Calcium carbonate master batch wind shear forming detection system based on machine vision

Abstract

The invention relates to the technical field of image analysis, in particular to a calcium carbonate master batch wind shear forming detection system based on machine vision. A calcium carbonate master batch wind shear forming detection system based on machine vision comprises an edge pixel collection module, a boundary trend recognition module, a distortion track repair module, an internal texture analysis module and a forming anomaly detection module. According to the method, the gray difference and gradient angle of continuous pixels are calculated, an edge distribution diagram is constructed, boundary detail expression is enhanced, the pixel arrangement direction and gray change analysis are combined, continuous edge line segments are extracted, the boundary recognition precision is improved, a closed contour is repaired according to line segment connection offset, and structural integrity is ensured; the method comprises the following steps: calculating a gray change rate in an internal region block, drawing a texture fluctuation map, accurately marking a sudden change region, realizing abnormal particle identification, fusing pixel-level gray features and region structure analysis, and improving the precision and integrity of form perception and defect identification.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and particularly to a calcium carbonate masterbatch air-shearing forming detection system based on machine vision. Background Art

[0002] The technical field of image analysis includes collecting, processing, and analyzing image data to achieve the recognition and understanding of the state, characteristics, form, and changes of objects. The core contents of this technical field include multiple links such as image preprocessing, target detection, feature extraction, image segmentation, morphological analysis, and target recognition. Image analysis technology relies on cross-disciplines such as computer vision, pattern recognition, and deep learning, and is widely used in multiple application scenarios such as industrial inspection, medical imaging, security monitoring, intelligent manufacturing, and autonomous driving. In the process of industrial production, image analysis technology can automatically identify the surface structure, size, form, etc. of products, providing important data support for production quality control.

[0003] Among them, the calcium carbonate masterbatch air-shearing forming detection system based on machine vision refers to collecting real-time images of calcium carbonate masterbatch during the air-shearing forming process by building an optical imaging unit, and using means such as image filtering, edge detection, contour extraction, and connected component analysis to determine the morphological size distribution state of masterbatch particles. This detection system constructs a two-dimensional image feature space, combines morphological operations to achieve accurate extraction of particle contours during the air-shearing forming process, and identifies particle boundaries based on threshold segmentation and region growing methods, thereby completing the determination of forming quality and data output.

[0004] In the practice of existing technologies, image analysis focuses on contour localization and target detection, and rarely involves fine-grained analysis of the gray-scale change and gradient direction of edge pixels, resulting in insufficient performance of boundary details. Particle edge recognition is often interfered by image noise and boundary blur, and it is difficult to obtain stable structural information. Contour analysis usually focuses on connected components, lacking an analysis mechanism for weak offsets between boundary line segments, which easily leads to contour closure interruption or misconnection phenomena, reducing the structural restoration degree. The detection methods for internal particle textures generally rely on indicators such as the mean or range of the gray-scale values of the entire image, ignoring the local region change rate, and are not conducive to depicting the microscopic mutation characteristics inside the structure. In practical applications, some particles may form boundary micro-deviations or local anomalies in internal textures due to air-shearing disturbances. Existing means cannot effectively distinguish such detail deviations, resulting in misjudgments or omissions in the quality assessment process, affecting the accuracy of detection results and the timeliness of production control. Summary of the Invention

[0005] The purpose of the present invention is to solve the drawbacks existing in the prior art, and to propose a calcium carbonate masterbatch air-shearing forming detection system based on machine vision.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A calcium carbonate masterbatch air-shearing forming detection system based on machine vision includes: The edge pixel collection module extracts the particle edge image area, detects continuous edge pixel points, calculates the sequence length, gray level range, gray level difference, and gradient angle, comprehensively reflects the distribution characteristics of edge pixels, and generates an edge pixel distribution map; The boundary trend recognition module, based on the edge pixel distribution map, analyzes the pixel arrangement direction and gray level change, judges the correlation between direction offset and gray level change, extracts edge line segments that meet continuity and direction consistency and their start and end coordinates, and generates a boundary trend map; The distortion trajectory repair module analyzes the connection offset of adjacent line segments in the boundary trend map, calculates the normal and tangential position differences of the line segment endpoints, and corrects the offset line segments according to the overall boundary trend to generate a boundary contour map; The internal texture analysis module, based on the boundary contour map, extracts the internal gray level information of the particle, calculates the gray level distribution range and change rate of the region in blocks, analyzes the gray level fluctuation trend, draws the internal texture structure of the particle, and generates a texture distribution map; The forming abnormality detection module, based on the texture distribution map, detects the gray level change rate of the region, identifies the gray level mutation in the continuous region, and generates an air-shearing forming detection result.

[0007] As a further solution of the present invention, the edge pixel distribution map includes pixel sequence length characteristics, gray level value distribution range, adjacent pixel gray level difference, and pixel gradient angle characteristics. The boundary trend map includes edge line segment direction distribution, line segment start and end coordinates, line segment continuity characteristics, and direction offset. The boundary contour map includes line segment connection relationship, overall boundary trend, normal position deviation, and tangential position deviation. The texture distribution map includes gray level distribution structure, regional gray level change rate, gray level fluctuation trend, and texture structure distribution. The air-shearing forming detection result includes gray level mutation region, continuous abnormal region, abnormal region marking information, and overall abnormal structure of the particle.

[0008] As a further solution of the present invention, the edge pixel collection module includes: The edge extraction sub-module obtains the particle edge image area, detects the positions of continuous edge pixel points, screens the pixel points with gray level mutation as edge points, records the image coordinates and gray level values, and generates an edge pixel point set; The sequence construction sub-module, according to the edge pixel point set, constructs a pixel sequence according to the adjacent position relationship, screens the sequence length greater than the connection threshold, and counts the gray level values of each segment of the sequence to generate a gray level change interval list; The angle calculation sub-module calculates the gradient angle of each pair of pixels based on the difference in adjacent pixel coordinates and grayscale in each sequence of the grayscale change interval list, accumulates the mean of all gradient angles, and generates an edge pixel distribution map; The connection threshold quantization is defined as: the connection threshold is the minimum number of consecutive edge pixel points in the sequence, usually set to 3 - 5 pixel lengths.

[0009] As a further aspect of the present invention, the boundary trend recognition module includes: The direction analysis sub-module obtains the edge pixel distribution map, analyzes the connection direction of consecutive pixel points according to the arrangement order of pixels in the image, records the direction change trend, and generates a pixel arrangement direction sequence; The grayscale difference judgment sub-module calculates the grayscale difference value between adjacent pixel positions according to the pixel arrangement direction sequence, judges the degree of difference between the grayscale change value and the direction change amount, filters out the pixel segments with difference values lower than the set continuity threshold, and obtains a continuous edge segment set; The trend extraction sub-module calls the continuous edge segment set, extracts the start and end coordinates of each segment of edge pixels, splices them in order to generate an edge path, arranges the path structure information, and generates a boundary trend map; The degree of difference = |grayscale change value / direction change amount|.

[0010] As a further aspect of the present invention, the distortion trajectory repair module includes: The endpoint deviation extraction sub-module, based on the boundary trend map, selects the connection positions of adjacent line segments, calls the endpoint coordinate information and the corresponding normal and tangential direction data, calculates the normal difference and tangential difference of the connection endpoints, analyzes the offset trend, and obtains the endpoint direction deviation value; The connection offset recognition sub-module compares the difference magnitudes in the continuous deviation sequence according to the endpoint direction deviation value, filters out the line segment set with deviations higher than the direction threshold, forms an offset interval index, and obtains the offset paragraph interval value; The closed contour generation sub-module calls the offset paragraph interval value, repairs the connection relationship of the associated boundary line segments, adjusts the boundary structure, recalculates the angle difference, endpoint offset amount, and segment length change amount of the repaired boundary, calculates the connection reconstruction error degree through operations, and combines the minimum error structure reconnection order to generate a boundary contour map; The direction threshold is determined according to the statistical distribution of the normal and tangential angle changes during the natural connection of edge line segments, and usually ranges from 5° to 15°.

[0011] As a further aspect of the present invention, the internal texture analysis module includes: The grayscale acquisition sub-module extracts the internal area of the particle based on the boundary contour map, divides it into equal-area sub-blocks, obtains the pixel grayscale values in the sub-blocks, counts the peak value, low value, and average value, and obtains the regional grayscale distribution range value; The grayscale fluctuation calculation sub-module calculates the grayscale difference and average difference of the sub-blocks according to the regional grayscale distribution range value, judges the grayscale change trend between adjacent sub-blocks, and obtains the regional grayscale change rate; The texture map generation sub-module calls the regional grayscale change rate, locates the position where the grayscale change rate mutates, combines the grayscale trend of the sub-blocks, draws the overall grayscale hierarchy, and generates the texture distribution map; The specific division of the equal-area sub-blocks uses an 8×8 pixel grid division.

[0012] As a further solution of the present invention, the forming abnormality detection module includes: The grayscale trend recognition sub-module reads the grayscale change values of the particle area block by block based on the texture distribution map, compares the growth amplitude of the grayscale difference between adjacent areas, extracts the grayscale mutation direction and continuous position, and obtains the continuous grayscale mutation trend value; The mutation area marking sub-module calls the continuous grayscale mutation trend value, detects the continuous particle numbers and coordinate positions in the mutation section, records the range of the grayscale change abnormal area, constructs a particle local marking layer, and obtains the particle mutation marking interval value; The structure screening and judgment sub-module screens the particle structure images corresponding to the grayscale mutation area according to the particle mutation marking interval value, judges whether there is a mutation convergence phenomenon in the whole structure, outputs the structure mutation intensity level, and generates the wind shear forming detection result.

[0013] As a further solution of the present invention, the determination criterion for the mutation convergence phenomenon is that more than 3 adjacent sub-blocks simultaneously undergo grayscale mutation.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, through the calculation of the grayscale difference and gradient angle of continuous pixels, an edge distribution map is constructed to enhance the expression of boundary details. Combining the pixel arrangement direction and grayscale change analysis, continuous edge segments are extracted to improve the boundary recognition accuracy. The closed contour is repaired according to the line segment connection offset to ensure the structural integrity. The grayscale change rate is calculated for the internal area in blocks, and the texture fluctuation map is drawn. The mutation area is accurately marked to realize the recognition of abnormal particles. By integrating pixel-level grayscale features and regional structure analysis, the accuracy and integrity of morphology perception and defect recognition are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is the system flow chart of the present invention; Figure 2Flow chart of the edge pixel collection module of the present invention; Figure 3 Flow chart of the boundary trend recognition module of the present invention; Figure 4 Flow chart of the distorted trajectory repair module of the present invention; Figure 5 Flow chart of the internal texture analysis module of the present invention; Figure 6 Flow chart of the molding abnormality detection module of the present invention. Specific implementation manners

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. are based on the orientation or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0018] Please refer to Figure 1 , a calcium carbonate masterbatch air-shearing molding detection system based on machine vision includes: The edge pixel collection module acquires the particle edge image area, detects the positions of continuous edge pixels, arranges them to form a pixel sequence, calculates the sequence length, statistically analyzes the gray value range of the sequence, calculates the gray difference between adjacent pixels, calculates the gradient angle between pixel positions in the sequence, and generates an edge pixel distribution map; The boundary trend recognition module analyzes the arrangement direction of continuous pixels based on the edge pixel distribution map, calculates the gray value change between pixel positions, judges the distribution difference between the gray value change and the direction offset value, filters out the edge segments that meet the continuity condition, extracts the start and end coordinates of the edge segments, and generates a boundary trend map; The distorted trajectory repair module analyzes the offset direction at the connection position of adjacent line segments based on the boundary trend map, calculates the normal and tangential position deviations of the line segment endpoints, identifies and repairs the offset line segment area based on the overall distribution of the particle boundary, and generates a boundary contour map by connecting the repaired line segments; The internal texture analysis module obtains the image grayscale values of the internal area of the particles based on the boundary contour map, calculates the grayscale distribution range of each area in blocks, calculates the regional grayscale change rate, analyzes the grayscale fluctuation trend, and draws the grayscale distribution structure diagram to generate the texture distribution map; Based on the texture distribution map, the forming anomaly detection module detects the regional grayscale change rate block by block, identifies the grayscale mutation trend of continuous areas, marks the particle areas where mutations occur, screens the parts where the overall structure grayscale of the particles mutates, and generates the air shear forming detection result.

[0019] The edge pixel distribution map includes the pixel sequence length feature, the grayscale value distribution range, the grayscale difference between adjacent pixels, and the pixel gradient angle feature. The boundary orientation map includes the edge line segment direction distribution, the start and end coordinates of the line segment, the line segment continuity feature, and the direction offset. The boundary contour map includes the line segment connection relationship, the overall boundary orientation, the normal position deviation, and the tangential position deviation. The texture distribution map includes the grayscale distribution structure, the regional grayscale change rate, the grayscale fluctuation trend, and the texture structure distribution. The air shear forming detection result includes the grayscale mutation area, the continuous abnormal area, the abnormal area marking information, and the overall abnormal structure of the particles.

[0020] Please refer to Figure 2 , the edge pixel collection module includes: The edge extraction sub-module obtains the particle edge image area, detects the positions of edge continuous pixel points, screens the pixel points with significant grayscale mutations as edge points, records the image coordinates and grayscale values, and generates the edge pixel point set; The edge extraction sub-module obtains the particle edge image area. First, it selects the target particle area from the grayscale image. For example, in a microscopic image with a resolution of 640×480, the coordinate range [(120, 150)-(520, 450)] of the particle area is selected through the set region of interest window. For all pixel points within this range, the grayscale value changes of each row are traversed. For each pixel point , the grayscale values of its 4 neighboring pixels in the up, down, left, and right directions are called and the difference calculation is performed with the current pixel grayscale value. The specific execution is as follows: Let the current pixel grayscale value be , and the grayscale values of its upper, lower, left, and right neighborhood pixels are , , , , then the grayscale differences in the four directions are , , , . According to the definition of "significant grayscale mutation" in the judgment condition as the difference ≥ 5, the current pixel does not meet the requirement to be an edge point; for another pixel point , the grayscale is , the pixel grayscale values above, below, to the left, and to the right of it are 70, 72, 67, and 68 respectively. The differences are 20, 18, 23, and 22 respectively, which meet the mutation judgment condition. Thus, this point is marked as an edge point, and the execution result is recorded as: edge point coordinates (260, 300), grayscale value 90. Repeat the above process to perform pixel difference judgment and extraction on the entire image. The set of all pixel points that meet the mutation criteria is called the "edge pixel point set", and each point item contains ternary data , in this example, a total of 1354 edge pixels are extracted after processing.

[0021] The sequence construction sub-module constructs a pixel sequence according to the edge pixel point set, in accordance with the adjacent position relationship, screens the sequence length greater than the connection threshold, and counts the grayscale values of each segment of the sequence to generate a list of grayscale change intervals; The sequence construction sub-module constructs a pixel sequence according to the above edge pixel point set. The specific operation is to first sort the pixel points according to the image coordinates in a line-by-line scanning manner, traverse from top to bottom and from left to right, and sequentially judge whether there are other edge points in the eight-neighborhood of each pixel point. If so, connect this point with the adjacent point to form a sequence unit. For example, if points A (120, 150), B (121, 150), and C (122, 151) all exist in the edge point set and the grayscale mutation is satisfied, then construct the sequence A→B→C, and continue to search along the connection direction until it cannot be extended; the length of each pixel sequence is accumulated through the Euclidean distance. For example, the distance from A to B is 1 pixel, and the distance from B to C is , and the cumulative length of the entire sequence is , if the length of this sequence ≥ the connection threshold, set the connection threshold to 5 pixels, that is, if the cumulative length ≥ 5, then retain this sequence, otherwise discard it; the threshold setting method is based on the minimum connected length of the edge coherent segments of the statistical image at the visually recognizable level. This value comes from the shortest length distribution of the effective edge segments in 30 images, and the maximum frequency appears at 5 pixels, so it is set as the reference standard; then, for each retained sequence, count its grayscale change interval. The grayscale change amount is defined as the absolute value of the grayscale difference between the starting and ending pixels. For example, the starting point grayscale of sequence D is 95, and the ending point grayscale is 145, with a change of 50, which is a high-change segment and is included in the grayscale change interval list. Each item in this list records the sequence number, starting and ending point coordinates, sequence length, and grayscale change amount. For example, a certain sequence ranges from point (140, 160) to (145, 164), the sequence length is 6.4 pixels, and the grayscale change amount is 52, which meets the characteristic change standard and enters the next step of analysis.

[0022] The angle calculation sub-module calculates according to the adjacent pixel coordinate differences and grayscale differences of each segment of the sequence in the grayscale change interval list, using the formula: ; Calculate the gradient angle of each pair of pixels, accumulate the mean of all gradient angles, and generate an edge pixel distribution map; Among them, θ represents the gradient angle of adjacent pixels, represents the gray value of the i-th pixel, 、 Represents the horizontal and vertical coordinates of the i-th pixel in the image, where i is the pixel index in the sequence; The angle calculation submodule calculates the gradient angle of adjacent pixels for each sequence in the grayscale change interval, and assumes that the grayscale values of any two adjacent pixels are 、 , the corresponding coordinates are 、 , the gradient angle calculation formula is: ; In this formula: Represents the grayscale difference between two adjacent pixels; Denominator Represents the Euclidean distance between two pixels; final Indicates the grayscale change angle per unit distance of pixel gradient change.

[0023] The specific example is as follows: Select a pixel sequence S, which contains three pairs of pixels: The first pair: coordinate points (12, 45) and (18, 48), grayscale values 90 and 145, grayscale difference ,distance ; The angle ; The second pair: coordinate points (18, 48), (22, 50), grayscale values 145, 150, grayscale difference ,distance ; angle ; The third pair: coordinate points (22, 50), (25, 54), grayscale values 150, 160, grayscale difference ,distance ; angle ; Calculate the average gradient angle: ; This value represents the average gray-scale change intensity of the sequence over the entire path and can be used later for constructing the gradient feature map. Among the parameters, the "mutation" criterion is that the gray-scale difference ≥ 45. The first pair of pixels in Sequence 1 meets this condition and participates in the angular aggregation statistics. The set range of the gradient angle threshold is [3°, 10°]. From multiple groups of image data, the gradient angles of pixel pairs in the edge-significant regions are on average concentrated around 6.5° with a standard deviation of 2.5°. The current value of 3.77° falls within the threshold range and belongs to the medium-intensity edge segment. This result indicates that this pixel segment can be used as a medium-strong edge line segment to participate in the edge directional distribution statistics. <> <>

[0024] Please refer to <> Figure 3 , the boundary trend recognition module includes: <> The direction analysis sub-module obtains the edge pixel distribution map, analyzes the connection directions of consecutive pixel points according to the arrangement order of the pixels in the image, records the direction change trend, and generates a pixel arrangement direction sequence; <> After obtaining the edge pixel distribution map, the direction analysis sub-module first sorts the coordinates of the identified edge pixels in the image in the order from top to bottom and from left to right. Each edge point will be assigned a corresponding position number after sorting. For example, the coordinates of the first edge point are (120, 135) and the number is 1. Then it checks in turn whether there is a next edge pixel in its 8-neighborhood. If there is, for example, the next edge point is found at (121, 135), then its direction is determined to be the horizontal direction of 0°. Then it continues to search for the next edge point. If it appears at (122, 136), the direction is 45° in the upper right direction, and so on. The direction information of all connected pixel pairs is recorded to form a direction sequence. The direction values are limited to 8 angles: 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°. Each angle represents a standard connection direction. The record of the direction change trend is based on the continuous angle difference of the direction sequence. For example, if the direction sequence of a certain edge pixel segment is [0°, 0°, 45°, 90°, 90°], then the direction changes are 0, 45, 45, 0. The direction jump points are recorded at the positions where the direction difference is greater than a certain threshold. The system sets the jump judgment threshold to 30°. This value is derived from the maximum direction difference in the average direction continuous segment statistically obtained from the sample image, which is 28°. Therefore, 30° is set as the reference standard for judging jumps, so that the vast majority of normal continuous segments will not be mislabeled as jump segments, and only when the direction change is obvious will it be recorded as a jump point. In actual operation, if in the pixel sequence (100, 150) → (101, 150) → (102, 151) → (103, 152) → (103, 153), the directions are 0°, 45°, 45°, 90° in turn, then the direction jumps are 0 → 45 and 45 → 90, both of which are 45°, so they are recorded as two jump points. Each segment with a continuous direction change less than this threshold is considered to have a consistent direction trend. Finally, this trend is integrated in sequence to construct a complete pixel arrangement direction sequence, providing a basis for the joint judgment of gray difference and direction difference in the next step.

[0025] According to the pixel arrangement direction sequence, the gray difference judgment sub-module calculates the gray difference between adjacent pixel positions, judges the degree of difference between the gray change value and the direction change amount, and filters out the pixel segments with difference values lower than the set continuity threshold to obtain a set of continuous edge segments; The grayscale difference judgment sub-module takes out adjacent pixel pairs one by one according to the pixel arrangement direction sequence constructed in the previous stage, and synchronously extracts their grayscale values and corresponding direction angles. For each pair of pixels, two steps are performed: First, calculate the absolute difference between the grayscale values of the two pixels. For example, if the grayscale value of pixel A is 110 and the grayscale value of pixel B is 115, the grayscale difference is 5. Second, calculate the change in the direction value corresponding to this pair of pixels. For example, the direction from A to B is 0°, and the direction from B to C is 45°, so the direction difference is 45. Then, calculate the difference between the grayscale difference and the direction difference. For example, in the above example, the difference is |5 - 45| = 40. This value is used to determine whether this pixel pair belongs to a continuous edge segment. Therefore, a "continuity threshold" needs to be set. This threshold is obtained by measuring the joint difference value distribution of grayscale difference and direction difference in 100 edge image samples. The statistical results show that most of the difference values of continuous edge segments are concentrated between 0 and 18, with a distribution mean of 10.7 and a standard deviation of 5.2. Therefore, the continuity threshold is set to 20 to cover more than 85% of the continuous edge sample range. In processing actual images, if the grayscale difference of a pixel pair is 7, the direction difference is 5, and the difference value is 2, which meets the threshold condition, it is retained; conversely, if the grayscale difference is 10, the direction difference is 60, and the difference value is 50, then this pixel pair is discarded and will not be part of the continuous edge segment. In this way, all adjacent pixel pairs in the direction sequence are traversed to construct a set of continuous edge segments composed of pixel pairs that meet the threshold conditions. For example, for the edge point sequence (200, 150) → (201, 151) → (202, 152), with grayscale values of 95, 100, 102 respectively and direction values of 45°, 45°, the calculated differences are 5 and 2 respectively, which meet the threshold requirements, so this segment is included in the set of continuous edge segments.

[0026] The trend extraction sub-module calls the set of continuous edge segments, extracts the start and end coordinates of each edge pixel, splices them in order to generate an edge path, arranges the path structure information, and generates a boundary trend map; The trend extraction sub-module calls the set of continuous edge segments, extracts the coordinate information of the first and last two endpoints of each continuous pixel sequence, that is, the start coordinate (x1, y1) and the end coordinate (x2, y2). Then, based on the segment number as the order basis, it sequentially judges whether adjacent two segments meet the splicing conditions. The splicing condition is set that the Euclidean distance between the start and end points does not exceed 3 pixels. This value comes from the fact that the maximum effective connection distance in the 8-neighborhood of the edge image is 2.82 pixels (from the maximum diagonal pixel distance (calculated), rounded to 3 pixels for determining whether adjacent segments can be regarded as connectable regions. For example, the end point of segment 1 is (180, 220), and the start point of segment 2 is (181, 221). Their Euclidean distance is 1.41, which is less than 3 and meets the condition. Perform the splicing operation to merge the two segments into a complete path. Continue to analyze the number of direction change points within the path. The rule for counting the number of direction changes is that when there is a direction jump of more than 30° within the path, it is recorded as one change. If the path direction sequence is [0°, 0°, 45°, 90°, 90°], then there are 2 direction changes, and the path direction stability parameter is 2. The subsequent path information to be recorded includes: path ID, starting coordinates, ending coordinates, path length, number of pixels, and number of direction changes. The path length is obtained by accumulating the distances between pairs of pixels. For example, the distances of 5 pairs of pixel segments are 1.00, 1.41, 1.00, 1.41, 1.00 in sequence, and the total length is 5.82 pixels. The recording results are stored in the form of a data structure. After organizing all path segments, a boundary trend map is generated according to the image coordinates, and finally, a continuous trend information diagram of the particle edge is formed.

[0027] Please refer to Figure 4 , the distorted trajectory repair module includes: Based on the boundary trend map, the end - point deviation extraction sub - module selects the connection positions of adjacent line segments, calls the end - point coordinate information and the corresponding normal and tangential direction data, calculates the normal difference and tangential difference of the connection end - points, analyzes the offset trend, and obtains the end - point direction deviation value; Based on the boundary trend map, the end - point deviation extraction sub - module first reads the sequence number of each boundary line segment in the boundary trend map and the coordinate information of its start and end points to obtain the connection position coordinates between each pair of adjacent line segments. Call the spatial coordinate data of the end - points in the boundary map, such as end - point A (x1 = 12.3, y1 = 45.6) and end - point B (x2 = 15.7, y2 = 48.9), and extract their corresponding tangential direction vectors and normal direction vectors. For example, the tangential vector of line segment AB is t1=(0.6, 0.8), and the normal vector is n1=( - 0.8, 0.6). Then for the next adjacent line segment BC, its tangential vector is set as t2=(0.707, 0.707), and the normal vector is set as n2=( - 0.707, 0.707). Calculate the normal difference and tangential difference between the end - points at the connection of adjacent line segments by vector components. The action is to subtract the components of n1 and n2 to get Δn=(n2x - n1x, n2y - n1y)=( - 0.707 + 0.8, 0.707 - 0.6)=(0.093, 0.107), and the tangential difference is Δt=(t2x - t1x, t2y - t1y)=(0.707 - 0.6, 0.707 - 0.8)=(0.107, - 0.093). Then use the vector modulus to calculate the normal deviation as , and the tangential deviation is , compare this deviation value with the set direction reference scale to judge the offset trend. The direction reference scale is set to 0.12 according to the stability of particle boundaries in actual applications. If the normal deviation or tangential deviation is greater than 0.12, it is recorded as a significant offset position. Through the above example, we get , so mark this connection point as the offset mutation end point. On this basis, count the offset differences of all connection points, and finally form an end point direction deviation value set {0.084, 0.091, 0.141, 0.057} for subsequent connection offset recognition and analysis.

[0028] The connection offset recognition sub-module compares the difference sizes in the continuous deviation sequence according to the end point direction deviation values, screens the line segment set with deviations higher than the direction threshold, forms an offset interval index, and obtains the offset paragraph interval value; The connection offset recognition sub-module receives the end point direction deviation value set {0.084, 0.091, 0.141, 0.057} output in the previous stage, compares the difference sizes of the continuous connection positions one by one, and screens the line segment combinations with deviation values exceeding the direction threshold as potential offset paragraph areas. The direction threshold is set with reference to 0.1, which is calculated according to the fluctuation range of the connection direction in the actual boundary map. The setting method is as follows: calculate the deviation value distribution at 100 line segment connection points, take the average value of the first 80% of the data and multiply it by 1.2 times for relaxation, which is the direction threshold. For example, if the average value of the first 80% of the deviation values in the sample data distribution is 0.083, then the direction threshold is , in actual execution, traverse the set {0.084, 0.091, 0.141, 0.057}, judge the relationship between each value and the threshold 0.1. Among them, 0.141 is greater than 0.1, and the other three values are less than 0.1. Therefore, only keep the paragraph with a connection deviation of 0.141, index its line segment number and record it as the connection position between line segment 3 and 4, mark it as the start and end points of the offset interval, and check the connection differences of the previous and next 1 segment each by step length and judge its continuity. If the previous segment is 0.091 close to the threshold, include it in the interval together. Finally, construct the offset paragraph interval value as: [line segment 2 to line segment 4], and form the corresponding index set [2, 3, 4] according to the line segment order in the original boundary map for the repair operation of the next stage closed contour.

[0029] The closed contour generation sub-module calls the offset paragraph interval value, repairs the connection relationship of the associated boundary line segments. After adjusting the boundary structure, recalculate the angle difference, end point offset amount and segment length change amount of the repaired boundary, using the formula: ; Calculate to obtain the connection reconstruction error degree, and generate a boundary contour map in combination with the minimum error structure reconnection order; Among them, 、 respectively represent the horizontal and vertical differences of the corresponding end points before and after repair in the boundary projection direction. is the angle difference of the edge segment direction before and after repair. is the direction angle reference value. is the edge segment length difference, and δ is the projection scale reference constant. The closed contour generation sub-module calls the obtained offset paragraph interval values [2, 3, 4], extracts the front and rear end point coordinates and line segment angles of each connection segment, performs the connection repair action, calls the end point coordinates and boundary direction information before and after the line segment connection, performs the reconnect operation on each connection segment in turn, and calculates the corresponding horizontal and vertical offsets Δu and Δv, the edge segment direction angle difference Δφ, and the edge segment length difference Δλ before and after the repair connection. Taking the third segment as an example, the front connection end point A is (x1 = 23.1, y1 = 67.4), the rear connection end point is (x2 = 25.5, y2 = 69.5), the horizontal offset , the vertical offset , the direction angles before and after connection are θ1 = 82.4° and θ2 = 87.3° respectively, then the angle difference is , the original edge segment length is 12.3 pixels, and after repair it is 10.0 pixels, so the length difference , the direction angle reference value , the projection reference constant , substitute into the formula: ; The result shows that the reconstruction error degree of the connection segment is -0.1266. A negative value indicates that there is no sudden change in the direction and length changes, allowing this segment to be retained as a connection path. Continue to calculate the Λ value for all repair segments in turn. For example, in another connection segment , , , , the execution formula is as follows: ; Finally, sort the Λ value results of all paragraphs and take the minimum value to form the reconstructed connection path, integrate the boundary point sequences after all repair connections, perform the angle recalculation operation, calculate the direction difference, length difference and offset of all end points again, and judge whether there is an angular mutation area greater than . If it exists, record it as a structural mutation node. If all nodes are less than this threshold, complete the operation of outputting the reconstructed closed boundary. The benefit of the formula is that it participates in the operation through the combination of horizontal and vertical position offsets and angle changes, and introduces an inhibitory term for edge segment length changes to achieve error constraint, so that the geometric continuity and structural smoothness of the boundary can be considered when judging the connection order, ensuring the closure rationality and consistency of the final contour.

[0030] Please refer to Figure 5 , the internal texture analysis module includes: The grayscale acquisition sub-module extracts the internal area of the particles based on the boundary contour map, divides it into equal-area sub-blocks, obtains the pixel grayscale values in the sub-blocks, calculates the peak value, the underestimated value, and the average value, and obtains the regional grayscale distribution range value. The grayscale acquisition sub-module, based on the boundary contour map, first extracts all the pixel point data enclosed within the contour line. The extraction method is to traverse all the pixel coordinates in the image and determine whether they fall within the contour polygon. If the condition is met, the pixel grayscale value is recorded into the internal pixel set. Assume the image is 256×256 pixels, and the contour area covers approximately 64×64 pixels, that is, a total of 4096 pixels. The system divides this area into 4×4 equal-area sub-blocks, each with an area of 16×16 pixels, i.e., 256 pixels. The sub-blocks are numbered from sub-block 1 to sub-block 16 in the order from top to bottom and from left to right. Then, the grayscale value sets of each sub-block are read one by one. The maximum value in each set is taken as the peak value, the minimum value as the underestimated value, and the average of all values as the average grayscale value. For example, the grayscale value set extracted from sub-block 1 is {112, 115, 120,..., 245}, a total of 256 values. After sorting and calculating the average, the peak value of this sub-block is 245, the underestimated value is 112, and the average value is 178. Repeat the above operations for the remaining sub-blocks to obtain the peak, underestimated, and average values of each sub-block. On this basis, calculate the grayscale difference for each sub-block, that is, the difference between the peak value and the underestimated value. For example, the grayscale difference of sub-block 1 is 133, sub-block 2 is 111, sub-block 3 is 90, sub-block 4 is 136. Then, establish horizontal and vertical adjacency relationships between these differences and the average value for subsequent construction of the regional grayscale distribution structure. Finally, calculate that the maximum peak value in all sub-blocks is 251, the minimum underestimated value is 112, and the regional grayscale distribution range value is 139, which is used as a reference benchmark for judging grayscale fluctuations in the next module.

[0031] The grayscale fluctuation calculation sub-module calculates the grayscale difference and the average difference of the sub-blocks according to the regional grayscale distribution range value, judges the grayscale change trend between adjacent sub-blocks, and obtains the regional grayscale change rate. The grayscale fluctuation calculation sub-module receives the regional grayscale distribution range value 139 and the sub-block grayscale difference and average data provided by the previous module. First, it performs a normalization operation on each sub-block grayscale difference. The normalization method is to divide the sub-block grayscale difference by the regional grayscale distribution range value. Assuming the difference of sub-block 1 is 133, the normalized result is 0.957; the difference of sub-block 2 is 111, and the normalized result is 0.799; for sub-block 3, the difference is 90, and the result is 0.647; for sub-block 4, the difference is 136, and the result is 0.978. After normalization, the grayscale fluctuation index sequence is {0.957, 0.799, 0.647, 0.978}. Then, it analyzes the grayscale change trend between adjacent sub-blocks. The method is to calculate the difference in the average grayscale values of two adjacent sub-blocks. For example, the average value of sub-block 1 is 178, and that of sub-block 2 is 182, so the difference is 4; the difference between sub-block 2 and 3 is 6; the difference between sub-block 3 and 4 is 12. Further, it determines whether these differences represent a sudden change in the grayscale change rate. According to the regional grayscale distribution range value, a threshold standard is set. The grayscale rate mutation threshold is set to 5% of the regional grayscale range value, that is, 139×0.05 = 6.95, approximately rounded to 7. If the difference in the average values between adjacent sub-blocks is greater than this value, it is marked as a mutation point. In this example, the difference between sub-block 3 and sub-block 4 is 12, higher than 7, so it is marked as a rate mutation point; the difference between sub-block 1 and 2 is 4, and the difference between sub-block 2 and 3 is 6, both lower than 7, so they do not form mutation points. Finally, the grayscale change trend identification result is that only the grayscale change rate mutation occurs between sub-block 3 and 4, which is further used to locate the texture boundary feature area.

[0032] The texture map generation sub-module calls the regional grayscale change rate, locates the position of the sudden change in the grayscale change rate, combines the grayscale trend of the sub-blocks, draws the overall grayscale hierarchy, and generates a texture distribution map; The texture map generation sub-module receives the grayscale change rate analysis result, and combines the grayscale change index and the mutation identification position of each sub-block to draw the grayscale hierarchy structure diagram of the entire image. The execution method is to first find the mutant sub-blocks in the image coordinates, namely sub-block 3 and sub-block 4, and locate their actual area positions in the image as the 3rd column and the 4th column of the 1st row. Insert texture change boundary lines between these sub-blocks. Subsequently, the texture intensity of each sub-block is classified according to the normalized grayscale difference value of each sub-block. The texture intensity levels are defined as follows: the low-intensity area is between 0 and 0.4, the medium-intensity area is between 0.4 and 0.75, and the high-intensity area is between 0.75 and 1.0. According to this classification method, the differences of sub-blocks 1 and 4 are 0.957 and 0.978, respectively, and both are determined to be high-intensity areas. Sub-block 2 is 0.799, belonging to the high-intensity edge area, and sub-block 3 is 0.647, belonging to the medium-intensity area. Different grayscales are used to fill different levels in the grayscale hierarchy diagram. The low-intensity area is filled with light gray, the medium-intensity area is filled with medium gray, and the high-intensity area is filled with dark gray. Finally, the texture map shows a boundary separation line between sub-blocks 3 and 4, and shows dark areas at the positions of sub-blocks 1 and 4 to indicate the areas with drastic texture changes. At the same time, the medium-intensity and high-intensity transition zones are retained to construct a complete texture distribution image and output it as the input layer data for subsequent image analysis.

[0033] Please refer to Figure 6 , the forming abnormality detection module includes: Based on the texture distribution map, the grayscale trend recognition sub-module reads the grayscale change values of the particle regions block by block, compares the growth amplitude of the grayscale difference between adjacent regions, extracts the grayscale mutation direction and the continuous position, and obtains the continuous grayscale mutation trend value; The grayscale trend recognition sub-module reads the grayscale change values of each sub-block in the particle region block by block based on the texture distribution map generated in the previous stage. This value is determined by the standardized grayscale difference output in the previous stage. For example, the grayscale change values of the sub-block sequence are {0.52, 0.57, 0.69, 0.78, 0.82, 0.95}. The module reads and calculates the difference growth amplitude between adjacent sub-blocks in sequence according to the sub-block numbers. The calculation method is the ratio of the difference between the grayscale values of two adjacent sub-blocks to the grayscale value of the previous sub-block. For example, the amplitude between sub-block 1 and sub-block 2 is (0.57 - 0.52) / 0.52 ≈ 0.096, between sub-block 2 and sub-block 3 is (0.69 - 0.57) / 0.57 ≈ 0.210, between sub-block 3 and sub-block 4 is (0.78 - 0.69) / 0.69 ≈ 0.130, between sub-block 4 and sub-block 5 is (0.82 - 0.78) / 0.78 ≈ 0.051, between sub-block 5 and sub-block 6 is (0.95 - 0.82) / 0.82 ≈ 0.158. After summarizing the above groups of data, they are compared with the growth amplitude threshold in turn. This threshold is set to 0.15, and the setting basis is the upper bound of the statistical results of the maximum natural growth ratio in the stable regions of multiple sample images. The actual value is 0.15. If the growth amplitude between a certain adjacent group is greater than this threshold, it is judged as a mutation growth point. In this example, the amplitude between sub-block 2 and sub-block 3 is 0.210, and between sub-block 5 and 6 is 0.158, both greater than 0.15. Therefore, the mutation directions are recorded as from sub-block 2 to 3 and from sub-block 5 to 6. Then, all mutation points are combined. If the mutation directions are continuous, their continuous positions are further extracted. For example, the change trends of sub-blocks 2, 3, and 4 are growth, growth, and growth respectively, forming a continuous trend segment, and the start and end sub-block numbers 2 to 4 are recorded. This process judges all continuous mutation trend segments in turn, and finally forms a set of continuous grayscale mutation trend values {(2, 4), (5, 6)}.

[0034] The mutation region marking sub-module calls the continuous grayscale mutation trend values, detects the continuous particle numbers and coordinate positions in the mutation section, records the range of the grayscale change abnormal region, constructs a particle local marking layer, and obtains the particle mutation marking interval value; The mutation region marking sub-module receives a set of continuous gray mutation trend values, extracts the start and end numbers of all mutation segments among them, and extracts the actually corresponding image block regions according to the pixel coordinates indexed by the sub-block mapping in the texture map. For example, the coordinate range of the image region corresponding to sub-block numbers 2 to 4 is 32 to 96 pixels in the x direction and 0 to 32 pixels in the y direction. The module delimits a rectangular region box in the image matrix according to this coordinate interval, and records the number of the starting sub-block, the upper left corner coordinates (x1, y1), and the lower right corner coordinates (x2, y2). The same operation is continued for each trend segment in the set. Suppose the region corresponding to the trend segment (5, 6) is 96 to 128 pixels in the x direction and 0 to 32 pixels in the y direction, then its coordinates are recorded as (x1 = 96, y1 = 0), (x2 = 128, y2 = 32). The coordinate sets of these mutation segments are combined to form a gray mutation marking region group. Then, a transparent layer is drawn in the original image to cover the above coordinate regions. The layer type is a rectangular mask layer. Each layer corresponds to a mutation section number, and the number of the particle sub-blocks included in the layer is numbered and recorded. Finally, the particle local marking layer information is obtained. The number of sub-blocks marked in this layer is used for the aggregation analysis in the subsequent structure judgment module. For example, if the sub-block sequence (2, 3, 4) is a continuous region, the particle numbers are the 2nd to the 4th blocks under particle ID01, and the mutation particle numbers are recorded as [01-2, 01-3, 01-4]. The finally output particle mutation marking interval values include the mutation segment number, the sub-block number range, the layer coordinate information, and the particle ID mapping index, which are used as the input basis for the subsequent structure screening.

[0035] The structure screening and judgment sub-module screens the particle structure images corresponding to the gray mutation regions according to the particle mutation marking interval values, judges whether there is a mutation convergence phenomenon in the whole structure, outputs the structure mutation intensity level, and generates the wind shear forming detection result; The structure screening and judgment sub-module receives the particle mutation marker interval value, screens the particle structure images corresponding to each mutation region, and the screening action is to traverse all regions in the marker layer and extract the corresponding image content one by one. After extraction, the structural aggregation of each structural block is judged. The judgment criterion is whether there are multiple sub-blocks concentrated within the contour of the same particle structure in the detected marker region, and calculate the percentage of the total area of the contour occupied by these sub-blocks. For example, if the particle structure number is 01, the total area of its boundary contour is 2048 pixels, the sub-block numbers in the marker interval are 2, 3, and 4, and the area of each sub-block is 256 pixels, with a total of 768 pixels, then the calculated proportion is 768 / 2048 = 0.375. This proportion is compared with the structure mutation judgment threshold. The structure mutation judgment threshold is set to 0.30, and the setting basis is the maximum non-mutation area ratio of the normal texture region in the structure image. After analyzing multiple sample particles, the upper bound of 0.30 is taken as the threshold reference. If the actual proportion is greater than this value, it is judged as a mutation convergence region. In this example, 0.375 is greater than 0.30, so it is determined as a mutation convergence. The system further outputs the mutation intensity level according to the convergence degree. The mutation intensity level is divided into low (proportion 0 - 0.15), medium (0.15 - 0.30), and high (0.30 - 1.0). The above proportion of 0.375 corresponds to the high level. Finally, the wind shear forming detection result is generated, recording the particle number 01 and the mutation intensity level as high. The output result is used for the subsequent structural sorting process identification stage.

[0036] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A calcium carbonate masterbatch air-shearing forming detection system based on machine vision, characterized in that, The system includes: The edge pixel collection module extracts the particle edge image area, detects continuous edge pixel points, calculates the sequence length, gray scale range, gray scale difference, and gradient angle, comprehensively reflects the distribution characteristics of edge pixels, and generates an edge pixel distribution map; The boundary trend recognition module analyzes the pixel arrangement direction and gray scale change based on the edge pixel distribution map, judges the correlation between direction offset and gray scale change, extracts edge line segments that meet continuity and direction consistency and their start and end coordinates, and generates a boundary orientation map; The distortion trajectory repair module analyzes the connection offset of adjacent line segments in the boundary orientation map, calculates the difference between the normal and tangential positions of the line segment endpoints, and corrects the offset line segments according to the overall boundary trend to generate a boundary contour map; The internal texture analysis module extracts the internal gray scale information of the particle based on the boundary contour map, calculates the gray scale distribution range and change rate of the region in blocks, analyzes the gray scale fluctuation trend, draws the internal texture structure of the particle, and generates a texture distribution map; The forming abnormality detection module detects the gray scale change rate of the region based on the texture distribution map, identifies the gray scale mutation in the continuous region, and generates a wind cut forming detection result.

2. The calcium carbonate masterbatch air-shearing forming detection system based on machine vision according to claim 1, wherein The edge pixel distribution map includes pixel sequence length characteristics, gray scale value distribution range, adjacent pixel gray scale difference, and pixel gradient angle characteristics. The boundary orientation map includes edge line segment direction distribution, line segment start and end coordinates, line segment continuity characteristics, and direction offset. The boundary contour map includes line segment connection relationship, overall boundary trend, normal position deviation, and tangential position deviation. The texture distribution map includes gray scale distribution structure, regional gray scale change rate, gray scale fluctuation trend, and texture structure distribution. The wind cut forming detection result includes gray scale mutation region, continuous abnormal region, abnormal region marking information, and overall particle abnormal structure.

3. The calcium carbonate masterbatch air knife forming detection system based on machine vision according to claim 2, wherein, The edge pixel collection module includes: The edge extraction sub-module obtains the particle edge image area, detects the positions of continuous edge pixel points, screens out the pixel points with gray scale mutation as edge points, records the image coordinates and gray scale values, and generates an edge pixel point set; The sequence construction sub-module constructs pixel sequences according to the adjacent relationship of positions in the edge pixel point set, screens out the sequences with a length greater than the connection threshold, and counts the gray scale values of each segment of the sequence to generate a list of gray scale change intervals; The angle calculation sub-module calculates the gradient angle of each pair of pixels according to the adjacent pixel coordinate difference and gray scale difference of each segment of the sequence in the list of gray scale change intervals, accumulates the mean value of all gradient angles, and generates an edge pixel distribution map; The quantization definition of the connection threshold: The connection threshold is the minimum number of continuous edge pixel points in the sequence, and is set to 3-5 pixel lengths.

4. The calcium carbonate masterbatch air-shearing forming detection system based on machine vision according to claim 3, wherein, The boundary trend recognition module includes: The direction analysis sub-module obtains the edge pixel distribution map, analyzes the connection direction of continuous pixel points according to the arrangement order of pixels in the image, records the direction change trend, and generates a pixel arrangement direction sequence; The gray scale difference judgment sub-module calculates the gray scale difference between adjacent pixel positions according to the pixel arrangement direction sequence, judges the difference degree between the gray scale change value and the direction change amount, and screens out the pixel segments with a difference value lower than the set continuity threshold to obtain a set of continuous edge segments; The edge direction extraction sub-module calls the continuous edge segment set, extracts the start and end coordinates of each edge pixel, splices them in order to generate an edge path, organizes the path structure information, and generates a boundary direction map; The degree of difference = |gray value change / direction change amount|.

5. The calcium carbonate masterbatch air-shearing forming detection system based on machine vision according to claim 4, wherein The distortion trajectory repair module includes: The endpoint deviation extraction sub-module, based on the boundary direction map, selects the connection positions of adjacent line segments, calls the endpoint coordinate information and the corresponding normal and tangential direction data, calculates the normal difference and tangential difference of the connection endpoints, analyzes the offset trend, and obtains the endpoint direction deviation value; The connection offset recognition sub-module, according to the endpoint direction deviation value, compares the difference magnitudes in the continuous deviation sequence, filters the set of line segments with deviations higher than the direction threshold, forms an offset interval index, and obtains the offset paragraph interval value; The closed contour generation sub-module calls the offset paragraph interval value, repairs the connection relationship of the associated boundary line segments, adjusts the boundary structure, recalculates the angle difference, endpoint offset amount, and segment length change amount of the repaired boundary, calculates the connection reconstruction error degree through operations, and combines the minimum error structure reconnection order to generate a boundary contour map; The direction threshold is determined according to the statistical distribution of the normal and tangential angle changes during the natural connection of edge line segments, and the value range is 5° to 15°.

6. The calcium carbonate masterbatch air-shearing forming detection system based on machine vision according to claim 5, characterized in that, The internal texture analysis module includes: The gray value acquisition sub-module, based on the boundary contour map, extracts the internal area of the particle, divides it into equal-area sub-blocks, obtains the pixel gray values in the sub-blocks, statistically calculates the peak value, low value, and average value, and obtains the regional gray value distribution range; The gray value fluctuation calculation sub-module calculates the gray value difference and average difference of the sub-blocks according to the regional gray value distribution range, judges the gray value change trend between adjacent sub-blocks, and obtains the regional gray value change rate; The texture map generation sub-module calls the regional gray value change rate, locates the positions where the gray value change rate mutates, combines the gray value trend of the sub-blocks, draws the overall gray level structure, and generates a texture distribution map; The specific division of the equal-area sub-blocks uses an 8×8 pixel grid division.

7. The calcium carbonate masterbatch air-shearing forming detection system based on machine vision according to claim 6, characterized in that, The forming anomaly detection module includes: The gray value trend recognition sub-module, based on the texture distribution map, reads the gray value changes of the particle area block by block, compares the growth amplitude of the gray value differences between adjacent areas, extracts the gray value mutation direction and continuous position, and obtains the continuous gray value mutation trend value; The mutation area marking sub-module calls the continuous gray value mutation trend value, detects the continuous particle numbers and coordinate positions within the mutation section, records the range of the gray value change abnormal area, constructs a particle local marking layer, and obtains the particle mutation marking interval value; The structure screening and judgment sub-module, according to the particle mutation marking interval value, screens the particle structure images corresponding to the gray value mutation areas, judges whether there is a mutation convergence phenomenon in the whole structure, outputs the structure mutation intensity level, and generates the wind shear forming detection result.

8. The calcium carbonate masterbatch air-shearing forming detection system based on machine vision according to claim 7, characterized in that, The determination criterion for the mutation convergence phenomenon is that gray value mutations occur simultaneously in more than 3 adjacent sub-blocks.

Images