Cyst particle analysis system based on AI algorithm

Abstract

The invention relates to the technical field of particle analysis, in particular to a cyst particle analysis system based on an AI algorithm, and the system comprises a particle image screening module, a boundary feature module, a structure deduction module, a particle recognition module and a performance analysis module. According to the method, the gray contrast and edge change rate analysis is carried out on an encapsulated lubricant image frame sequence in a liquid or semi-solid state to realize accurate positioning of a particle target area, an edge angle continuous change slope sequence is utilized to determine a particle structure boundary sudden change position, and the particle structure edge is deduced according to a boundary trend continuous behavior. The method comprises the following steps: accurately establishing a closed structure path, identifying an adhesion region, carrying out decoupling division on particles by using included angle range and direction consistency indexes, determining the boundary contour of the particles, rapidly and accurately marking a structure dissimilatory aggregation region through the area pixel number and gray gradient characteristics of a structure unit, and improving the particle structure identification precision. And the misjudgment risk caused by particle overlapping and adhesion is reduced.

Description

Technical Field

[0001] The present invention relates to the field of particle analysis technology, and in particular to a cyst particle analysis system based on an AI algorithm. Background Art

[0002] The field of particle analysis technology primarily involves the quantitative or qualitative detection and evaluation of parameters such as the distribution state, morphological characteristics, size distribution, composition, structural characteristics, and quantity of solid particles in liquid, gas, or solid media. This field is widely used in the chemical, pharmaceutical, materials, food, and microelectronics industries. It primarily utilizes optical microscopy, laser scattering, image recognition, X-ray energy spectrum analysis, and mass spectrometry to establish particle size distribution curves, particle morphological parameter models, and particle composition analysis reports. This allows for a refined, multi-dimensional analysis of target particle samples, providing data support for material performance control, product quality assessment, and process optimization.

[0003] The cyst particle analysis system is designed to systematically detect and analyze particles contained in encapsulated lubricants. The system identifies the presence, quantity, size, distribution, and structural characteristics of cyst particles. It is particularly suited for evaluating whether particles in encapsulated lubricants have potential functional effects or adversely affect lubrication performance. It analyzes the structural integrity, particle size distribution, shell thickness, and distribution stability of cyst particles within the lubricant, thereby determining the lubricating performance and service stability of the lubricant.

[0004] Traditional analysis systems rely on the detection of the overall appearance and rough distribution of particles. They lack clarity in identifying the structural mutation locations of particle boundaries, edge trends, and adhesion areas between particles, resulting in large boundary positioning errors. Traditional systems lack clear quantitative analysis methods for particle structural details such as tangent angle changes, boundary continuity, and mutation. This leads to ambiguous judgments of particle structure and difficulty in effectively identifying the actual boundaries of adhering particles, significantly increasing the misjudgment rate of particle overlap and adhesion. For example, the phenomenon of particle overlap cannot be accurately identified and separated, affecting the accuracy of particle count and size statistics, resulting in inaccurate evaluation of lubricant functionality and service performance, reducing the efficiency of lubrication performance monitoring and management, and posing potential risks to lubricant application stability assessments. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a cyst particle analysis system based on AI algorithm.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a cyst particle analysis system based on an AI algorithm, the system comprising: The particle image screening module obtains a sequence of image frames of the encapsulated lubricant in a liquid or semi-solid state, locates the central grayscale peak coordinates of the particle target area, and collects the pixel index information of the corresponding coordinates in the image matrix to generate a particle image grayscale index table; The boundary feature module calls the particle image grayscale index table, extracts the boundary pixel sequence within the range, determines the boundary trend change law based on the boundary angle difference range, and marks the boundary behavior mutation points to generate boundary direction change annotation information; The structure deduction module obtains the corresponding tangential grayscale value fluctuation direction sequence based on the boundary direction change annotation information, inputs the sample set into the boundary prediction model, determines the difference between the stable behavior sequence in the trajectory node distribution and the mean absolute error distribution of the boundary fitting curve based on the learning weights in the model, extracts the low-error fitting path to construct a closed connected structure, and obtains the adhesion structure reconstruction judgment information; The particle recognition module calls the adhesion structure reconstruction judgment information, calculates the angle range difference value and direction vector consistency coefficient between the node and the image grayscale centroid path, screens the structure boundary basis point, and generates a particle boundary decoupling node group.

[0007] As a further solution of the present invention, the particle image grayscale index table includes the particle grayscale peak point, image pixel position index, boundary grayscale transition value and target area grayscale density value; the boundary direction change annotation information is specifically a boundary mutation marking layer, an angle mutation feature point set, a slope change distribution trajectory and a boundary angle difference distribution map; the adhesion structure reconstruction judgment information includes a closed path pixel sequence diagram, a trajectory stability difference matrix, a node connectivity judgment relationship set and a behavior trajectory error distribution map; the particle boundary decoupling node group specifically refers to a structural boundary node pair set, a grayscale center of gravity direction comparison set, an angle deviation analysis table and a boundary basis position index set.

[0008] As a further solution of the present invention, the particle image screening module includes: The image frame acquisition submodule acquires a sequence of image frames of the encapsulated lubricant in a liquid or semi-solid state, detects the image of each frame, determines whether the grayscale distribution range of the image falls within the image noise filtering threshold, selects a set of image frames that meet the distribution range requirements, and generates information on the number of image frames screened; The grayscale feature positioning submodule calls the image frame to filter the quantitative value information, analyzes the grayscale difference value of the pixel block area and the edge grayscale change rate in the image, determines that the area with the grayscale difference value above the grayscale contrast reference value is the particle target area, searches for the pixel with the maximum grayscale value in the particle target area and locates the grayscale peak coordinates, and generates a particle grayscale peak coordinate set; The pixel index extraction submodule retrieves the corresponding position index in the image according to the particle grayscale peak coordinate set, marks the image pixel row and column numbers and grayscale peak value range of the coordinate point, and integrates the record entries to generate a particle image grayscale index table.

[0009] As a further solution of the present invention, the boundary feature module includes: The boundary pixel extraction submodule calls the particle image grayscale index table, performs grayscale difference judgment on the neighborhood of pixels around each grayscale peak coordinate based on the recorded grayscale peak coordinates and edge grayscale change rate, selects boundary pixels whose grayscale gradient direction is consistent with the grayscale peak diffusion direction, numbers them, and generates a boundary pixel extraction quantity value; The tangent slope calculation submodule extracts the number of boundary pixels and sequentially obtains the pixel positions of three points in a group. The slope of the line connecting the first and third points in each group is calculated, and the slope is used as the tangent direction angle of the middle point. The difference between adjacent angles is calculated using the tangent direction angle to obtain a continuous angle change trend sequence for each point, thereby generating a boundary angle change slope value. Based on the boundary angle change slope value, the mutation annotation recognition submodule compares and analyzes the angle change slope of each boundary point with the angle fluctuation difference of the three consecutive points before and after, determines that the pixel point with an angle difference greater than the set boundary angle mutation threshold is the mutation point, and annotates the position of the mutation point to generate boundary direction change annotation information.

[0010] As a further solution of the present invention, the structure deduction module includes: The behavior trajectory construction submodule extracts the tangential grayscale change direction sequence of each mutation point based on the boundary direction change annotation information, obtains the normal angle difference between adjacent points, cross-screens the direction mutation rate and normal angle mutation amplitude in the tangential sequence, extracts the point sequence that meets the mutation synchronization judgment rule, constructs a candidate trajectory point set, and generates the boundary behavior trajectory sample value; The path prediction fitting submodule calls the boundary behavior trajectory sample value, inputs the candidate trajectory point sequence into the boundary prediction model, applies the learned weights obtained by training to the sample point sequence in the model to determine the response intensity distribution, and calculates the mean absolute error between the fitting curve of the sample path and the true boundary. For path samples whose error values are lower than the error distribution stability threshold, a node sequence is extracted to obtain a stable path prediction fitting value. The connectivity relationship judgment submodule is based on the stable path prediction fitting value and the node sequence coordinate information, and calculates the tangent direction and extension angle between the nodes respectively, compares the node direction vector with the adjacent normal angle direction, and judges that the node connection pairs whose angle range is within the threshold of the clamping angle interval are the connection structure basis points, generates a connectivity relationship group, and establishes the adhesion structure reconstruction judgment information.

[0011] As a further solution of the present invention, the particle identification module includes: The connected node extraction submodule obtains boundary nodes distributed in the intersection area of the non-closed path based on the adhesion structure reconstruction judgment information, selects points in the adjacent node sequence whose angle fluctuation difference is greater than the angle change recognition threshold as mutation nodes, and obtains the number of boundary mutation nodes; The directional consistency calculation submodule calls the number of boundary mutation nodes, obtains the direction vector of each node and the grayscale centroid, calculates the angle range of the node connection path, obtains the cosine value of the angle between the node path direction and the centroid direction, calculates the directional consistency score of the node, marks the nodes with directional consistency scores lower than the directional mutation threshold as candidate structural nodes, and generates directional consistency screening results; The structural boundary judgment submodule judges whether the angle change direction between adjacent nodes is continuous based on the direction consistency screening result, extracts the node pairs that meet the continuity direction switching threshold condition in the angle change sequence as the structural boundary reference points, and obtains the particle boundary decoupling node group.

[0012] As a further solution of the present invention, the formula for calculating the direction consistency score of the node is specifically: ; in, represents the direction consistency score of the i-th node, Represents the direction angle of the path from the i-th node to the grayscale centroid, represents the average value of the node direction angle, Represents the angle between the path direction of the i-th node and the direction of the grayscale center of gravity, Represents the normalized value of the pixel distance from the i-th node to the grayscale centroid path.

[0013] As a further embodiment of the present invention, the system further comprises: The performance analysis module calculates the particle density alienation index based on the particle boundary decoupling node group, according to the number of boundary area pixels and the maximum grayscale gradient point of each structural particle unit, and compares it with the standard distribution density of the image background area. It also marks the high-offset area, statistically distributes the position and number of the structural alienation aggregation units in the image, and obtains the particle behavior interference distribution information; The particle behavior interference distribution information includes interference frame classification labels, gradient distribution density maps, risk level mapping areas, and particle behavior change sections.

[0014] As a further solution of the present invention, the performance analysis module includes: The particle structure calculation submodule is based on the particle boundary decoupling node group and the structural boundary point list of each particle. It obtains the number of pixels within the closed boundary of the region as the number of boundary area pixels, and performs first-order gradient calculation on the grayscale change value in each pixel area. The point where the grayscale gradient has the maximum value is recorded, and the feature record item set of the structural particle is constructed to generate the particle structure feature quantity value. The density alienation calculation submodule calls the particle structure feature quantity value, performs normalization offset judgment based on the maximum grayscale gradient point and the number of boundary area pixels of the particle, combined with its grayscale center of gravity density and the image background standard distribution density value, calculates the density alienation score of each particle unit, and compares it with the alienation recognition threshold. Particles with density alienation scores greater than the alienation recognition threshold are selected as alienated units to obtain the density alienation evaluation result; Based on the density alienation assessment results, the interference aggregation labeling submodule extracts the corresponding image position coordinates and sets number labels according to the screened alienation unit sequence, constructs a coordinate distribution layer, extracts the cluster contour boundaries of the numbered positions, and integrates the center point position of the cluster area with the label to obtain particle behavior interference distribution information.

[0015] As a further solution of the present invention, the formula for calculating the density alienation score of each particle unit is specifically: ; in, represents the density alienation score of the kth particle, represents the grayscale centroid density of the kth particle, Indicates the standard density of the background area of the image, Represents the distribution value of the kth particle in the jth dimension of the structure extension direction vector, Represents the mean value of the direction vector of the image background area in the jth dimension, Represents the normalized value of the number of pixels of the boundary area of the kth particle.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by performing grayscale contrast and edge change rate analysis on a sequence of image frames of an encapsulated lubricant in a liquid or semi-solid state, precise positioning of the target particle area is achieved, the position of the particle structure boundary mutation is determined using a sequence of continuously changing slopes of edge angles, a characteristic image of boundary direction change is constructed, the particle structure edge is deduced based on the continuous behavior of the boundary trend, a closed structure path is accurately established and adhesion areas are identified, the particles are decoupled and divided using angle range and direction consistency indicators, the particle boundary contour is clarified, and a particle distribution vector group is established and a particle density alienation index is calculated through the area pixel number and grayscale gradient characteristics of the structural unit, so that the structural alienation aggregation area is quickly and accurately marked, the particle structure recognition accuracy is improved, the risk of misjudgment caused by particle overlap and adhesion is reduced, the abnormal particle distribution area is more effectively monitored, and the accuracy of particle functionality evaluation and the prediction ability of lubricant service stability are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 is a system flow chart of the present invention; Figure 2 Schematic diagram of the system framework of the present invention; Figure 3 This is a flow chart of the particle image screening module of the present invention; Figure 4 This is a flow chart of the boundary feature module of the present invention; Figure 5 This is a flow chart of the structure deduction module of the present invention; Figure 6 This is a flow chart of the particle recognition module of the present invention; Figure 7 Flowchart of the performance analysis module of the present invention. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0020] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0021] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0022] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0023] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0024] See also Figure 1 , a cyst particle analysis system based on AI algorithm, the system includes: The particle image screening module obtains a sequence of image frames of the encapsulated lubricant in a liquid or semi-solid state, detects the grayscale contrast distribution value and edge grayscale change rate in each frame, locates the central grayscale peak coordinates of the particle target area, and collects the pixel index information of the corresponding coordinates in the image matrix to generate a particle image grayscale index table; The boundary feature module calls the particle image grayscale index table and extracts the boundary pixel sequence within the range based on the grayscale peak coordinates and the edge grayscale change rate. It calculates the tangent direction angle of each pair of adjacent boundary pixels and the slope sequence of continuous angle changes. It determines the boundary trend change pattern based on the boundary angle difference range, marks the boundary behavior mutation points, and generates boundary direction change annotation information. The structural deduction module, based on the boundary direction change annotation information and the continuous behavior trend of the mutation point, obtains the corresponding tangential grayscale value fluctuation direction sequence, and combines the normal angle mutation behavior to establish a candidate point behavior trajectory sample set. The sample set is input into the boundary prediction model. The difference between the stable behavior sequence in the trajectory node distribution and the mean absolute error distribution of the boundary fitting curve is determined based on the learning weights within the model. The low-error fitting path is extracted to construct a closed connected structure. The relationship between the ductility angle of the path node distribution and the normal mutation direction is determined to establish a structural connectivity judgment matrix and obtain the adhesion structure reconstruction judgment information. Mean absolute error (MAE) is a metric used in image behavior prediction models to measure the error between predicted and true values. Learning weights are the feature map values generated by AI models during training. They are used to assess the reliability of the mapping between the input sequence and the target output. They are commonly found in convolutional neural networks and graph neural network structures. The particle recognition module uses the adhesion structure reconstruction judgment information and calculates the angle range and direction vector consistency coefficient between the node and the image grayscale centroid path based on the mutation nodes distributed at the intersection of non-closed paths in the connectivity judgment matrix. Node pairs with angle range values lower than the set direction consistency threshold and consistency coefficient higher than the connectivity strength benchmark are selected as structural demarcation points to generate particle boundary decoupling node groups. The directional vector consistency coefficient is used to measure the directional proximity between two vectors and is often used to determine the similarity of the edge structure of an image. The grayscale centroid path is a set of paths formed by connecting the centroid of a particle region with any boundary point under the weight of grayscale intensity. It is widely used in image clustering and structure recognition analysis. The performance analysis module is based on the particle boundary decoupling node group. It constructs a particle distribution vector group according to the number of boundary area pixels and the maximum grayscale gradient point of each structural particle unit. It calculates the particle density alienation index based on the particle grayscale center of gravity density and the structure extension direction vector set, and compares it with the standard distribution density of the image background area. It also marks the high-offset area, statistically distributes the position and number of the structural alienation aggregation units in the image, and obtains the particle behavior interference distribution information.

[0025] The particle image grayscale index table includes the particle grayscale peak point, image pixel position index, boundary grayscale transition value and target area grayscale density value. The boundary direction change annotation information is specifically the boundary mutation mark layer, angle mutation feature point set, slope change distribution trajectory and boundary angle difference distribution map. The adhesion structure reconstruction judgment information includes the closed path pixel sequence map, trajectory stability difference matrix, node connectivity judgment relationship set and behavior trajectory error distribution map. The particle boundary decoupling node group specifically refers to the structure boundary node pair set, grayscale center of gravity direction comparison set, angle deviation analysis table and boundary basis position index set. The particle behavior interference distribution information includes the interference frame classification label, gradient distribution density map, risk level mapping area and particle behavior change segment.

[0026] See also Figure 2 and Figure 3 ,The particle image screening module includes an image frame acquisition submodule, a grayscale feature positioning submodule, and a pixel index extraction submodule; The image frame acquisition submodule acquires a sequence of image frames of the encapsulated lubricant in a liquid or semi-solid state, detects the image of each frame, determines whether the grayscale distribution range of the image falls within the image noise filtering threshold, selects a set of image frames that meet the distribution range requirements, and generates information on the number of image frames screened; To obtain an image frame sequence of the encapsulated lubricant in a liquid or semi-solid state, a high-speed image sensor is first used to capture the lubricating medium in a flowing state frame by frame, and the original dynamic image frame is intercepted into a continuous static frame image at a sampling rate of 20fps. The image frame resolution is set to 1024×1024 pixels, and the grayscale value range of each pixel in the image matrix is set to 0–255, and its corresponding grayscale level is defined as an 8-bit unsigned integer value. Then, the grayscale distribution range of each pixel area in the image matrix is detected, and the grayscale histogram of each frame image is obtained and the grayscale variance is calculated to determine whether there is a problem of insufficient difference between the grayscale concentrated area and the dispersed area in the image frame. The image noise filtering threshold is used as the boundary standard. The image noise filtering threshold is 25. If the image grayscale variance is lower than this value, the image is judged to be an invalid frame and is eliminated. The image frames with image grayscale variance higher than the threshold are retained. After the screening logic is executed, all image frames that meet the requirements are integrated into an image frame set, and the difference in the number of frames before and after screening is recorded. The screening process example is: if a total of 120 frames of images are collected, it is judged that 86 frames have grayscale variances above 40, 12 frames are between 25-40, and 22 frames are below 25, then only the first 86 frames enter the subsequent analysis process, and the remaining images are eliminated, and the image frame screening number value is 86; The grayscale feature positioning submodule calls the image frame to filter the quantitative value information, analyzes the grayscale difference value of the pixel block area and the grayscale change rate of the edge in the image, determines that the area with the grayscale difference value above the grayscale comparison reference value is the particle target area, searches for the pixel with the maximum grayscale value in the particle target area, locates the grayscale peak coordinates, and generates the particle grayscale peak coordinate set; Call each frame image matrix in the image frame screening quantity value, obtain the image block matrix divided into 8×8 pixels in each frame image, calculate the average grayscale value of each image block and compare the difference with the maximum grayscale value in the edge area of the image block, construct the grayscale difference value matrix, set the grayscale contrast benchmark value to 20, if the difference value exceeds the benchmark value, the image block is determined to be a candidate particle target area, and then calculate the edge grayscale change rate of each candidate image block, use the central difference formula to derive the pixel grayscale value of the block along the x-axis and y-axis respectively, and obtain the gradient value. Set the grayscale change rate threshold to 15. If there is at least one directional gradient value greater than the threshold in the image block, the image block is retained as the particle valid area. Then, the grayscale value of each pixel in the particle valid area is searched, and the grayscale peak point, that is, the pixel coordinate point corresponding to the maximum grayscale value, is located. Its row and column numbers are recorded in the grayscale peak coordinate list. In the actual example, if there are 300 image blocks in the image frame, 128 blocks are retained after grayscale difference value screening, of which 92 blocks meet the grayscale change rate threshold. Finally, the grayscale peak points in the 92 particle areas are extracted as the target coordinates to generate the particle grayscale peak coordinate set; The pixel index extraction submodule retrieves the corresponding position index in the image based on the particle grayscale peak coordinate set, marks the image pixel row and column numbers and grayscale peak value range of the coordinate point, and integrates the record entries to generate a particle image grayscale index table; According to the coordinate point information recorded in the particle grayscale peak coordinate set, the image matrix position index value corresponding to each coordinate point is extracted, and the image matrix pixel array is matched for each point row and column position respectively. The grayscale value range is locked to the pixel neighborhood grayscale interval of ±10. The maximum and minimum grayscale values in the neighborhood are extracted and their distribution positions are recorded. The grayscale peak value of each particle, the grayscale range of the surrounding pixels, and the position index number constitute the particle index entry. All entries are integrated to form a unified index table. The index table fields include image frame number, particle number, grayscale peak row and column number, grayscale value, peak pixel neighborhood grayscale range, block number, etc. For example, if the particle grayscale peak coordinate set contains 85 sets of coordinate data, the corresponding image blocks are numbered B1 to B85, and after extracting the neighborhood grayscale range for each peak point, the maximum and minimum values are recorded as 238 and 211 respectively. Then the particle record entry is written into the index table. After performing the same operation on all 85 sets of data, the particle image grayscale index table is generated.

[0027] See also Figure 2 and Figure 4 ,The boundary feature module includes boundary pixel extraction submodule, tangent slope calculation submodule, and mutation annotation ,recognition submodule; The boundary pixel extraction submodule calls the particle image grayscale index table and performs grayscale difference judgment on the neighborhood of pixels around each grayscale peak coordinate based on the recorded grayscale peak coordinates and edge grayscale change rate. It selects boundary pixels whose grayscale gradient direction is consistent with the grayscale peak diffusion direction and numbers them to generate the number of boundary pixel extraction values. To call the grayscale peak coordinates and edge grayscale change rate recorded in the particle image grayscale index table, first extract each grayscale peak coordinate point in the particle image frame from the index table. , and then define a neighborhood with a radius of 3 pixels around this point, that is, extract a 7×7 pixel block consisting of ±3 pixels horizontally and ±3 pixels vertically with this coordinate as the center, and compare the grayscale values of all pixels in the neighborhood with the grayscale value of the center point Perform difference calculation to obtain the grayscale difference of each pixel , and combined with its relative position, determine whether the grayscale diffusion direction of the pixel is consistent with the center grayscale gradient direction. The grayscale diffusion direction is determined by the maximum gradient direction of the adjacent 3×3 pixels. The direction consistency is determined as follows: the angle between the direction vector of the line connecting the pixel and the center point and the main gradient direction is less than 30°. This judgment can be determined by comparing whether the direction cosine value is greater than 0.866. If this condition is met, the grayscale diffusion direction is considered to be consistent, and the pixel coordinates are further recorded and assigned a number. In a specific example, if 100 particle grayscale peak points are detected in the image frame, after the 7×7 neighborhood grayscale diffusion direction consistency judgment, 12 boundary pixels are extracted from each point on average, and the final total number is 1200 boundary pixels, generating the number of boundary pixel extraction values; The tangent slope calculation submodule extracts the number of boundary pixels and sequentially obtains the pixel positions of three points in a group. It calculates the slope of the line connecting the coordinates of the first and third points in each group and uses the slope as the tangent direction angle of the middle point. It then uses the tangent direction angle to calculate the difference between adjacent angles, obtaining a continuous angle change trend sequence for each point and generating the boundary angle change slope value. According to the boundary pixel extraction value of all numbered pixel point sequences, three consecutive numbered points are extracted as a group, and the coordinate positions of the first and third points in each group are recorded as and , use the coordinate difference to calculate the slope of the connecting line , use the slope value as the tangent direction angle of the middle pixel point , the unit is angle system, after all points have completed the slope angle calculation, extract the adjacent angle difference of each point in sequence index order , construct a tangent angle change sequence, where the angle difference less than 5° is determined as a boundary smooth point, the angle difference greater than 15° is determined as a curvature mutation candidate point, and the point between 5° and 15° is considered a normal boundary transition point. In the specific example, if there are 1200 numbered boundary pixel points, a total of 1198 groups of angle sequences are generated. The angle change range statistics are: 842 groups of change values are less than 5°, 236 groups are between 5° and 15°, and 120 groups are greater than 15°. Finally, the continuous angle change trend sequence of the tangent direction of all boundary points in the frame and their adjacent points is calculated to generate the boundary angle change slope value; The mutation annotation recognition submodule compares and analyzes the angle change slope of each boundary point with the angle fluctuation difference of three consecutive points before and after it based on the boundary angle change slope value. It identifies the pixel point with an angle difference greater than the set boundary angle mutation threshold as the mutation point, annotates the position of the mutation point, and generates boundary direction change annotation information. Based on the tangent angle difference data recorded in the boundary angle change slope value, first extract the tangent angle change value of each boundary point with the previous and next points. , calculate the mean of the three-point change difference As the angle fluctuation reference value of the point, the current angle difference of the point is then subtracted from the mean to determine whether the difference is greater than the boundary angle mutation threshold. The mutation threshold is set to 10°, which is set with reference to the minimum morphological mutation angle in common particle contours. In the example, if the angle difference of three points at a certain point is 8°, 14°, and 12°, the average is 11.33°. If the current point angle is 22°, the difference is 10.67°, which is greater than the mutation threshold. The point is determined to be a mutation point, and the corresponding pixel position is marked as the mutation point. All mutation point sets are finally summarized into the boundary annotation layer and numbered. In a frame of image, if there are 1200 boundary pixels, the number of common mutation points is between 60 and 85, and the boundary direction change annotation information is obtained.

[0028] See also Figure 2 and Figure 5 ,The structure deduction module includes the behavior trajectory construction submodule, the path prediction and ,fitting submodule, and the connectivity relationship judgment submodule; The behavior trajectory construction submodule extracts the tangential grayscale change direction sequence of each mutation point based on the boundary direction change annotation information, obtains the normal angle difference between adjacent points, cross-screens the direction mutation rate and normal angle mutation amplitude in the tangential sequence, extracts the point sequence that meets the mutation synchronization judgment rule, constructs a candidate trajectory point set, and generates the boundary behavior trajectory sample value; Based on the mutation point set recorded in the boundary direction change annotation information, the image coordinate information of the mutation point is first extracted. Each mutation point is identified by its pixel index in the image matrix. Indicates that a linear detection window is constructed around the point along the tangent direction of plus or minus 5 pixels, and the grayscale value of each pixel in the window is obtained. , construct the gray value direction sequence in order, and then perform adjacent difference operation on the sequence to calculate the gray value change gradient If the gradient shows the same sign and increasing and decreasing trend around the center point, the tangential grayscale change direction is determined to be continuous, and the sequence is recorded as a valid tangential grayscale direction sequence. At the same time, with the mutation point as the center, the angle between it and the two adjacent mutation points is extracted, and the normal angle difference is calculated. , if the angle difference is greater than the set angle mutation threshold , then mark the point as a normal mutation node. After the above processing is performed on all mutation points, two sequences are constructed according to the tangential direction mutation rate and the normal angle mutation amplitude of the mutation point. The two sequences are cross-screened. The screening strategy is that the corresponding index value at the intersection meets the requirements of mutation rate higher than 5 gray levels / pixel and angle mutation greater than 18 degrees. The screening result is the mutation synchronization point sequence. In the sample, if there are 160 mutation points in the boundary direction change annotation information, 41 candidate points are retained after synchronization screening. Their position coordinates, direction sequence, and angle data are integrated to construct a candidate trajectory point set and generate the boundary behavior trajectory sample value. The path prediction fitting submodule calls the boundary behavior trajectory sample value, inputs the candidate trajectory point sequence into the boundary prediction model, applies the learned weights obtained from the training to the sample point sequence in the model to determine the response intensity distribution, and calculates the mean absolute error between the fitting curve of the sample path and the true boundary. For path samples with error values below the error distribution stability threshold, the node sequence is extracted to obtain the stable path prediction fitting value; Call all candidate trajectory point sequences in the boundary behavior trajectory sample value, first construct trajectory samples for each point sequence in sequence, each trajectory sample contains no less than 5 consecutive mutation points, and then input the constructed point sequence into the trained boundary prediction model. The model contains the response intensity mapping weight matrix W obtained by convolutional network training. Each input trajectory sample point sequence is multiplied by the corresponding model weight vector to generate a node response value sequence. , then construct a quadratic spline function to approximate the trajectory sample to generate a fitting path, and calculate the mean absolute error (MAE) between the fitting path and the original boundary contour. MAE is defined as ,in represents the actual boundary point coordinates, Indicates the fitting path coordinates. If the error value is less than the set stable error threshold Pixels, the path is marked as a stable fitting path. After repeating the above operation for all samples, the path sample point sequence that meets the error condition is extracted, and its point coordinates are archived to obtain the stable path prediction fitting value; The connectivity judgment submodule predicts the fitting value of the stable path and calculates the tangent direction and extension angle between nodes according to the node sequence coordinate information. It compares the angle between the node direction vector and the adjacent normal angle, and determines that the node connection pairs whose angle range is within the threshold of the angle range are the connection structure basis points, generates a connectivity relationship group, and establishes the adhesion structure reconstruction judgment information. According to the coordinate information of all node sequences in the stable path prediction fitting value, the spatial coordinates between each two adjacent nodes are extracted in turn to construct the vector , after normalizing the vector, calculate its direction angle, which is the normal direction vector of the point The angle between , if the angle If the point pair falls within the set range of the clamping angle interval threshold, which is set to [40°, 70°], it is determined that the point pair is a connection structure basis point, and its coordinate position, vector number and corresponding angle value are recorded and filled in the relationship matrix. After performing the above judgment on each frame of the image, all point pairs that meet the conditions will be automatically written into the adhesion structure connectivity relationship matrix, and the matrix will be grouped by identification numbers to establish the adhesion structure reconstruction judgment information.

[0029] See also Figure 2 and Figure 6 ,The particle recognition module includes a connected node extraction submodule, a direction consistency calculation submodule, and a structure boundary judgment submodule; The connected node extraction submodule obtains boundary nodes distributed in the intersection area of non-closed paths based on the adhesion structure reconstruction judgment information, selects points in the adjacent node sequence whose angle fluctuation difference is greater than the angle change recognition threshold as mutation nodes, and obtains the number of boundary mutation nodes; Based on the connectivity judgment matrix recorded in the adhesion structure reconstruction judgment information, the boundary closure status information of each path node in the image frame is first extracted, and the coordinate trajectory of the path marked as "unclosed" is tracked to obtain all path termination points and their adjacent structures, and to determine whether the area is in the path junction area. The judgment standard for path junction is that the distance between the end of the path and any path node that is not its own structure is less than 3 pixels and the angle is greater than 30°, then it is judged as a junction area. After obtaining all the boundary nodes in the junction area, the direction vector of the continuous node sequence in the area is constructed into a coordinate slope sequence, and the difference processing is performed based on the difference of the direction angle sequence. If the angle fluctuation difference between three consecutive points is greater than the set angle change recognition threshold , then the point is marked as a mutation node. The threshold is set to 12°, which is set according to the minimum angle jump amplitude observed in the particle boundary. If the angle fluctuation is less than 12°, the morphological change is not representative. If it exceeds, it is regarded as a structural inflection point. The value varies with the roughness of the particle boundary. In the coarse boundary sample, the value can be increased to 15°. In the actual execution process, for example, there are 197 groups of non-closed paths in the image frame, and 654 nodes in the junction area are calculated. After calculating the angle sequence, it is found that the angle change of 87 nodes is greater than 12°, which is judged as a mutation node, and the number of mutation nodes is recorded as the number of boundary mutation nodes. The directional consistency calculation submodule calls the number of boundary mutation nodes, obtains the direction vector of each node and the grayscale centroid, calculates the angle range of the node connection path, obtains the cosine value of the angle between the node path direction and the centroid direction, calculates the directional consistency score of the node, marks the nodes with directional consistency scores lower than the directional mutation threshold as candidate structural nodes, and generates directional consistency screening results; The formula for calculating the direction consistency score of a node is as follows: ; in, represents the direction consistency score of the i-th node, Represents the direction angle of the path from the i-th node to the grayscale centroid, represents the average value of the node direction angle, Represents the angle between the path direction of the i-th node and the direction of the grayscale center of gravity, is the cosine value of the direction vector consistency, Represents the normalized value of the pixel distance from the i-th node to the grayscale centroid path; Call all mutation nodes in the boundary mutation node quantity value, and perform the following operations on each node: first obtain the direction vector to the image grayscale center of gravity coordinate point, recorded as , and then calculate the direction vector of the connection path before and after the mutation node , calculate the cosine of the angle using the vector angle formula , calculate the angle from each node to the center of gravity , the angle values of all nodes are averaged to get , calculate the difference between each node and the average angle and square it, combined with the cosine value of the direction vector angle Substituting into the formula: ; in is the normalized value of the distance from the node to the grayscale centroid pixel. The normalization process is to divide the original Euclidean distance by the diagonal pixel length of the image frame. The image frame resolution is 512×512 pixels, so the longest diagonal is Pixels, taking node k=12 as an example, the path direction angle is , the average direction angle is , the direction cosine value is , the pixel distance is 390 pixels, corresponding to the normalized distance is , substituting into: ; Direction consistency identification threshold used The setting basis is the lower limit of the minimum non-jitter node consistency score interval corresponding to the sample in the stable area of the image grayscale center offset. Nodes below this threshold are considered to have obvious direction offsets. Under different image frame density levels, the threshold adjustment range is 0.12–0.18. The specific value is controlled by the background noise variance. When the noise variance exceeds 30, it is recommended to increase it to 0.18. In this example, the node k=12 , does not belong to the candidate structural nodes. In the example, a total of 87 mutation nodes are extracted. After calculation, there are 28 nodes that meet the threshold, which constitute the candidate set and generate the direction consistency screening results; The structural boundary judgment submodule determines whether the angle change direction between adjacent nodes is continuous based on the direction consistency screening results, extracts the node pairs that meet the continuity direction switching threshold condition in the angle change sequence as the structural boundary reference points, and obtains the particle boundary decoupling node group; According to the candidate node set selected from the direction consistency screening results, read the numbers of every two adjacent nodes in the node list, obtain their direction vectors to construct the angle change sequence, and the determination basis of the angle change direction is the direction angle difference between the two adjacent nodes If the angle change direction has a polarity reversal between the front and rear nodes (i.e., from positive to negative or from negative to positive), and the angle change amplitude is greater than the direction switching threshold , it is considered as continuous direction switching behavior. The threshold setting refers to the actual angle range of the boundary contour reversal behavior on the edge of the particle image. The minimum reversal angle measured in 50 groups of samples is 19.7°, so 20° is selected as the benchmark threshold. This value does not change with the image resolution but is negatively correlated with the clarity of the sampled particle boundary. When the boundary blur increases, the reversal angle needs to be relaxed to 24°. The node pairs that meet this behavior are extracted and their image matrix position indexes are recorded as the boundary reference points of the particle structure. In the example, 13 groups of valid node pairs are formed among the 28 candidate nodes, and finally output as the particle boundary decoupling node group.

[0030] See also Figure 2 and Figure 7 ,The performance analysis module includes the particle structure calculation submodule, the density ,alienation measurement submodule, and the interference aggregation marking submodule; The particle structure calculation submodule is based on the particle boundary decoupling node group. According to the structural boundary point list of each particle, the number of pixels within the closed boundary of the region is obtained as the number of boundary area pixels. The first-order gradient calculation is performed on the grayscale change value in each pixel area, and the point where the grayscale gradient has the maximum value is recorded. The feature record item set of the structural particles is constructed to generate the particle structure feature quantity value. Based on the structural boundary point list of each particle in the particle boundary decoupling node group, the region is first filled according to the closed contour formed by the boundary nodes, all the pixels in the contour are extracted as the closed area mask, and then the number of valid pixels in the area is calculated as the boundary area pixel number of the current particle. , assuming the image frame resolution is 1024×1024, and the average size of each particle area is 20 pixels in diameter, then the pixels of the closed area of the circular particle are approximately Pixel, then the first-order gradient calculation is performed on the grayscale values of all pixels in the closed area, and the central difference method is used to derive the x and y directions respectively, and the gradient size is obtained. , search for the maximum gradient value in each particle area, the corresponding pixel is the point with the strongest grayscale mutation, record its grayscale value, coordinate index, and the number of the particle to which it belongs, and integrate them with the number of pixels in the boundary area as a feature item. Assuming that the number of particles in the image frame is 85, 85 feature records containing the above information will be generated in the end, and the feature quantity value of the particle structure will be generated; The density alienation calculation submodule calls the particle structure feature quantity value, obtains the particle grayscale center of gravity position based on the maximum point of the particle grayscale gradient and the number of pixels in the boundary area, and constructs a particle distribution vector group based on the central area of the image frame. It sequentially calculates the relative offset amplitude of the particle in the structure extension direction vector set, and combines its grayscale center of gravity density with the standard distribution density value of the image background to perform normalized offset judgment. It calculates the density alienation score of each particle unit and compares it with the alienation recognition threshold. Particles with density alienation scores greater than the alienation recognition threshold are selected as alienated units to obtain the density alienation assessment result. The formula for calculating the density alienation score of each particle unit is as follows: ; in, represents the density alienation score of the kth particle, represents the grayscale centroid density of the kth particle, Indicates the standard density of the background area of the image, Represents the distribution value of the kth particle in the jth dimension of the structure extension direction vector, Represents the mean value of the direction vector of the image background area in the jth dimension, represents the normalized value of the number of pixels of the boundary area of the kth particle; Call the maximum gray gradient point and boundary area pixel number of all particles in the particle structure feature quantity value , get the coordinate position of the particle grayscale center of gravity , based on the image frame center coordinates Calculate the offset vector , establish a Cartesian coordinate system with the center of the image as the origin, and construct the structural extension direction vector group of all particles , where each dimension represents a direction (such as horizontal, vertical, diagonal, etc.), calculate the particle distribution vector and the background average distribution vector The sum of squares of the Euclidean distances, the density of grayscale centroids and background density Calculate the absolute value of the difference and substitute it into the formula: , in is the normalized value of the boundary area, and the total number of pixels in the image is , if the area of the kth particle is 314 pixels, then ,set up 、 、 , substituting into the formula we get: ; The density alienation identification threshold in this formula The setting basis is the 75% percentile value of the difference distribution range between the background area density and the average particle density in the image frame, which means that under non-abnormal interference conditions, 75% of the particles are The value should be lower than this value. If it exceeds this value, it will be considered as an alienation anomaly. The threshold range varies with the image brightness mean. It should be increased to 0.30 when the brightness standard deviation is greater than 20, and should be reduced to 0.24 when the background variance is less than 15. In this example, , the particle is determined to be an alienated unit, and finally all particles that meet this condition are screened out to obtain the density alienation evaluation result; Based on the density alienation assessment results, the interference aggregation labeling submodule extracts the corresponding image position coordinates and sets number labels according to the sequence of screened alienated units. It then constructs a coordinate distribution layer, extracts cluster contour boundaries for the numbered positions, and integrates the center point position of the cluster area with the label to obtain particle behavior interference distribution information. According to the sequence of alienated units screened in the density alienation evaluation results, the pixel coordinate position of each unit in the image frame is extracted and numbered in sequence. , establish a coordinate index table, mark the coordinates corresponding to each number on the layer, use the minimum circumscribed circle method to perform cluster boundary recognition processing on all numbered points, identify the closed contour line formed by the edge points of the cluster area, obtain the center of gravity position of the contour and match it with the corresponding number one by one, output the center point coordinates of the cluster area and its number marking information on the image, integrate the center positions of all cluster contours and the corresponding particle information, and obtain the particle behavior interference distribution information.

[0031] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0032] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0033] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0034] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0035] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0036] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of the device or unit, which can be electrical, mechanical or other forms.

[0037] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0038] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0039] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical disks.

[0040] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. Cyst particle analysis system based on AI algorithm, characterized by: The system comprises: The particle image screening module obtains a sequence of image frames of the encapsulated lubricant in a liquid or semi-solid state, locates the central grayscale peak coordinates of the particle target area, and collects the pixel index information of the corresponding coordinates in the image matrix to generate a particle image grayscale index table; The boundary feature module calls the particle image grayscale index table, extracts the boundary pixel sequence within the range, determines the boundary trend change law based on the boundary angle difference range, and marks the boundary behavior mutation points to generate boundary direction change annotation information; The structure deduction module obtains the corresponding tangential grayscale value fluctuation direction sequence based on the boundary direction change annotation information, inputs the sample set into the boundary prediction model, determines the difference between the stable behavior sequence in the trajectory node distribution and the mean absolute error distribution of the boundary fitting curve based on the learning weights in the model, extracts the low-error fitting path to construct a closed connected structure, and obtains the adhesion structure reconstruction judgment information; The particle recognition module calls the adhesion structure reconstruction judgment information, calculates the angle range difference value and direction vector consistency coefficient between the node and the image grayscale centroid path, screens the structure boundary basis point, and generates a particle boundary decoupling node group.

2. The cyst particle analysis system based on AI algorithm according to claim 1, characterized in that: The particle image grayscale index table includes the particle grayscale peak point, image pixel position index, boundary grayscale transition value and target area grayscale density value; the boundary direction change annotation information specifically includes the boundary mutation marking layer, the angle mutation feature point set, the slope change distribution trajectory and the boundary angle difference distribution map; the adhesion structure reconstruction judgment information includes the closed path pixel sequence map, the trajectory stability difference matrix, the node connectivity judgment relationship set and the behavior trajectory error distribution map; the particle boundary decoupling node group specifically refers to the structure boundary node pair set, the grayscale center of gravity direction comparison set, the angle deviation analysis table and the boundary basis position index set.

3. The cyst particle analysis system based on AI algorithm according to claim 2, characterized in that: The particle image screening module includes: The image frame acquisition submodule acquires a sequence of image frames of the encapsulated lubricant in a liquid or semi-solid state, detects the image of each frame, determines whether the grayscale distribution range of the image falls within the image noise filtering threshold, selects a set of image frames that meet the distribution range requirements, and generates information on the number of image frames screened; The grayscale feature positioning submodule calls the image frame to filter the quantitative value information, analyzes the grayscale difference value of the pixel block area and the edge grayscale change rate in the image, determines that the area with the grayscale difference value above the grayscale contrast reference value is the particle target area, searches for the pixel with the maximum grayscale value in the particle target area and locates the grayscale peak coordinates, and generates a particle grayscale peak coordinate set; The pixel index extraction submodule retrieves the corresponding position index in the image according to the particle grayscale peak coordinate set, marks the image pixel row and column numbers and grayscale peak value range of the coordinate point, and integrates the record entries to generate a particle image grayscale index table.

4. The cyst particle analysis system based on AI algorithm according to claim 3, characterized in that: The boundary feature module includes: The boundary pixel extraction submodule calls the particle image grayscale index table, performs grayscale difference judgment on the neighborhood of pixels around each grayscale peak coordinate based on the recorded grayscale peak coordinates and edge grayscale change rate, selects boundary pixels whose grayscale gradient direction is consistent with the grayscale peak diffusion direction, numbers them, and generates a boundary pixel extraction quantity value; The tangent slope calculation submodule extracts the number of boundary pixels and sequentially obtains the pixel positions of three points in a group. The slope of the line connecting the first and third points in each group is calculated, and the slope is used as the tangent direction angle of the middle point. The difference between adjacent angles is calculated using the tangent direction angle to obtain a continuous angle change trend sequence for each point, thereby generating a boundary angle change slope value. Based on the boundary angle change slope value, the mutation annotation recognition submodule compares and analyzes the angle change slope of each boundary point with the angle fluctuation difference of the three consecutive points before and after, determines that the pixel point with an angle difference greater than the set boundary angle mutation threshold is the mutation point, and annotates the position of the mutation point to generate boundary direction change annotation information.

5. The cyst particle analysis system based on AI algorithm according to claim 4, characterized in that: The structure deduction module includes: The behavior trajectory construction submodule extracts the tangential grayscale change direction sequence of each mutation point based on the boundary direction change annotation information, obtains the normal angle difference between adjacent points, cross-screens the direction mutation rate and normal angle mutation amplitude in the tangential sequence, extracts the point sequence that meets the mutation synchronization judgment rule, constructs a candidate trajectory point set, and generates the boundary behavior trajectory sample value; The path prediction fitting submodule calls the boundary behavior trajectory sample value, inputs the candidate trajectory point sequence into the boundary prediction model, applies the learned weights obtained by training to the sample point sequence in the model to determine the response intensity distribution, and calculates the mean absolute error between the fitting curve of the sample path and the true boundary. For path samples whose error values are lower than the error distribution stability threshold, a node sequence is extracted to obtain a stable path prediction fitting value. The connectivity relationship judgment submodule is based on the stable path prediction fitting value and the node sequence coordinate information, and calculates the tangent direction and extension angle between the nodes respectively, compares the node direction vector with the adjacent normal angle direction, and judges that the node connection pairs whose angle range is within the threshold of the clamping angle interval are the connection structure basis points, generates a connectivity relationship group, and establishes the adhesion structure reconstruction judgment information.

6. The cyst particle analysis system based on AI algorithm according to claim 5, characterized in that: The particle identification module includes: The connected node extraction submodule obtains boundary nodes distributed in the intersection area of the non-closed path based on the adhesion structure reconstruction judgment information, selects points in the adjacent node sequence whose angle fluctuation difference is greater than the angle change recognition threshold as mutation nodes, and obtains the number of boundary mutation nodes; The directional consistency calculation submodule calls the number of boundary mutation nodes, obtains the direction vector of each node and the grayscale centroid, calculates the angle range of the node connection path, obtains the cosine value of the angle between the node path direction and the centroid direction, calculates the directional consistency score of the node, marks the nodes with directional consistency scores lower than the directional mutation threshold as candidate structural nodes, and generates directional consistency screening results; The structural boundary judgment submodule judges whether the angle change direction between adjacent nodes is continuous based on the direction consistency screening result, extracts the node pairs that meet the continuity direction switching threshold condition in the angle change sequence as the structural boundary reference points, and obtains the particle boundary decoupling node group.

7. The cyst particle analysis system based on AI algorithm according to claim 6, characterized in that: The formula for calculating the direction consistency score of the node is specifically: ； in, represents the direction consistency score of the i-th node, Represents the direction angle of the path from the i-th node to the grayscale centroid, represents the average value of the node direction angle, Represents the angle between the path direction of the i-th node and the direction of the grayscale center of gravity, Represents the normalized value of the pixel distance from the i-th node to the grayscale centroid path.

8. The cyst particle analysis system based on AI algorithm according to claim 7, characterized in that: The system further comprises: The performance analysis module calculates the particle density alienation index based on the particle boundary decoupling node group, according to the number of boundary area pixels and the maximum grayscale gradient point of each structural particle unit, and compares it with the standard distribution density of the image background area. It also marks the high-offset area, statistically distributes the position and number of the structural alienation aggregation units in the image, and obtains the particle behavior interference distribution information; The particle behavior interference distribution information includes interference frame classification labels, gradient distribution density maps, risk level mapping areas, and particle behavior change sections.

9. The cyst particle analysis system based on AI algorithm according to claim 8, characterized in that: The performance analysis module includes: The particle structure calculation submodule is based on the particle boundary decoupling node group and the structural boundary point list of each particle. It obtains the number of pixels within the closed boundary of the region as the number of boundary area pixels, and performs first-order gradient calculation on the grayscale change value in each pixel area. The point where the grayscale gradient has the maximum value is recorded, and the feature record item set of the structural particle is constructed to generate the particle structure feature quantity value. The density alienation calculation submodule calls the particle structure feature quantity value, performs normalization offset judgment based on the maximum grayscale gradient point and the number of boundary area pixels of the particle, combined with its grayscale center of gravity density and the image background standard distribution density value, calculates the density alienation score of each particle unit, and compares it with the alienation recognition threshold. Particles with density alienation scores greater than the alienation recognition threshold are selected as alienated units to obtain the density alienation evaluation result; Based on the density alienation assessment results, the interference aggregation labeling submodule extracts the corresponding image position coordinates and sets number labels according to the screened alienation unit sequence, constructs a coordinate distribution layer, extracts the cluster contour boundaries of the numbered positions, and integrates the center point position of the cluster area with the label to obtain particle behavior interference distribution information.

10. The cyst particle analysis system based on AI algorithm according to claim 9, characterized in that: The formula for calculating the density alienation score of each particle unit is specifically: ； in, represents the density alienation score of the kth particle, represents the grayscale centroid density of the kth particle, Indicates the standard density of the background area of the image, Represents the distribution value of the kth particle in the jth dimension of the structure extension direction vector, Represents the mean value of the direction vector of the image background area in the jth dimension, Represents the normalized value of the number of pixels of the boundary area of the kth particle.

Images