A method and system for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition

Abstract

This invention discloses an online uniformity monitoring method and system for TMR mixing process based on dynamic particle morphology recognition, relating to the field of intelligent monitoring technology in feed processing. The method includes: during material mixing, real-time acquisition of dynamic images of the material within the mixing chamber; identification of material particles in the dynamic images; extraction of the characteristic length of each particle; and calculation of skewness and kurtosis based on the characteristic length; division of the effective observation area of ​​the dynamic images into multiple sub-regions based on the centroid coordinates of all material particles in the dynamic images; extraction of particle characteristic statistics within each sub-region; and construction of a multidimensional spatial uniformity index set to characterize the mixing state of the material; input of the skewness, kurtosis, and multidimensional spatial uniformity index set into a pre-trained evaluation model, outputting a comprehensive uniformity index and corresponding quality level.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology in feed processing, and particularly relates to an online uniformity monitoring method and system for TMR mixing process based on dynamic particle morphology recognition. Background Technology

[0002] The uniformity of total mixed ration (TMR) mixing is a core technological indicator in modern livestock farming, directly affecting feed utilization efficiency, animal health, and production performance. Uneven mixing can easily lead to picky eating and nutrient imbalances in animals, while excessive mixing can damage the physical structure of the feed and affect rumination. Currently, the mainstream assessment methods for TMR uniformity in the industry all have significant limitations.

[0003] Traditional manual sensory evaluation relies on the operator's personal experience, lacks objective quantitative standards, and results in subjective and poor repeatability, making it difficult to support standardized production management. While offline sampling and analysis methods offer some objectivity, they require interrupting production to take samples from different points in the mixing chamber for laboratory sieving or chemical testing. This process is not only cumbersome and time-consuming, with severely delayed results, but the sampling itself also disrupts the original three-dimensional distribution of the material, failing to provide real-time feedback for the production process.

[0004] Some automation technologies attempt to improve upon these problems, but fundamental flaws remain. For example, static surface analysis requires stopping stirring and mechanically flattening the material surface, then assessing uniformity by analyzing surface grayscale images. This method not only disrupts continuous production, but its measurement object is actually the processed surface optical properties (such as color and humidity), rather than the material's key physical morphological distribution (such as fiber length and grain integrity). Another online analysis technology based on a single optical feature attempts to achieve uninterrupted monitoring, but it largely relies on macroscopic color or texture statistics of the entire image. In complex working conditions with high dust levels, fluctuating lighting, and materials of similar color, its anti-interference capability is insufficient, and it also cannot effectively quantify the core physical uniformity indicator of "length and thickness."

[0005] In summary, existing technologies either fail to achieve real-time online monitoring or, while achieving online monitoring, fail to address the actual process requirement of uniform physical distribution. Therefore, the industry urgently needs an innovative technological solution that can achieve real-time, objective, and accurate assessment of the uniformity of the physical distribution of TMR materials without interrupting production. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes an online uniformity monitoring method for TMR stirring processes based on dynamic particle morphology recognition, comprising: S1, during the material mixing process, real-time dynamic images of the material in the mixing chamber are collected, material particles in the dynamic images are identified, the characteristic length of each material particle is extracted, and skewness and kurtosis are calculated based on the characteristic length. S2, based on the centroid coordinates of all material particles in the material dynamic image, divide the effective observation area of ​​the material dynamic image into multiple sub-regions; S3, extract the particle feature statistics in each sub-region and construct a set of multidimensional spatial uniformity indices to characterize the mixing state of materials; S4 takes a set of skewness, kurtosis and multidimensional spatial uniformity indices as input into the pre-trained evaluation model and outputs a comprehensive uniformity index and the corresponding quality level.

[0007] Furthermore, before identifying material particles in the dynamic image of the material in step S1, the process also includes: A positive pressure air curtain is continuously applied to the camera to prevent contaminants from adhering. The clarity of the material dynamic image is calculated by the pixel gradient change of the material dynamic image. When the clarity drops below a preset clarity threshold, a window cleaning or enhanced maintenance operation is automatically triggered.

[0008] Furthermore, the extraction of the characteristic length of each material particle in step S1 includes: calculating the minimum bounding rectangle for each material particle, and using the length of the longer side of the minimum bounding rectangle as the characteristic length of the corresponding material particle.

[0009] Furthermore, in step S2, dividing the effective observation area of ​​the material dynamic image into multiple sub-regions includes: calculating the distribution center point of all material particles based on the centroid coordinates of all material particles, determining the main direction of the distribution of all material particles through principal component analysis, using the distribution center point as the center of the effective observation area, rotating the coordinate system to align the main direction with the horizontal axis, and uniformly dividing the effective observation area into multiple sub-regions of equal area under the aligned coordinate system.

[0010] Furthermore, the particle feature statistics in step S3 include: the average feature length, length variation coefficient, and particle density of each sub-region.

[0011] Step S3 involves constructing a set of multidimensional spatial homogeneity indices to characterize the mixing state of materials, including: Calculate the first spatial homogeneity index: in, The first spatial homogeneity index, For the first The average feature length of each sub-region; Calculate the second spatial homogeneity index: in, As a second spatial homogeneity index, For the first The coefficient of variation of the length of each subregion; Calculate the homogeneity index of the third space: in, As a third spatial homogeneity index, For the first Particle density in each sub-region; Calculate the homogeneity index of the fourth space: in, As the fourth spatial homogeneity index, This represents the maximum absolute value of the difference in average feature length between sub-regions in each row. It is the maximum absolute value of the difference in average feature length between each column of sub-regions; The first spatial uniformity index, the second spatial uniformity index, the third spatial uniformity index, and the fourth spatial uniformity index constitute a multidimensional spatial uniformity index set.

[0012] Furthermore, it also includes S5: real-time display of the comprehensive uniformity index and the corresponding quality level, and triggering corresponding alarm or control signals according to preset rules.

[0013] This invention also proposes an online uniformity monitoring system for TMR stirring processes based on dynamic particle morphology recognition, comprising: The particle recognition module is used to collect real-time dynamic images of materials in the mixing chamber during the material mixing process, identify material particles in the dynamic images, extract the characteristic length of each material particle, and calculate skewness and kurtosis based on the characteristic length. The region division module is used to divide the effective observation area of ​​the material dynamic image into multiple sub-regions based on the centroid coordinates of all material particles in the material dynamic image; The index construction module is used to extract particle feature statistics in each sub-region and construct a set of multi-dimensional spatial uniformity indices to characterize the mixing state of materials. The monitoring module is used to input a set of skewness, kurtosis and multidimensional spatial uniformity indices into the pre-trained evaluation model and output a comprehensive uniformity index and the corresponding quality level.

[0014] Furthermore, the particle recognition module includes the following steps before recognizing material particles in a dynamic image of the material: A positive pressure air curtain is continuously applied to the camera to prevent contaminants from adhering. The clarity of the material dynamic image is calculated by the pixel gradient change of the material dynamic image. When the clarity drops below a preset clarity threshold, a window cleaning or enhanced maintenance operation is automatically triggered.

[0015] Furthermore, the particle recognition module extracts the feature length of each material particle by: calculating the minimum bounding rotation rectangle for each material particle, and using the length of the longer side of the minimum bounding rotation rectangle as the feature length of the corresponding material particle.

[0016] Furthermore, the region division module divides the effective observation area of ​​the material dynamic image into multiple sub-regions, including: calculating the distribution center point of all material particles based on the centroid coordinates of all material particles, determining the main direction of the distribution of all material particles through principal component analysis, using the distribution center point as the center of the effective observation area, rotating the coordinate system to align the main direction with the horizontal axis, and uniformly dividing the effective observation area into multiple sub-regions of equal area under the aligned coordinate system.

[0017] Compared with the prior art, the present invention has the following advantages and technical effects: Online monitoring of the mixing process enables non-invasive real-time monitoring, ensuring the continuity of production and the authenticity of data. By analyzing the individual physical morphology (characteristic length) of tens of thousands of particles in the dynamic image of the material, and combining it with the regional statistical method, the spatial distribution is refined and quantified, thereby essentially reflecting the mixing uniformity of the material in terms of "length and thickness". It features an automatic cleaning mechanism based on clarity detection to maintain reliable operation; It provides a quantitative comprehensive uniformity index and a visual interface, offering an objective basis for accurately determining the "mixing endpoint" and achieving energy conservation, consumption reduction, and quality improvement. Attached Figure Description

[0018] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a system structure diagram of Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of a mixing device. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0021] The following embodiments of the present invention are all applied to, for example... Figure 3 The mixing equipment shown is used, but it should be noted that... Figure 3 The mixing equipment described is only for illustrative purposes and does not serve as a limitation.

[0022] Example 1 like Figure 1 As shown in the figure, this embodiment proposes an online uniformity monitoring method for TMR stirring process based on dynamic particle morphology recognition, which specifically includes the following steps: S1, during the material mixing process, real-time dynamic images of the material in the mixing chamber are collected, material particles in the dynamic images are identified, the characteristic length of each material particle is extracted, and skewness and kurtosis are calculated based on the characteristic length. Preferably, in this embodiment, dynamic images of materials are acquired in the following manner: The default frame rate is 5 frames per second, and it can be automatically adjusted according to the material flow rate within the range of 1 to 30 frames per second; the resolution of each frame image is 1920×1080 pixels, using 8-bit grayscale format (i.e., 256 levels of grayscale); the acquisition process is completely synchronized with the mixing operation, starting from the start of mixing and stopping at the end of mixing, realizing continuous recording of the entire process.

[0023] Specifically, before identifying material particles in the dynamic image of the material in step S1, the process also includes: A positive pressure air curtain is continuously applied to the camera to prevent contaminants from adhering. The clarity of the material dynamic image is calculated by the pixel gradient change of the material dynamic image. When the clarity drops below a preset clarity threshold, a window cleaning or enhanced maintenance operation is automatically triggered.

[0024] For example, this embodiment can automatically trigger window cleaning or enhanced maintenance operations through the following steps: 1. Initialization and Image Acquisition: After stirring begins, the camera captures grayscale images at a resolution of 1280×720 at a rate of 5 frames per second; A positive pressure of 0.3 kPa is maintained at the air curtain to prevent pollutants from adhering. The brightness of the LED light source is dynamically adjusted by using a PID control algorithm based on real-time feedback from the photosensitive sensor, ensuring that the illuminance of the working area is stable within the range of 500±50 lux. To eliminate the influence of lens edge distortion, the central 1536×864 pixel region of each frame image (approximately 80% of the original image area) is extracted as the selected region for subsequent processing.

[0025] 2. Intelligent cleaning trigger decision: Based on the severity of the clarity degradation, a tiered cleaning strategy is implemented, namely: Benchmark calibration: Acquire 100 frames of images with the camera in a clean state, calculate the sharpness of each frame, and obtain the average sharpness of the 100 frames. Clarity serves as a benchmark for cleanliness; The Brenner gradient algorithm was used to calculate the sharpness of each frame in 100 images. The calculation formula is as follows: in, For the sharpness of each frame of the image, For pixels grayscale value, For pixels grayscale value; Degradation rate calculation: Real-time calculation of current resolution Rate of decrease in clarity relative to cleaning baseline : 3. Three-level triggering mechanism: when A warning will be issued when the level is ≥15%. when When the pressure is ≥30%, the standard cleaning mode is triggered (0.6 kPa, lasting 5 seconds).

[0026] when When the pressure is ≥50%, a powerful cleaning mode is triggered (1.0 kPa air pressure for 10 seconds). For example, the air curtain pressure is instantly increased to 1.0 kPa and lasts for 10 seconds to remove stubborn deposits with strong airflow, and then the pressure is automatically restored to normal.

[0027] 4. Cleaning execution: The gas pressure is adjusted by controlling the solenoid valve, and the cleaning nozzle below the camera is activated to complete the lens cleaning action.

[0028] 5. Cleaning effect verification and feedback After the cleaning operation is completed, wait for a predetermined time until the airflow stabilizes, then continuously acquire multiple frames of images (e.g., acquire 5 frames immediately) and evaluate the effect. Calculate the sharpness restoration rate: in, For clarity restoration rate, For clarity before cleaning, The clarity after cleaning.

[0029] Verification standard: If the sharpness restoration rate If the cleaning rate is less than 70%, the cleaning is deemed incomplete and will be automatically rechecked and restarted after 30 minutes.

[0030] Specifically, step S1 involves extracting the characteristic length of each material particle by: calculating the minimum bounding rectangle for each material particle and using the length of the longer side of the minimum bounding rectangle as the characteristic length of the corresponding material particle.

[0031] Preferably, the acquired material dynamic images are subjected to Gaussian filtering to smooth noise, followed by contrast-limited adaptive histogram equalization (CLAHE) to enhance image details. Then, Canny edge detection and Niblack local adaptive thresholding algorithm are fused to separate the foreground target from the background. Finally, the average area of ​​connected components in the binary image is calculated. Based on this, the structuring element radius of the morphological operation is dynamically determined, i.e., the opening operation radius is... The radius of the closed operation is To effectively separate mutually adhering particles, the minimum bounding rectangle of each individual particle's connected component is calculated. The length of the longer side of this rectangle is used as the feature length of the particle, and the length of the first pixel is recorded. The centroid coordinates of each particle Finally, based on the characteristic lengths of all particles, the skewness of the overall length distribution is calculated. With kurtosis .

[0032] Preferably, in this embodiment, the skewness of the overall length distribution is calculated in the following manner. With kurtosis : in, This represents the number of all material particles in the material dynamic image. The mean of the characteristic lengths of all material particles. is the standard deviation of the characteristic length of all material particles.

[0033] S2, based on the centroid coordinates of all material particles in the material dynamic image, divide the effective observation area of ​​the material dynamic image into multiple sub-regions. For example, divide the effective observation area into multiple sub-regions in the form of a nine-square grid, with each grid serving as a sub-region. Specifically, step S2, which divides the effective observation area of ​​the material dynamic image into multiple sub-regions, includes: calculating the distribution center point of all material particles based on the centroid coordinates of all material particles, determining the main direction of the distribution of all material particles through principal component analysis, using the distribution center point as the center of the effective observation area, rotating the coordinate system to align the main direction with the horizontal axis, and uniformly dividing the effective observation area into multiple sub-regions of equal area under the aligned coordinate system.

[0034] S3, extract the particle feature statistics in each sub-region and construct a set of multidimensional spatial uniformity indices to characterize the mixing state of materials; Specifically, the particle feature statistics in step S3 include: the average feature length, length variation coefficient, and particle density of each sub-region.

[0035] Step S3 involves constructing a set of multidimensional spatial homogeneity indices to characterize the mixing state of materials, including: Calculate the first spatial homogeneity index: in, The first spatial homogeneity index, For the first The average feature length of each sub-region; Calculate the second spatial homogeneity index: in, As a second spatial homogeneity index, For the first The coefficient of variation of the length of each subregion; Preferably, this embodiment calculates in the following way : in, For the first The first sub-region The characteristic length of each material particle; Calculate the homogeneity index of the third space: in, As a third spatial homogeneity index, For the first Particle density in each sub-region; Preferably, this embodiment calculates in the following way : in, For the first The number of material particles in each sub-region For the first The effective area of ​​each sub-region; Calculate the homogeneity index of the fourth space: in, As the fourth spatial homogeneity index, This represents the maximum absolute value of the difference in average feature length between sub-regions in each row. It is the maximum absolute value of the difference in average feature length between each column of sub-regions; The first spatial uniformity index, the second spatial uniformity index, the third spatial uniformity index, and the fourth spatial uniformity index constitute a multidimensional spatial uniformity index set.

[0036] S4 takes a set of skewness, kurtosis and multidimensional spatial uniformity indices as input into the pre-trained evaluation model and outputs a comprehensive uniformity index and the corresponding quality level.

[0037] Preferably, the skewness With kurtosis Together with four spatial uniformity indicators , , , Together they form a feature vector .

[0038] Then, the feature vector The input is fed into a pre-trained evaluation model (such as XGBoost or a neural network) that has been pre-trained based on historical data. The pre-trained evaluation model analyzes the feature vectors. The inherent correlations and weights are used to output a comprehensive uniformity index (FEI) ranging from 0 to 100. An exemplary calculation of the comprehensive uniformity index FEI can be expressed as: ,in , , , , , denoted as weight coefficients, where all weight coefficients are automatically learned through model training.

[0039] Regarding model training, for example, it could be: Training data: Collected data from 1000 batches of mixing, including image sequences and laboratory test results; Data annotation: Convert the uniformity of the test results into a FEI value of 0-100; Model selection: XGBoost regression model; Parameter settings: n_estimators=100, max_depth=6, learning_rate=0.1; Performance requirements: 5-fold cross-validation R² ≥ 0.85, MAE ≤ 5 points.

[0040] The model outputs a comprehensive uniformity index (FEI) (0-100 points).

[0041] Finally, the mixing quality is intelligently graded based on the comprehensive uniformity index (FEI) score: 85 points and above is "excellent", 70 to 84 points is "good", 55 to 69 points is "medium", and below 55 points is "poor".

[0042] S5: Displays the comprehensive uniformity index and corresponding quality level in real time, and triggers corresponding alarms or control signals according to preset rules.

[0043] Preferably, the following key information is displayed in real time: 1. Real-time updated Comprehensive Evenness Index (FEI) values; 2. The current corresponding quality level; 3. A visual view of sub-regions (e.g., a nine-square grid) reflecting the spatial distribution of particles; 4. Show the trend of the comprehensive evenness index FEI over time.

[0044] For example, preset rules that trigger corresponding alarms or control signals can be: When the overall uniformity index (FEI) is detected to remain stable at 85 points ("Excellent") or above for 30 consecutive seconds, a "mixing meets the standard" alarm will be automatically triggered. This alarm is intended to notify the operator that it is safe to stop mixing or carry out subsequent feeding operations, thereby achieving the goal of saving energy and improving production efficiency.

[0045] Example 2 like Figure 2 As shown in the figure, this embodiment proposes an online uniformity monitoring system for TMR stirring process based on dynamic particle morphology recognition, which specifically includes the following modules: The particle recognition module is used to collect real-time dynamic images of materials in the mixing chamber during the material mixing process, identify material particles in the dynamic images, extract the characteristic length of each material particle, and calculate skewness and kurtosis based on the characteristic length. The region division module is used to divide the effective observation area of ​​the material dynamic image into multiple sub-regions based on the centroid coordinates of all material particles in the material dynamic image; The index construction module is used to extract particle feature statistics in each sub-region and construct a set of multi-dimensional spatial uniformity indices to characterize the mixing state of materials. The monitoring module is used to input a set of skewness, kurtosis and multidimensional spatial uniformity indices into the pre-trained evaluation model and output a comprehensive uniformity index and the corresponding quality level.

[0046] The control module is used to display the comprehensive uniformity index and the corresponding quality level in real time, and to trigger corresponding alarms or control signals according to preset rules.

[0047] Since the system technical solution of this embodiment 2 is based on the technical solution of embodiment 1, it will not be described again.

[0048] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition, characterized in that, include: S1, During the material mixing process, real-time dynamic images of the material in the mixing chamber are collected, material particles in the dynamic images are identified, the characteristic length of each material particle is extracted, and skewness and kurtosis are calculated based on the characteristic length. S2, based on the centroid coordinates of all material particles in the material dynamic image, divide the effective observation area of ​​the material dynamic image into multiple sub-regions; S3, extract the particle feature statistics in each sub-region and construct a set of multidimensional spatial uniformity indices to characterize the mixing state of materials; S4, input the set of skewness, kurtosis and multidimensional spatial uniformity indices into the pre-trained evaluation model, and output the comprehensive uniformity index and the corresponding quality level.

2. The method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition according to claim 1, characterized in that, Before identifying material particles in the material dynamic image in step S1, the method further includes: continuously applying positive pressure to the camera to prevent contaminants from adhering; The clarity of the material dynamic image is calculated by the pixel gradient change of the material dynamic image. When the clarity drops below a preset clarity threshold, a window cleaning or enhanced maintenance operation is automatically triggered.

3. The method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition according to claim 1, characterized in that, Step S1 involves extracting the characteristic length of each material particle by: calculating the minimum bounding rectangle for each material particle, and using the length of the longer side of the minimum bounding rectangle as the characteristic length of the corresponding material particle.

4. The method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition according to claim 1, characterized in that, Step S2, which divides the effective observation area of ​​the material dynamic image into multiple sub-regions, includes: calculating the distribution center point of all material particles based on the centroid coordinates of all material particles, determining the main direction of the distribution of all material particles through principal component analysis, using the distribution center point as the center of the effective observation area, rotating the coordinate system to align the main direction with the horizontal axis, and uniformly dividing the effective observation area into multiple sub-regions of equal area under the aligned coordinate system.

5. The method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition according to claim 1, characterized in that, The particle feature statistics in step S3 include: the average feature length, length variation coefficient, and particle density of each sub-region. Step S3 involves constructing the set of multidimensional spatial homogeneity indices to characterize the mixing state of materials, including: Calculate the first spatial homogeneity index: in, The first spatial homogeneity index, For the first The average feature length of each sub-region; Calculate the second spatial homogeneity index: in, As a second spatial homogeneity index, For the first The coefficient of variation of the length of each subregion; Calculate the homogeneity index of the third space: in, As a third spatial homogeneity index, For the first The particle density of each sub-region; Calculate the homogeneity index of the fourth space: in, As the fourth spatial homogeneity index, This represents the maximum absolute value of the difference in average feature length between sub-regions in each row. It is the maximum absolute value of the difference in average feature length between each column of sub-regions; The first spatial uniformity index, the second spatial uniformity index, the third spatial uniformity index, and the fourth spatial uniformity index constitute a multidimensional spatial uniformity index set.

6. The method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition according to claim 1, characterized in that, It also includes S5: displaying the comprehensive uniformity index and the corresponding quality level in real time, and triggering corresponding alarm or control signals according to preset rules.

7. An online uniformity monitoring system for TMR stirring process based on dynamic particle morphology recognition, characterized in that, include: The particle recognition module is used to acquire dynamic images of materials in the mixing chamber in real time during the material mixing process, identify material particles in the dynamic images, extract the characteristic length of each material particle, and calculate skewness and kurtosis based on the characteristic length. The region division module is used to divide the effective observation area of ​​the material dynamic image into multiple sub-regions based on the centroid coordinates of all material particles in the material dynamic image; The index construction module is used to extract particle feature statistics in each sub-region and construct a set of multi-dimensional spatial uniformity indices to characterize the mixing state of materials. The monitoring module is used to input the set of skewness, kurtosis and multidimensional spatial uniformity indicators into the pre-trained evaluation model, and output the comprehensive uniformity index and the corresponding quality level.

8. The TMR stirring process online uniformity monitoring system based on dynamic particle morphology recognition according to claim 7, characterized in that, Before identifying material particles in the dynamic image of the material in the particle recognition module, the process also includes: continuously applying positive pressure to the camera to prevent contaminants from adhering; The clarity of the material dynamic image is calculated by the pixel gradient change of the material dynamic image. When the clarity drops below a preset clarity threshold, a window cleaning or enhanced maintenance operation is automatically triggered.

9. The method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition according to claim 7, characterized in that, Extracting the feature length of each material particle in the particle recognition module includes: calculating the minimum bounding rectangle for each material particle, and using the length of the longer side of the minimum bounding rectangle as the feature length of the corresponding material particle.

10. The method for online uniformity monitoring of TMR stirring process based on dynamic particle morphology recognition according to claim 7, characterized in that, The region division module divides the effective observation area of ​​the material dynamic image into multiple sub-regions, including: calculating the distribution center point of all material particles based on the centroid coordinates of all material particles, determining the main direction of the distribution of all material particles through principal component analysis, using the distribution center point as the center of the effective observation area, rotating the coordinate system to align the main direction with the horizontal axis, and uniformly dividing the effective observation area into multiple sub-regions of equal area under the aligned coordinate system.

Images