S5: Obtain the load uniformity index by performing gridded load rate statistics and dispersion calculation on the binary mask of the load area.

[0013] Furthermore, the acquisition of effective detection grayscale images and seed point sets by collecting microscopic images of composite powder under side illumination conditions includes:

[0014] An industrial microscope camera and a side light source are set up on a microscope imaging platform. The incident angle of the side light source relative to the surface of the composite powder sample is fixed at 45 degrees. Microscopic imaging is performed on the spray-dried illite and pine needle oil composite powder sample to acquire the original color image of the composite powder. The original color image is then converted to grayscale to obtain the corresponding effective detection grayscale image. The pixels with the minimum grayscale value are searched in the effective detection grayscale image. The preset number of pixels with the lowest grayscale value are used as seed points, and the set of all seed points is used as the seed point set.

[0015] Furthermore, the method of obtaining the artifact suppression factor by jointly evaluating the local gradient and curvature response of the effectively detected grayscale image includes:

[0016] By extracting and normalizing the local gradient feature data and Hessian curvature response feature data of the effectively detected grayscale images, pseudo-effect response evaluation data is obtained.

[0017] The artifact suppression factor is obtained by applying punitive enhancement modulation and lower limit constraint processing to the artifact evaluation data.

[0018] Furthermore, the process of extracting and normalizing the local gradient feature data and Hessian curvature response feature data of the effectively detected grayscale image to obtain pseudo-effect evaluation data includes:

[0019] For any target pixel in a grayscale image that is effectively detected, the local gradient feature vector corresponding to the target pixel is obtained by calculating the grayscale difference between the target pixel and its adjacent pixels in the horizontal and vertical directions, and the magnitude of the local gradient feature vector is used as the gradient intensity of the target pixel for evaluation.

[0020] The Hessian curvature response feature data corresponding to the target pixel is obtained by calculating the second-order difference and cross difference of the target pixel in the horizontal and vertical directions. The transpose of the local gradient feature data matrix is ​​then multiplied sequentially with the Hessian curvature response feature data matrix and the local gradient feature data matrix to obtain the curvature projection response evaluation corresponding to the target pixel. The absolute value of the curvature projection response evaluation is used as the numerator, and the result of adding the square of the gradient intensity evaluation to the smallest positive number that is prevented from being divided by zero is used as the denominator. The resulting fraction is used as the normalized curvature response evaluation corresponding to the target pixel.

[0021] The reciprocal of the calculation result of adding the cosine value corresponding to the incident angle to the smallest positive number to prevent division by zero is used as the incident angle compensation coefficient; the calculation result of multiplying the curvature response normalization evaluation by the incident angle compensation coefficient of the side illumination is used as the pseudo-effect evaluation data of the target pixel.

[0022] Furthermore, the method of obtaining the artifact suppression factor by performing punitive enhancement modulation and lower limit constraint processing on the artifact suppression data includes:

[0023] Set the adjustment coefficient for the penalty enhancement modulation; for any target pixel in the effectively detected grayscale image, extract the artifact evaluation data corresponding to the target pixel, and use the result of multiplying the adjustment coefficient of the penalty enhancement modulation with the artifact evaluation data as the penalty enhancement modulation amount evaluation; use a constant 1 as the lower limit constraint benchmark, and use the result of adding the constant 1 with the penalty enhancement modulation amount evaluation as the artifact suppression factor corresponding to the target pixel.

[0024] Furthermore, the process of obtaining a flux compensation factor by jointly evaluating regional growth direction consistency and local flux characteristics includes:

[0025] By performing directional consistency analysis on the regional growth direction vector data and the local gradient direction data of the effectively detected grayscale image, growth direction consistency evaluation data is obtained.

[0026] By extracting divergence and curl features from the local gradient field data of the effectively detected images, local flux feature evaluation data is obtained.

[0027] By gating and upper limit constraint processing of growth direction consistency assessment data and local flux characteristic assessment data, flux compensation factor is obtained.

[0028] Furthermore, the step of performing directional consistency analysis on the region growth direction vector data and the local gradient direction data of the effectively detected grayscale image to obtain growth direction consistency evaluation data includes:

[0029] For any target pixel in a grayscale image that is effectively detected, the target seed point corresponding to the target pixel is determined from the seed point set, and the two-dimensional pixel coordinates of the target pixel and the two-dimensional pixel coordinates of the target seed point are obtained.

[0030] The difference between the two-dimensional pixel coordinates of the target pixel and the two-dimensional pixel coordinates of the target seed point is used as the region growth direction vector data corresponding to the target pixel. The magnitude of the region growth direction vector data is used as the normalization scale. The result of dividing the region growth direction vector data by the normalization scale is used as the unit region growth direction vector data corresponding to the target pixel.

[0031] The dot product operation is performed between the growth direction vector data of the unit region corresponding to the target pixel and the local gradient vector to obtain the direction consistency dot product evaluation of the target pixel. The result of multiplying the magnitude of the growth direction vector data of the region corresponding to the target pixel and the magnitude of the local gradient vector and adding the result to the smallest positive number that is prevented from being divided by zero is used as the direction consistency normalization scale of the target pixel. The result of dividing the direction consistency dot product evaluation of the target pixel by the direction consistency normalization scale is used as the direction consistency normalization evaluation of the target pixel. The direction consistency normalization evaluation of the target pixel is mapped through the hyperbolic tangent function, and the corresponding mapping result is used as the growth direction consistency evaluation of the target pixel.

[0032] Furthermore, the step of extracting divergence and curl features from the local gradient field data of the effectively detected image to obtain local flux feature evaluation data includes:

[0033] Define the local range of a pixel. For any target pixel in the effectively detected image, obtain the local gradient field of the target pixel through the gradient vector of the pixel within the local range of the target pixel. The local gradient field of the target pixel includes at least the horizontal gradient component and the vertical gradient component of the pixel.

[0034] The first-order difference calculation is performed on the horizontal gradient component in the horizontal direction and the first-order difference calculation is performed on the vertical gradient component in the vertical direction. The result of the sum of the two calculations is used as the divergence feature evaluation corresponding to the target pixel.

[0035] The gradient components in the vertical direction are calculated by first-order difference in the horizontal direction, and the gradient components in the horizontal direction are calculated by first-order difference in the vertical direction. The absolute value of the difference between the two is used as the curl feature evaluation of the target pixel.

[0036] Divide the curl feature evaluation corresponding to the target pixel by the calculation result of the curl feature evaluation to obtain the local flux feature evaluation data corresponding to the target pixel.

[0037] Furthermore, the process of obtaining the flux compensation factor by gating and upper limit constraint processing of growth direction consistency assessment data and local flux characteristic assessment data includes:

[0038] A compensation intensity coefficient is set. For any target pixel in the effectively detected image, a constant 1 is used as the gating benchmark. The result of subtracting the local flux feature evaluation data from the constant 1 is used as the flux gating coefficient corresponding to the target pixel. The compensation intensity coefficient, the growth direction consistency evaluation, and the flux gating coefficient are multiplied, and the negative of the calculation result is used as the exponential decay driving quantity. The exponential decay driving quantity is mapped through an exponential function with the natural constant as the base to obtain the flux compensation factor corresponding to the target pixel.

[0039] Furthermore, the step of fusing and modulating the gray-level difference determination cost through the artifact suppression factor and the flux compensation factor to obtain the final growth determination cost and generate a binary mask for the load region includes:

[0040] Set a threshold for the grayscale difference determination cost; for any target pixel in the effectively detected image, obtain the grayscale value of the target pixel and the grayscale value of the target seed point corresponding to the target pixel, and use the absolute value of the difference between the grayscale value of the target pixel and the grayscale value of the target seed point as the grayscale difference determination cost corresponding to the target pixel.

[0041] The result of multiplying the gray-level difference determination cost corresponding to the target pixel, the artifact suppression factor, and the flux compensation factor is used as the final growth determination cost corresponding to the target pixel.

[0042] When the final growth determination cost is less than the threshold of the gray-level difference determination cost, the target pixel is marked as a load region pixel and the load region binary mask is updated; when the final growth determination cost is greater than or equal to the preset gray-level difference determination cost threshold, the target pixel is marked as an empty region pixel and the load region binary mask remains unchanged; traverse all target pixels in the effectively detected gray-level image to obtain the load region binary mask.

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

[0044] The present invention discloses an image recognition-based method for detecting the uniformity of illite and pine needle oil loading. In the online detection scenario of submicron-sized illite and pine needle oil composite powder after spray drying, this solution introduces a curvature mutation-sensitive penalty mechanism for geometrically self-occluding pseudo-dark areas into the region growth determination cost. This effectively suppresses the high-frequency shadows formed by microscopic stacking gaps, even if they overlap with the wetting dark areas in grayscale, due to the sharp groove-type curvature characteristics of their grayscale surfaces. This significantly reduces the erroneous inclusion area in the loading region mask, reduces the false coverage increase and false dispersion fluctuation caused by the surface structure of agglomerates, and improves the robustness and consistency of the segmentation results to the differences in agglomerate morphology between different batches.

[0045] Meanwhile, this solution introduces a growth direction consistency and local flux quality gating compensation mechanism targeting the weak edges of the semi-volatile oil film. This prevents the nonlinear gradient decay caused by semi-transparent evaporation at the oil film edge from being misjudged as a termination boundary by a simple grayscale difference threshold. This allows for more continuous and complete extraction of the wetting transition zone, reducing underestimation of the load area and distribution breaks caused by early stoppages and missed detections. Based on a more realistic load mask, the statistical dispersion of the gridded local load rate can stably reflect the actual load uniformity differences, thereby improving the accuracy and traceability of online quality judgment and reducing the risk of process adjustment delays and batch rework caused by false alarms or missed alarms. Attached Figure Description

[0046] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0047] Figure 1 This is a flowchart illustrating a method for detecting the uniformity of illite and pine needle oil loading based on image recognition, as described in an embodiment of the present invention. Detailed Implementation

[0048] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0049] See Figure 1 This is a flowchart of a method for detecting the uniformity of illite and pine needle oil loading based on image recognition, as provided in Embodiment 1 of the present invention. Figure 1 As shown, a method for detecting the uniformity of illite and pine needle oil loading based on image recognition may include:

[0050] Step S1: Acquire effective detection grayscale images and seed point sets by acquiring microscopic images of composite powder under side lighting conditions.

[0051] An industrial microscope camera and a side light source are set up on a microscope imaging platform. The incident angle of the side light source relative to the surface of the composite powder sample is fixed at 45 degrees. Microscopic imaging is performed on the spray-dried illite and pine needle oil composite powder sample to acquire the original color image of the composite powder. The original color image is then converted to grayscale to obtain the corresponding effective detection grayscale image. The pixels with the minimum grayscale value are searched in the effective detection grayscale image. The preset number of pixels with the lowest grayscale value are used as seed points, and the set of all seed points is used as the seed point set.

[0052] Thus, the acquisition of effective detection grayscale images and seed point sets by collecting microscopic images of composite powder under side lighting conditions has been completed.

[0053] Step S2: Obtain the artifact suppression factor by jointly evaluating the local gradient and curvature response of the effectively detected grayscale image.

[0054] In the microscopic image recognition of submicron-sized illite and pine needle oil mixed powders, the core challenge lies in accurately distinguishing the geometric shadows between the pine needle oil-infiltrated areas and the aggregated particles. Because illite powder forms spherical aggregates after spray drying, these aggregates are not perfect spheres; their surfaces are rough structures composed of countless submicron-sized layers stacked together, and physical gaps exist between the aggregates. Under side illumination from an industrial camera, these deep physical gaps and stacking dead angles create high-frequency geometric self-occlusion shadows. These shadowed areas exhibit extremely low grayscale values ​​in the image, highly confusing with the light-absorbing pine needle oil-infiltrated areas in the grayscale dimension. This causes conventional algorithms to easily misjudge the dried gaps as oil spots, resulting in false positives in the detection results. To address this problem, this step utilizes the physical properties of liquid surface tension as a criterion. As a liquid with a certain viscosity, pine needle oil, when wetting the surface of powder, fills the microscopic gaps due to capillary action, resulting in a continuous and smooth surface morphology in the wetting area. Conversely, the gaps in dried agglomerates retain the original particle stacking structure, with their edges exhibiting sharp V-shaped grooves in micro-geometry, exhibiting significant local curvature abrupt changes. Therefore, by analyzing the second-order geometric features of the local gray-level surface of the image, an optimization factor is constructed to respond to the sharp groove structure and suppress the smooth liquid surface response, effectively intercepting dry shadows during the growth process and eliminating misjudgments.

[0055] In summary, this invention first extracts and normalizes the local gradient feature data and Hessian curvature response feature data of the effectively detected grayscale image to obtain pseudo-effect evaluation data. Specifically, for any target pixel in the effectively detected grayscale image, the local gradient feature vector corresponding to the target pixel is obtained based on the grayscale difference calculation of adjacent pixels in the horizontal and vertical directions, and the magnitude of the local gradient feature vector is used as the gradient intensity evaluation of the target pixel. The Hessian curvature response feature data corresponding to the target pixel is obtained through second-order difference and cross-difference calculations in the horizontal and vertical directions. The local gradient feature vector is then used as the gradient intensity evaluation of the target pixel. The transpose of the data matrix is ​​multiplied sequentially with the Hessian curvature response feature data matrix and the local gradient feature data matrix to obtain the curvature projection response evaluation corresponding to the target pixel. The absolute value of the curvature projection response evaluation is used as the numerator, and the result of adding the square of the gradient intensity evaluation to the smallest positive number to prevent division by zero is used as the denominator. The resulting fraction is used as the curvature response normalization evaluation corresponding to the target pixel. The reciprocal of the result of adding the cosine value of the incident angle to the smallest positive number to prevent division by zero is used as the incident angle compensation coefficient. The result of multiplying the curvature response normalization evaluation with the incident angle compensation coefficient of the side illumination is used as the pseudo-effect evaluation data of the target pixel.

[0056] After obtaining the artifact suppression data, the artifact suppression factor is obtained by further processing the artifact suppression data with penalized enhancement modulation and lower limit constraint. Specifically, the adjustment coefficient of penalized enhancement modulation is set. In this embodiment of the invention, the adjustment coefficient of penalized enhancement modulation is set to 30. This coefficient can be adjusted according to the actual scenario and is not required. For any target pixel in the effectively detected grayscale image, the artifact suppression data corresponding to the target pixel is extracted. The result of multiplying the adjustment coefficient of penalized enhancement modulation with the artifact suppression data is used as the penalized enhancement modulation amount evaluation. Using a constant 1 as the lower limit constraint benchmark, the result of adding the constant 1 with the penalized enhancement modulation amount evaluation is used as the artifact suppression factor corresponding to the target pixel.

[0057] In one implementation, assume the first The grayscale gradient vector of each pixel is ;No. The Hessian matrix of each pixel is ; Angle of incidence is The adjustment coefficient for the punitive enhancement modulation is: Then the first The expression for calculating the artifact suppression factor for each pixel is:

[0058]

[0059] in, Indicates the first The artifact suppression factor per pixel; The modulation coefficient representing the penalty enhancement modulation; Indicates the first The grayscale gradient vector of each pixel; Indicates the first Transpose of the grayscale gradient vector of each pixel; Indicates the first A Hessian matrix of pixels; Indicates the first The square of the magnitude of the grayscale gradient vector of each pixel; Angle representing the angle of incidence; The cosine value of the incident angle; The representation method is a very small positive number with a denominator of 0, which is set in the embodiments of the present invention. .

[0060] It should be noted that the optimization factor was designed to distinguish between dry cracks and wet surfaces. Firstly, to capture the sharp geometric features unique to dry cracks, a core term was introduced into the formula. This term mathematically represents the projection of the second derivative of the image's grayscale surface along the direction of the steepest gradient. Dry aggregates exhibit deep V-shaped gaps, with their grayscale values ​​undergoing dramatic drops and rises across the gap edges, resulting in a very high second derivative value in that direction. In contrast, the pine needle oil-soaked area, due to the surface tension of the liquid, presents a smooth U-shape or plane, with a relatively small second derivative value along the gradient. Therefore, this term is used to quantify the sharpness of local regions. Secondly, to eliminate the interference of local contrast differences in the image on geometric shape judgment, a denominator term is included in the formula. A normalization process is achieved by dividing the projection of the second derivative by the square of the gradient magnitude, ensuring the factorization... It only responds to the geometric topology of the image, i.e., whether it is sharp enough, and is not affected by fluctuations in the absolute gray value of the gap itself. This means that regardless of whether the gap is dark black or light gray, as long as its geometric structure is sharp, it can be identified. Finally, considering that shadow formation is closely related to lighting conditions, the formula introduces an illumination angle compensation term. Under side lighting, the slit structure perpendicular to the light source will produce the deepest shadow. The tendency towards a smaller value increases the overall compensation term, further amplifying the penalty for this type of shadow most easily confused. In summary, when the algorithm traverses the gaps in dry aggregates, although their grayscale values ​​may be very low, their extremely high second derivative along the gradient results in... The value will be significantly greater than This drastically increases the growth cost, forcing the algorithm to stop growing in that area; however, for real oil regions, due to their smooth surface, The value remains at The surrounding area allows the algorithm to grow normally.

[0061] Thus, the artifact suppression factor was obtained by jointly evaluating the local gradient and curvature response of the effectively detected grayscale image.

[0062] Step S3: Obtain the flux compensation factor by jointly evaluating the consistency of regional growth direction and local flux characteristics.

[0063] In step S2, after successfully eliminating the interference of high-frequency drying gaps by introducing a geometric curvature artifact suppression factor, the region growth algorithm can accurately identify the obvious oil core area. However, this process also tightens the geometric constraints of the growth criterion, thus exposing the contradiction in the identification of the pine needle oil loading edge. Because the pine needle oil undergoes a dynamic penetration process accompanied by volatilization on the surface of submicron illite powder, the outermost edge of its loading area is not a distinct boundary, but a semi-wetting transition zone formed by the micron-level attenuation of the oil film thickness.

[0064] Within this transition zone, as the oil film thins, the leveling effect relied upon in step S2 gradually weakens, meaning the curvature features are no longer significantly smooth, and the grayscale value gradually approaches the dry background due to substrate transmission. At this point, if the conventional judgment logic is continued or only the geometric constraints of step S2 are relied upon, the algorithm will prematurely terminate growth before fully covering the transition zone due to increased grayscale difference and a lack of strong geometric smoothness, resulting in a detected load area smaller than the actual wetting area—a premature termination phenomenon. Therefore, based on the removal of false targets in step S2, a proactive compensation mechanism based on liquid diffusion dynamics is needed to address this weak edge signal. Utilizing the radial flow field consistency unique to the diffusion process, hidden vector direction information is extracted as new growth momentum in edge regions where grayscale and geometric features are insignificant, thereby achieving detection from core region confirmation to edge region completion.

[0065] In summary, this invention first performs direction consistency analysis on the region growth direction vector data and the local gradient direction data of the effectively detected grayscale image to obtain growth direction consistency evaluation data. Specifically, for any target pixel in the effectively detected grayscale image, a target seed point corresponding to the target pixel is determined from the seed point set, and the two-dimensional pixel coordinates of the target pixel and the target seed point are obtained. The difference between the two-dimensional pixel coordinates of the target pixel and the target seed point is used as the region growth direction vector data corresponding to the target pixel. The magnitude of the region growth direction vector data is used as the normalization scale, and the result of dividing the region growth direction vector data by the normalization scale is used as the unit corresponding to the target pixel. The region growth direction vector data is used as follows: The dot product operation is performed between the unit region growth direction vector data corresponding to the target pixel and the local gradient vector to obtain the direction consistency dot product evaluation for the target pixel. The result of multiplying the magnitude of the region growth direction vector data corresponding to the target pixel by the magnitude of the local gradient vector and adding it to the smallest positive number (avoiding division by zero) is used as the direction consistency normalization scale for the target pixel. The direction consistency dot product evaluation for the target pixel is divided by the direction consistency normalization scale to obtain the direction consistency normalization evaluation for the target pixel. The direction consistency normalization evaluation for the target pixel is mapped using the hyperbolic tangent function, and the resulting mapping is used as the growth direction consistency evaluation for the target pixel.

[0066] After obtaining the growth direction consistency assessment, the divergence and curl features of the local gradient field data of the effective detection image are extracted to obtain local flux feature assessment data. Specifically, the local range of the pixel is defined. In this embodiment of the invention, the local range of the pixel is defined as follows: For any target pixel in the effectively detected image, the local gradient field of the target pixel is obtained through the gradient vector of the pixels within a local range. The local gradient field of the target pixel includes at least the horizontal and vertical gradient components of the pixel. The first-order difference is calculated for the horizontal gradient component in the horizontal direction and the first-order difference is calculated for the vertical gradient component in the vertical direction. The sum of the two calculations is used as the divergence feature evaluation corresponding to the target pixel. The first-order difference is calculated for the vertical gradient component in the horizontal direction and the first-order difference is calculated for the horizontal gradient component in the vertical direction. The absolute value of the difference between the two calculations is used as the curl feature evaluation corresponding to the target pixel. The curl feature evaluation corresponding to the target pixel is divided by the divergence feature evaluation result as the local flux feature evaluation data corresponding to the target pixel.

[0067] After obtaining the local flux feature evaluation data, the flux compensation factor is obtained by gating and upper limit constraint processing of the growth direction consistency evaluation data and the local flux feature evaluation data. Specifically, a compensation intensity coefficient is set. In this embodiment of the invention, the compensation intensity coefficient is set to 2. For any target pixel in the effectively detected image, a constant 1 is used as the gating benchmark. The result of subtracting the local flux feature evaluation data from the constant 1 is used as the flux gating coefficient corresponding to the target pixel. The compensation intensity coefficient, the growth direction consistency evaluation data and the flux gating coefficient are multiplied together, and the negative of the calculation result is used as the exponential decay driving quantity. The exponential decay driving quantity is mapped through an exponential function with the natural constant as the base to obtain the flux compensation factor corresponding to the target pixel.

[0068] In one implementation, a seed point is assumed. Pointing to the The growth direction vector of the unit region of each pixel is: ;No. The local gradient field of a local region of pixels is ;No. The divergence feature corresponding to each pixel is evaluated as follows: ;No. The curl evaluation for each pixel is: The compensation strength coefficient is Then the first The formula for calculating the flux compensation factor corresponding to each pixel is:

[0069]

[0070] in, Indicates the first The flux compensation factor corresponding to each pixel; Indicates the compensation strength coefficient; Represents seed point Pointing to the The growth direction vector of a unit region for each pixel; Indicates the first Local gradient field of a local region of a pixel; Indicates the first Curl evaluation corresponding to each pixel; Indicates the first Evaluation of divergence features corresponding to each pixel; Indicates the magnitude of the vector; This represents an exponential function with the natural constant e as the base. This represents the hyperbolic tangent function.

[0071] It should be noted that, firstly, considering the characteristic of blurred gray-scale features but clear diffusion direction in the edge region, the first part of the formula... The orientation alignment is detected using vector dot product. In the transition zone, although the thinning of the oil film weakens the filling effect, the brightness change trend still strictly follows the physical diffusion law of brightening from the center outwards. At this time, the gradient direction is highly aligned with the growth direction, and the dot product is close to... After mapping with the hyperbolic tangent function, a baseline compensation is provided, telling the algorithm that although the grayscale appears light and uneven, its direction of change is correct. Secondly, to prevent this compensation mechanism from being falsely triggered when encountering dry background textures with coincident orientations, i.e., to avoid disrupting the steps... To mitigate the effects, the formula introduces a laminar flow quality filter: This is the key factor in distinguishing between ordered diffusion and random texture. Real liquid permeation fields exhibit laminar flow characteristics of high divergence and low vortex, causing this ratio to approach [a certain value]. Full compensation is allowed; however, the rough texture of the dry substrate, even when oriented, exhibits high twist due to the disorder of its microstructure, leading to an increase in this ratio and thus automatically closing the compensation channel. In summary, As a supplement to step S2, the factor specifically takes over the edge regions where geometric features fail. Under the premise of confirming that the local flow field conforms to the physical diffusion law (unidirectional and low-cyclonic), it actively lowers the judgment threshold and reintegrates the weak edge signals that would otherwise be truncated by step S2 or the threshold into the load region, thereby achieving complete coverage of the entire wetting process.

[0072] Thus, the flux compensation factor was obtained by jointly evaluating the consistency of regional growth direction and local flux characteristics.

[0073] Step S4: The gray-scale difference determination cost is fused and modulated by the artifact suppression factor and the flux compensation factor to obtain the final growth determination cost and generate a binary mask for the load region.

[0074] After obtaining the artifact suppression factor and flux compensation factor corresponding to the target pixel, a threshold for gray-level difference determination cost is set. In this embodiment of the invention, the threshold for gray-level difference determination cost is set to 20. For any target pixel in the effectively detected image, the gray value of the target pixel and the gray value of the target seed point corresponding to the target pixel are obtained. The absolute value of the difference between the gray value of the target pixel and the gray value of the target seed point is used as the gray-level difference determination cost corresponding to the target pixel. The result of multiplying the gray-level difference determination cost, the artifact suppression factor, and the flux compensation factor corresponding to the target pixel is used as the final growth determination cost corresponding to the target pixel. When the final growth determination cost is less than the threshold for gray-level difference determination cost, the target pixel is marked as a load region pixel and the load region binary mask is updated. When the final growth determination cost is greater than or equal to the preset gray-level difference determination cost threshold, the target pixel is marked as an empty region pixel and the load region binary mask remains unchanged. All target pixels in the effectively detected gray-level image are traversed to obtain the load region binary mask.

[0075] Thus, the gray-scale difference determination cost is fused and modulated using the artifact suppression factor and flux compensation factor to obtain the final growth determination cost and generate a binary mask for the load region.

[0076] Step S5: Obtain the load uniformity index by performing gridded load rate statistics and dispersion calculation on the binary mask of the load area.

[0077] After extracting the load region of the entire image in step S4, a binarized load distribution mask image is obtained. To objectively evaluate the load uniformity of pine needle oil on the surface of submicron illite powder, it is necessary to quantify the spatial dispersion rather than solely relying on the total coverage. This step employs a gridded variance analysis method: the binarized mask image is divided into... Equal sub-grid regions (in this embodiment) );

[0078] Calculate the percentage of pine needle oil-loaded pixels within each sub-grid region (i.e., local load rate). );

[0079] Calculate the standard deviation of the local load rate for all subgrids. The formula is:

[0080]

[0081] in This represents the average load factor across the entire graph. Finally, it is expressed as the standard deviation. As an indicator of uniformity, The smaller the value, the more uniform the pine needle oil is dispersed on the illite surface and the better the loading effect; conversely, it indicates the presence of local agglomeration or missed coating.

[0082] Thus, the load uniformity index was obtained by performing gridded load rate statistics and dispersion calculation on the binary mask of the load region.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting the uniformity of illite and pine needle oil loading based on image recognition, characterized in that, The method includes: Step S1: Acquire effective detection grayscale images and seed point sets by acquiring microscopic images of composite powder under side illumination conditions; Step S2: Obtain the artifact suppression factor by jointly evaluating the local gradient and curvature response of the effectively detected grayscale image; Step S3: Obtain the flux compensation factor by jointly evaluating the consistency of regional growth direction and local flux characteristics; Step S4: The gray-scale difference determination cost is fused and modulated by the artifact suppression factor and the flux compensation factor to obtain the final growth determination cost and generate a binary mask for the load region. Step S5: Obtain the load uniformity index by performing gridded load rate statistics and dispersion calculation on the binary mask of the load area.

2. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 1, characterized in that, The process involves acquiring microscopic images of composite powder under side-illuminated conditions to obtain effective detection grayscale images and seed point sets, including: An industrial microscope camera and a side light source are set up on a microscope imaging platform. The incident angle of the side light source relative to the surface of the composite powder sample is fixed at 45 degrees. Microscopic imaging is performed on the spray-dried illite and pine needle oil composite powder sample to acquire the original color image of the composite powder. The original color image is then converted to grayscale to obtain the corresponding effective detection grayscale image. The pixels with the minimum grayscale value are searched in the effective detection grayscale image. The preset number of pixels with the lowest grayscale value are used as seed points, and the set of all seed points is used as the seed point set.

3. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 1, characterized in that, The method of obtaining an artifact suppression factor by jointly evaluating the local gradient and curvature response of the effectively detected grayscale image includes: By extracting and normalizing the local gradient feature data and Hessian curvature response feature data of the effectively detected grayscale images, pseudo-effect response evaluation data is obtained. The artifact suppression factor is obtained by applying punitive enhancement modulation and lower limit constraint processing to the artifact evaluation data.

4. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 3, characterized in that, The process involves extracting and normalizing the local gradient feature data and Hessian curvature response feature data of the effectively detected grayscale image to obtain pseudo-effect evaluation data, including: For any target pixel in a grayscale image that is effectively detected, the local gradient feature vector corresponding to the target pixel is obtained by calculating the grayscale difference between the target pixel and its adjacent pixels in the horizontal and vertical directions, and the magnitude of the local gradient feature vector is used as the gradient intensity of the target pixel for evaluation. The Hessian curvature response feature data corresponding to the target pixel is obtained by calculating the second-order difference and cross difference of the target pixel in the horizontal and vertical directions. The transpose of the local gradient feature data matrix is ​​then multiplied sequentially with the Hessian curvature response feature data matrix and the local gradient feature data matrix to obtain the curvature projection response evaluation corresponding to the target pixel. The absolute value of the curvature projection response evaluation is used as the numerator, and the result of adding the square of the gradient intensity evaluation to the smallest positive number that is prevented from being divided by zero is used as the denominator. The resulting fraction is used as the normalized curvature response evaluation corresponding to the target pixel. The reciprocal of the calculation result of adding the cosine value corresponding to the incident angle to the smallest positive number to prevent division by zero is used as the incident angle compensation coefficient; the calculation result of multiplying the curvature response normalization evaluation by the incident angle compensation coefficient of the side illumination is used as the pseudo-effect evaluation data of the target pixel.

5. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 3, characterized in that, The method of obtaining an artifact suppression factor by performing punitive enhancement modulation and lower limit constraint processing on the artifact suppression data includes: Set the adjustment coefficient for the penalty enhancement modulation; for any target pixel in the effectively detected grayscale image, extract the artifact evaluation data corresponding to the target pixel, and use the result of multiplying the adjustment coefficient of the penalty enhancement modulation with the artifact evaluation data as the penalty enhancement modulation amount evaluation; use a constant 1 as the lower limit constraint benchmark, and use the result of adding the constant 1 with the penalty enhancement modulation amount evaluation as the artifact suppression factor corresponding to the target pixel.

6. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 1, characterized in that, The process of obtaining a flux compensation factor by jointly evaluating regional growth direction consistency and local flux characteristics includes: By performing directional consistency analysis on the regional growth direction vector data and the local gradient direction data of the effectively detected grayscale image, growth direction consistency evaluation data is obtained. By extracting divergence and curl features from the local gradient field data of the effectively detected images, local flux feature evaluation data is obtained. By gating and upper limit constraint processing of growth direction consistency assessment data and local flux characteristic assessment data, flux compensation factor is obtained.

7. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 6, characterized in that, The process involves performing directional consistency analysis on the region growth direction vector data and the local gradient direction data of the effectively detected grayscale image to obtain growth direction consistency evaluation data, including: For any target pixel in a grayscale image that is effectively detected, the target seed point corresponding to the target pixel is determined from the seed point set, and the two-dimensional pixel coordinates of the target pixel and the two-dimensional pixel coordinates of the target seed point are obtained. The difference between the two-dimensional pixel coordinates of the target pixel and the two-dimensional pixel coordinates of the target seed point is used as the region growth direction vector data corresponding to the target pixel. The magnitude of the region growth direction vector data is used as the normalization scale. The result of dividing the region growth direction vector data by the normalization scale is used as the unit region growth direction vector data corresponding to the target pixel. The dot product operation is performed between the growth direction vector data of the unit region corresponding to the target pixel and the local gradient vector to obtain the direction consistency dot product evaluation of the target pixel. The result of multiplying the magnitude of the growth direction vector data of the region corresponding to the target pixel and the magnitude of the local gradient vector and adding the result to the smallest positive number that is prevented from being divided by zero is used as the direction consistency normalization scale of the target pixel. The result of dividing the direction consistency dot product evaluation of the target pixel by the direction consistency normalization scale is used as the direction consistency normalization evaluation of the target pixel. The direction consistency normalization evaluation of the target pixel is mapped through the hyperbolic tangent function, and the corresponding mapping result is used as the growth direction consistency evaluation of the target pixel.

8. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 6, characterized in that, The process of extracting divergence and curl features from the local gradient field data of the effectively detected images to obtain local flux feature evaluation data includes: Define the local range of a pixel. For any target pixel in the effectively detected image, obtain the local gradient field of the target pixel through the gradient vector of the pixel within the local range of the target pixel. The local gradient field of the target pixel includes at least the horizontal gradient component and the vertical gradient component of the pixel. The first-order difference calculation is performed on the horizontal gradient component in the horizontal direction and the first-order difference calculation is performed on the vertical gradient component in the vertical direction. The result of the sum of the two calculations is used as the divergence feature evaluation corresponding to the target pixel. The gradient components in the vertical direction are calculated by first-order difference in the horizontal direction, and the gradient components in the horizontal direction are calculated by first-order difference in the vertical direction. The absolute value of the difference between the two is used as the curl feature evaluation of the target pixel. Divide the curl feature evaluation corresponding to the target pixel by the calculation result of the curl feature evaluation to obtain the local flux feature evaluation data corresponding to the target pixel.

9. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 6, characterized in that, The process of obtaining a flux compensation factor by gating and fusion of growth direction consistency assessment data and local flux characteristic assessment data and applying upper limit constraints includes: A compensation intensity coefficient is set. For any target pixel in the effectively detected image, a constant 1 is used as the gating benchmark. The result of subtracting the local flux feature evaluation data from the constant 1 is used as the flux gating coefficient corresponding to the target pixel. The compensation intensity coefficient, the growth direction consistency evaluation, and the flux gating coefficient are multiplied, and the negative of the calculation result is used as the exponential decay driving quantity. The exponential decay driving quantity is mapped through an exponential function with the natural constant as the base to obtain the flux compensation factor corresponding to the target pixel.

10. The method for detecting the uniformity of illite and pine needle oil loading based on image recognition according to claim 1, characterized in that, The step of fusing and modulating the gray-level difference determination cost through an artifact suppression factor and a flux compensation factor to obtain the final growth determination cost and generate a binary mask for the load region includes: Set a threshold for the grayscale difference determination cost; for any target pixel in the effectively detected image, obtain the grayscale value of the target pixel and the grayscale value of the target seed point corresponding to the target pixel, and use the absolute value of the difference between the grayscale value of the target pixel and the grayscale value of the target seed point as the grayscale difference determination cost corresponding to the target pixel. The result of multiplying the gray-level difference determination cost corresponding to the target pixel, the artifact suppression factor, and the flux compensation factor is used as the final growth determination cost corresponding to the target pixel. When the final growth determination cost is less than the threshold of the gray-level difference determination cost, the target pixel is marked as a load region pixel and the load region binary mask is updated; when the final growth determination cost is greater than or equal to the preset gray-level difference determination cost threshold, the target pixel is marked as an empty region pixel and the load region binary mask remains unchanged; traverse all target pixels in the effectively detected gray-level image to obtain the load region binary mask.

Images