Plasma melt-blown powder spheroidizing method based on image recognition, medium and equipment

Abstract

The invention belongs to the technical field of metal powder preparation, and discloses a plasma melt-blown powder spheroidizing method, medium and equipment based on image recognition, the method comprises the following steps: S1, determining process parameters required by plasma melt-blown powder spheroidizing, and starting cooling water; s2, metal powder is contained in a powder feeder; s3, a vacuum pump valve is opened, and an Ar gas valve, a collector valve and a powder feeder valve are closed; s4, the vacuum system is stopped, an Ar gas valve is opened, the vacuum pump is closed, and Ar gas is introduced into the powder falling bin; s5, opening a powder feeder valve and a collector valve; s6, ignition is conducted after the Ar flow is adjusted; s7, a powder feeder is started, and plasma spheroidizing is started; s8, carrying out online intelligent diagnosis and control on the spheroidizing process; s9, after spheroidizing is finished, the powder feeder and the airflow of the powder feeder are closed; and S10, collecting a spheroidized product. The automatic level of the powder spheroidizing process can be remarkably improved, manual intervention is reduced, the production efficiency is improved, and the consistency and reliability of product quality are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of metal powder preparation, and in particular to a plasma melt-blown powder spheroidization processing method, medium and equipment based on image recognition. Background Art

[0002] In the field of powder metallurgy and materials science, plasma melt-blowing technology is a commonly used method for preparing high-quality powders, especially when preparing metal or alloy powders. These powders are widely used in many fields such as 3D printing, coating preparation, catalyst carriers and powder metallurgy. Plasma melt-blowing technology uses the heat energy of a high-temperature plasma flame to melt the powder particles and quickly solidify them to form spherical powders by feeding the powder into the flame. Although plasma melt-blowing technology has significant advantages in powder preparation, the complexity of its process control also brings a series of challenges. Traditional control of flame shape and process parameters relies on the operator's experience and manual adjustment, which is not only inefficient, but also difficult to ensure the consistency and repeatability of product quality. In addition, slight changes in flame shape may significantly affect the spheroidization effect of the powder, including the size, shape and composition uniformity of the particles, which in turn affects the mechanical properties and application performance of the final product.

[0003] The existing method mainly involves sampling and testing the spheroidized powder, and then using electron microscope scanning to observe and analyze the morphology, size and microstructure of the spherical powder particles. The specific steps are as follows: ① Sample preparation: First, you need to prepare an appropriate amount of spherical powder samples. Usually, the powder samples are evenly dispersed on a conductive tape or a dedicated sample holder to ensure its stability and conductivity in the electron microscope.

[0004] ② Surface conductive treatment: In order to reduce charge accumulation, it may be necessary to spray gold or carbon on the sample surface to improve its conductivity and prevent the sample from being charged under electron beam irradiation.

[0005] ③ Place in electron microscope: Place the prepared sample in a scanning electron microscope (SEM) or transmission electron microscope (TEM). These devices can provide high-resolution images to observe the surface morphology and internal structure of the sample.

[0006] ④Parameter setting: Set the working parameters of the electron microscope, such as acceleration voltage, working distance, magnification, etc., according to the powder characteristics to be observed.

[0007] ⑤ Image acquisition: Under the set parameters, the images of powder particles are collected. SEM can provide surface morphology and size information of powder particles, while TEM can observe the internal structure and crystal defects of particles.

[0008] ⑥Data analysis: Analyze the collected images through electron microscope software, measure the size, distribution, sphericity and other parameters of the particles, and evaluate whether the particles have cracks, holes or other defects.

[0009] ⑦ Report generation: Generate a detailed inspection report based on the image analysis results. The report usually contains particle statistics, morphological feature descriptions, and possible quality issues.

[0010] ⑧Quality Control: The results of electron microscope scanning can be used to control the quality of spherical powders to ensure that they meet specific industrial application standards.

[0011] Scanning electron microscope detection is an important means of quality assessment, but this method is too time-consuming, and during transportation, the spheroidized powder is too active, which may produce analytical results that are inconsistent with the actual results.

[0012] In the process of testing the quality of plasma melt-blown powder spheroidization, sampling inspection and electron microscope scanning inspection of spherical powder may have the following disadvantages: 1. Disadvantages of sampling inspection: Sampling may not fully represent the quality of the entire batch of powder, and there may be accidental errors; the sampling process may introduce contamination or change the original state of the powder, affecting the accuracy of the test results; sampling inspection requires time and resources, and may not be able to achieve real-time monitoring and rapid feedback; 2. Disadvantages of electron microscope scanning detection: Electron microscope scanning detection usually requires a complex sample preparation process, which may affect the natural state of the powder; scanning electron microscope can only provide two-dimensional images, which may not fully reflect the true three-dimensional morphology of powder particles; the test results may be limited by the skills and experience of the operator and are subjective; electron microscope detection may not be able to quantify sphericity, making it difficult to accurately evaluate sphericity; for large-scale production processes, electron microscope scanning detection may not meet the needs of high efficiency and high throughput; 3. Quantification of sphericity detection: Traditional SEM detection methods cannot quantify sphericity, making the results difficult to use for quality control and comparison; additional software tools are required to assist in quantitative analysis, which increases the complexity and time cost of detection; 4. Cost and resources: High-precision testing equipment and software are usually expensive and require professional operation and maintenance, which increases the cost of testing. Summary of the invention

[0013] The embodiments of the present invention provide a plasma melt-blown powder spheroidization processing method, medium and equipment based on image recognition to solve the problem that the powder spheroidization process in the prior art relies on empirical adjustment and offline detection, and the flame state cannot be monitored in real time, resulting in unstable quality of spheroidized powder and difficulty in meeting high-precision requirements. In order to have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary is not a general review, nor is it intended to identify key / important components or describe the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a preface to the detailed description that follows.

[0014] According to a first aspect of an embodiment of the present invention, a plasma melt-blown powder spheroidization processing method based on image recognition is provided; the method comprises the following steps: S1: Equipment inspection and preparation: Determine the process parameters required for plasma melt-blown powder spheroidization, and after confirmation, turn on the cooling water; S2: Put the metal powder into the powder feeder and close the powder bin; S3: Open the vacuum pump valve, close the Ar gas valve, collector and powder feeder valve, start the vacuum system, and evacuate at room temperature. When the air pressure in the powder drop bin drops below 10Pa, maintain it; S4: Stop the vacuum system, open the Ar gas valve, close the vacuum pump valve, and introduce Ar gas into the powder dropping bin to make the pressure in the powder dropping bin reach +2000Pa. Stop introducing Ar gas and let it stand; S5: Open the powder feeder valve and collector valve; S6: Adjust the Ar flow rate and ignite, monitor the pressure in the powder drop bin, and release the pressure in time if it is too high; S7: Turn on the powder feeder and start plasma spheroidization; S8: Online intelligent diagnosis and control of the spheroidization process; S9: After the spheroidization is completed, the powder feeder and its airflow are turned off, the plasma power supply is turned off, and then the Ar gas is stopped; S10: After cooling to room temperature, turn off the cooling water in the powder dropping bin, disassemble the powder dropping bin and collector, collect the spheroidized product, and store the product in a sealed bag.

[0015] In one embodiment, the method, the step S9 of performing online intelligent diagnosis and control on the spheroidization process, specifically comprises: Step 1: original flame image data acquisition; Step 2: flame image data preprocessing: preprocessing is performed based on the collected original flame image data to obtain a clear flame image with a clear flame outline; Step 3: Flame shape feature extraction: Use image analysis algorithms to identify the edge, area, and color of the flame and quantify the morphological characteristics of the flame; at the same time, record the stability and energy distribution of the flame; Step 4: Dataset segmentation: The data obtained in step 3 after feature engineering of the flame shape features is segmented. According to the ratio of 8:2, 80% of the data is allocated as a training set to train the model, and the remaining 20% ​​constitutes a validation set to evaluate the accuracy of the model. Step 5: Analysis of the correlation between flame characteristics and spheroidized powder quality: Combining the flame shape characteristics and the quality evaluation data of spheroidized powder, using statistical methods, predict the correlation between different flame morphological characteristics and spheroidized powder quality; using feature selection technology, identify key features that significantly affect the quality of spheroidized powder; Step 6: Model construction and performance optimization: Using the extracted flame characteristics and spheroidized powder quality data, a convolutional neural network model based on deep learning is constructed, and combined with machine learning algorithms such as decision trees, random forests, and support vector machines to form a comprehensive prediction model; through cross-validation and hyperparameter tuning, the model is iteratively trained until the model has the best accuracy on the validation set; Step 7, feature importance assessment: covers the feature importance score calculated using the Permutation importance method, and based on the prediction accuracy of the convolutional neural network model, decision tree, random forest, and support vector machine model in step 6, weighted processing is performed on the feature importance score of each model to obtain the final importance score of the feature, and the key features with the highest scores are screened out based on this, which is used as the basis for analyzing the factors affecting the quality of the spheroidized powder finished product; Step 8: Automatic adjustment of process parameters: Based on the model prediction results, develop an intelligent feedback control system, analyze the prediction output in real time, and automatically adjust the key process parameters of the plasma melt-blowing equipment. Through precise control, optimize the flame shape and improve the production quality of spheroidized powder.

[0016] In one embodiment, the method, step 3 of flame shape feature extraction: using an image analysis algorithm to identify the edge, area, and color of the flame, and quantify the morphological characteristics of the flame; at the same time, recording the stability and energy distribution state of the flame, specifically includes: S31: Feature calculation: Calculate the number of pixels in the segmented flame area, calculate the length of the flame boundary, and calculate multiple parameters to describe the flame shape, as follows: When calculating the flame interview, first traverse each pixel of the image, and add 1 unit for each pixel that exists. The final value of the counter is the number of pixels in the flame area. Where m is the number of image rows; n is the number of image columns; I(i,j) is the pixel value at position (i,j); When calculating the boundary length of the flame area, first select a boundary pixel of the flame area and move pixel by pixel along the edge of the flame; then for each boundary pixel, check its 8 neighbors and determine the next boundary pixel; then record the positions of all boundary pixels; finally, calculate the Euclidean distance, which is calculated as follows: Where N is the total number of boundary pixels; ( , ) is the coordinate of the kth boundary pixel; ( , ) is the coordinate of the k+1th boundary pixel; The shape of the flame is calculated using circularity, rectangularity and eccentricity. The specific formula is as follows: Circle shape: ; Rectangularity: ; Eccentricity: ; S32: The gray level co-occurrence matrix is ​​used to calculate the texture features of the flame area, such as energy, contrast, uniformity and directionality. The calculation formula is as follows: Where L is the number of gray levels, P(i,j) is the frequency of co-occurrence of gray levels i and j in the gray level co-occurrence matrix; Where |ij| represents the difference between gray levels i and j, and P(i,j) is the frequency of co-occurrence of gray levels i and j; Directivity D(θ)=

[0017] in, is at an angle The frequency of gray levels i and j appearing together; is the change in angle from i to j; S33: The brightness center and brightness variance are calculated through the flame area brightness composition histogram to describe the brightness distribution of the flame. The specific formula is as follows: Where, the brightness center is the average brightness; b is the brightness value, ranging from 0 to L-1 (L is the maximum brightness, usually 255); H(b) is the number of pixels with brightness value i; S34: Calculate the height of the flame by determining the baseline and top of the flame; calculate the width of the flame by determining the widest part of the flame; calculate the morphological features based on the geometric shape of the flame in the image: in, ; in, ; Brightness standard deviation =

[0018] is the brightness value of the kth interval; when >3, it means that the brightness distribution is sharper than the normal distribution, that is, it has a higher peak and a thinner tail; <3, it means that the brightness distribution is flatter than the normal distribution, that is, it has a lower peak and a thicker tail; =3, it means that the brightness distribution is close to normal distribution; S35: Use Fourier transform analysis to evaluate the energy distribution of the flame and calculate the spectrum characteristics of the flame, such as frequency peak and spectrum width, as follows: Where (u, v) is the frequency domain coordinate, and F(u, v) is the two-dimensional Fourier transform of the image of the flame area; S36: The characteristic values ​​of flame area, flame boundary length, circularity, rectangularity, eccentricity, flame energy, flame contrast, flame uniformity, flame directionality, flame brightness center, flame brightness variance, flame height, flame width, brightness mean, brightness median, brightness maximum, brightness minimum, brightness variance, brightness standard deviation, brightness skewness, brightness kurtosis, flame frequency peak and flame frequency width are unified to the same frequency, and then normalized to a unified range to eliminate the influence of different characteristic dimensions; the normalized values ​​are obtained as follows: Where i=1,2,3,...,24, There are 24 eigenvalues; min( ) is the minimum value of the i-th eigenvalue; max ( ) is the maximum value of the i-th eigenvalue; S37: Use statistical tests and principal component analysis to select the most representative features; S38: Merge multiple representative features into a unified feature vector.

[0019] In one embodiment, the method, the step 7 feature importance assessment: covers the feature importance score calculated using the Permutation importance method, and based on the prediction accuracy of the convolutional neural network model, decision tree, random forest, and support vector machine model in step 6, the feature importance score of each model is weighted to obtain the final importance score of the feature, and the key features with the highest scores are screened out based on this, which is used as the basis for analyzing the factors affecting the quality of the spheroidized powder finished product, specifically including: Phase 1, permutation importance method: After the model training is completed, the importance of the features in the validation set is initially scored using a convolutional neural network model, decision tree, random forest or support vector machine model; then, the value of a feature in the validation set is randomly rearranged, and the model is predicted again. The importance of the feature is evaluated by comparing the change in prediction accuracy before and after the rearrangement; this method reflects the importance of the feature by the decrease in accuracy, and the score is recorded as score0; Phase 2: Weighting of model prediction accuracy: Select the best performing convolutional neural network model, decision tree, random forest, and support vector machine model on the validation set, record their prediction accuracy, and assign weights to each model based on these accuracies, calculated by dividing the sum of their respective accuracies by the total accuracy; use these weights, combined with the model’s feature importance function, to calculate the weighted importance score of each feature; Stage 3, determination of the final feature score: The score0 obtained in the above steps is combined with the feature importance scores obtained through different models to calculate the final score of each feature. This score reflects the comprehensive contribution of the feature to the model's predictive ability. Features with high scores will be considered as key factors affecting the quality of spheroidized powder.

[0020] In one embodiment, the method, step 2, flame image data preprocessing: preprocessing based on the collected original flame image data to obtain a clear flame image with a clear flame outline, specifically includes: S21: selecting a global threshold or multiple local thresholds according to the brightness characteristics of the flame and by analyzing the histogram of the image; S22: all pixels with values ​​higher than the threshold are set as foreground, and the rest are set as background; S23: manually selecting or automatically detecting one or more pixel points in the flame area as seeds, and determining which pixels can be merged into the area where the seed points are located according to the gray value, color and texture of the pixels; S24: Starting from the seed point, check the pixels around it, and if they meet the similarity criteria, merge them into the region; repeat this process until there are no more pixels to merge; S25: Select Sobel and Canny as operators, and apply the selected edge detection operators to the image to highlight the edges; S26: On the one hand, reduce the width of the edge to make it a single pixel wide line, and on the other hand, remove weak edges by setting a threshold and only keep strong edges; S27: Morphological methods are used to further smooth the flame edge and remove small noise points.

[0021] In one embodiment, the method, step S27 further smoothes the flame edge and removes small noise points by using a morphological method, specifically comprising: (1) Read the flame image; (2) Convert to grayscale image; (3) Binarize the image; (4) Define the structural element; (5) Morphological dilation; (6) Morphological erosion; (7) Opening operation: erosion first and then dilation; (8) Closing operation: dilation first and then erosion; (9) Morphological gradient: dilation image-erosion image; (10) Top hat transformation: original image-opening operation result; (11) Black hat transformation: closing operation result-original image; (12) Output result.

[0022] According to a second aspect of an embodiment of the present invention, a computer-readable storage medium is provided.

[0023] In some embodiments, the computer-readable storage medium includes a computer program for storing a computer program, wherein the computer program, when executed by a processor, implements the plasma melt-blown powder spheroidization processing method based on image recognition.

[0024] According to a third aspect of an embodiment of the present invention, a computer device is provided.

[0025] In some embodiments, the computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0026] The technical solution provided by the embodiment of the present invention may have the following beneficial effects: (1) The present invention realizes precise control of the spheroidization process by analyzing the flame shape in the plasma melt-blowing process in real time and establishing a correlation model between the flame shape and the powder spheroidization output and quality; (2) The present invention can significantly improve the automation level of the powder spheroidization process, reduce manual intervention, improve production efficiency, and ensure the consistency and reliability of product quality.

[0027] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0029] Figure 1 is a system configuration diagram according to an exemplary embodiment; Figure 2 is a flow chart showing online intelligent diagnosis and control of a spheroidization process according to an exemplary embodiment; Figure 3 The figure is a schematic diagram showing the structure of a computer device according to an exemplary embodiment. DETAILED DESCRIPTION

[0030] The following description and accompanying drawings fully illustrate the specific embodiments of this article so that those skilled in the art can practice them. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. The scope of the embodiments of this article includes the entire scope of the claims, as well as all available equivalents of the claims. Herein, the terms "first", "second", etc. are only used to distinguish one element from another, without requiring or implying any actual relationship or order between these elements. In fact, the first element can also be called the second element, and vice versa. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the structure, device or equipment including a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also include elements inherent to such structure, device or equipment. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the structure, device or equipment including the elements. Each embodiment is described in a progressive manner herein, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the embodiments can be referred to each other.

[0031] The terms "longitudinal", "lateral", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc. in this document indicate the orientation or position relationship based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing this document and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In the description of this document, unless otherwise specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a mechanical connection or an electrical connection, it can also be the internal communication of two elements, it can be a direct connection, or it can be an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0032] As used herein, the term "plurality" means two or more than two, unless otherwise specified.

[0033] In this document, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B.

[0034] In this article, the term "and / or" is a description of the association relationship between objects, indicating that three relationships may exist. For example, A and / or B means: A or B, or, A and B.

[0035] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0036] Example 1 Figure 1 An embodiment of the plasma melt-blown powder spheroidization processing method based on image recognition of the present invention is shown.

[0037] The embodiment of the present invention discloses a plasma melt-blown powder spheroidization processing method based on image recognition, comprising the following steps: Step 1: Original flame image data collection: At the beginning of the experiment, flame image data collection is carried out. In the experimental environment, high-precision camera equipment is deployed to capture dynamic images of flames in real time. During the collection process, the integrity and authenticity of the image data are ensured to provide high-quality original materials for subsequent analysis. At the same time, the collection timestamp is recorded to ensure the time sequence of the data.

[0038] Step 2: Preprocessing of flame image data: After acquisition, the original flame image is carefully preprocessed; image enhancement techniques such as contrast adjustment and noise removal are used to improve image quality and ensure the clarity of the flame outline; in addition, image areas irrelevant to the experimental objectives are eliminated to reduce the complexity of data processing. Image standardization is implemented to provide a consistent basis for feature extraction.

[0039] Step 3: Flame shape feature extraction: Based on image preprocessing, further perform flame shape feature extraction; use image analysis algorithms to identify the edge, area, color and other attributes of the flame, and quantify the morphological characteristics of the flame; at the same time, record the stability and energy distribution state of the flame to provide key data for building an estimation model.

[0040] Step 4: Dataset segmentation: To further evaluate the prediction performance of the model, the data after feature engineering is segmented. According to the ratio of 8:2, 80% of the data is allocated as the training set to train the model, and the remaining 20% ​​constitutes the validation set to evaluate the accuracy of the model.

[0041] Step 5. Analysis of the correlation between flame characteristics and spheroidized powder quality: Combining the flame shape characteristics and the quality evaluation data of spheroidized powder, using statistical methods, explore the potential relationship between different flame morphological characteristics and spheroidized powder quality; through feature selection technology, identify the key features that significantly affect the quality of spheroidized powder, laying the foundation for building a prediction model.

[0042] Step 6: Model construction and performance optimization: Using the extracted flame characteristics and spheroidized powder quality data, a convolutional neural network (CNN) model based on deep learning is constructed, and combined with machine learning algorithms such as decision trees, random forests, and support vector machines to form a comprehensive prediction model; through cross-validation and hyperparameter tuning, the model is iteratively trained until the model has the best accuracy on the validation set.

[0043] Step 7, feature importance assessment: covers the feature importance score calculated using the Permutation importance method, and weighted processing of the feature importance score of each model based on the prediction accuracy of the convolutional neural network model, decision tree, random forest, and support vector machine model in step 6. Through this process, we obtained the final importance score of the feature, and based on this, we selected the key features with the highest scores as the basis for analyzing the factors affecting the quality of the spheroidized powder product.

[0044] Step 8: Automatic adjustment of process parameters: Based on the model prediction results, an intelligent feedback control system is developed. The system can analyze the prediction output in real time and automatically adjust the key and relevant process parameters of the plasma melt-blown equipment, such as gas flow, power and other key variables. Through precise control, the flame shape is optimized, the production quality of spheroidized powder is improved, and the process flow is intelligent and automated.

[0045] In this embodiment, the method disclosed by the present invention comprises the following steps: Step 1: Original flame image data collection: Specifically, a high-resolution industrial camera is used to aim at the flame area in the plasma melt-blown equipment to capture the dynamic image of the flame in real time. The camera has high-temperature protection and stable imaging capabilities to adapt to the high-temperature environment generated by plasma. The camera is installed horizontally in front of the flame to ensure that the full view of the flame can be captured from the front.

[0046] The camera is fixed on the equipment platform with a tripod, parallel to the flame position. The camera is a 6-megapixel rolling shutter visual inspection CMOS industrial camera. The use of this camera can significantly improve the performance of the image recognition system and is suitable for various industrial automation, quality control and scientific research fields. First of all, its 6-megapixel high resolution can capture more details and improve the accuracy of image recognition; secondly, the rolling shutter can reduce motion blur, which is particularly suitable for capturing fast-moving objects and improving image quality; thirdly, CMOS sensors usually have the characteristics of low noise, high sensitivity and fast response, which helps to obtain clear images under various lighting conditions; in addition, industrial cameras are usually designed for harsh working environments, with dustproof, waterproof and vibration-resistant characteristics to ensure stable operation; finally, high-speed data transmission interfaces (such as GigE interfaces) ensure that image data can be quickly transmitted to the processing system.

[0047] Step 2: preprocessing is performed based on the collected flame image to obtain a clear flame image with a clear flame outline; Specifically, the collected original images are preprocessed by the image processing system to improve the accuracy of subsequent feature extraction. In the experiment, we used Gaussian filtering to remove noise from the image and converted the color image into a grayscale image (grayscale) to simplify the processing process. Finally, we used histogram equalization and other methods to improve the contrast of the image. In order to more clearly identify the shape of the flame later, we used three methods to process the image. First, we used threshold segmentation to quickly remove most of the background, then applied regional growth starting from the known part of the flame and gradually expanded to the entire flame area. Finally, edge detection was used to refine the outline of the flame to ensure that the edge of the flame was accurately identified.

[0048] The specific steps include: S21: selecting a global threshold or multiple local thresholds according to the brightness characteristics of the flame and by analyzing the histogram of the image; S22: all pixels with values ​​higher than the threshold are set as foreground, and the rest are set as background; S23: manually selecting or automatically detecting one or more pixel points in the flame area as seeds, and determining which pixels can be merged into the area where the seed points are located according to the gray value, color and texture of the pixels; S24: Starting from the seed point, check the pixels around it, and if they meet the similarity criteria, merge them into the region; repeat this process until there are no more pixels to merge; S25: Select Sobel and Canny as operators, and apply the selected edge detection operators to the image to highlight the edges; S26: On the one hand, reduce the width of the edge to make it a single pixel wide line. On the other hand, set a threshold to remove weak edges and keep only strong edges; S27: Morphological methods are used to further smooth the flame edge and remove small noise points; The morphological method includes: (1) reading a flame image; (2) converting it into a grayscale image (if necessary); (3) binarizing the image; (4) defining a structural element; (5) morphological dilation; (6) morphological erosion; (7) opening operation (erosion followed by dilation); (8) closing operation (dilation followed by erosion); (9) morphological gradient (dilation image-erosion image); (10) top hat transformation (original image-opening operation result); (11) black hat transformation (closing operation result-original image); and (12) outputting the result.

[0049] Step 3, extracting the characteristics of the flame shape from the clear flame image to obtain the flame stability and energy distribution state data; Specifically, the key features of the flame shape are extracted from the image generated by the image processing algorithm, such as flame height, width, morphological characteristics, brightness distribution, etc.; these features can reflect the stability and energy distribution state of the flame.

[0050] The specific steps are as follows S31: Feature calculation: Calculate the number of pixels in the segmented flame area (reflecting the flame area), calculate the length of the flame boundary (the edge complexity of the flame), and calculate multiple parameters to describe the flame shape. The details are as follows: When calculating the flame interview, first traverse each pixel of the image, and add 1 unit for each pixel that exists. The final value of the counter is the number of pixels in the flame area. Where m is the number of image rows; n is the number of image columns; I(i,j) is the pixel value at position (i,j); When calculating the boundary length of the flame area, we first select a boundary pixel of the flame area and move pixel by pixel along the edge of the flame; then for each boundary pixel, check its 8 neighbors and determine the next boundary pixel; then record the positions of all boundary pixels; finally, calculate the Euclidean distance. The detailed calculation formula is as follows: Where N is the total number of boundary pixels; ( , ) is the coordinate of the kth boundary pixel; ( , ) is the coordinate of the k+1th boundary pixel; The shape of the flame is calculated using circularity, rectangularity and eccentricity. The specific formula is as follows: Circle shape: ; Rectangularity: ; Eccentricity: ; S32: The gray level co-occurrence matrix is ​​used to calculate the texture features of the flame area, such as energy, contrast, uniformity and directionality. The specific calculation is as follows: Where L is the number of gray levels, P(i,j) is the frequency of co-occurrence of gray levels i and j in the gray level co-occurrence matrix; Where |ij| represents the difference between gray levels i and j, and P(i,j) is the frequency of co-occurrence of gray levels i and j; Directivity D(θ)=

[0051] in, is at an angle The frequency of gray levels i and j appearing together; is the change in angle from i to j; S33: The brightness center and brightness variance are calculated through the flame area brightness composition histogram to describe the brightness distribution of the flame. The specific formula is as follows: Where, the brightness center is the average brightness; b is the brightness value, ranging from 0 to L-1 (L is the maximum brightness, usually 255); H(b) is the number of pixels with brightness value i; S34: Calculate the height of the flame by determining the baseline and top of the flame; determine the widest part of the flame and calculate the width of the flame; calculate the morphological features based on the geometric shape of the flame in the image, such as: calculate the brightness mean, median, maximum and minimum values ​​of the flame area, which is used to analyze the uniformity or central tendency of the brightness distribution: in, ; in, ; Brightness standard deviation =

[0052] is the brightness value of the kth interval; when >3, it means that the brightness distribution is sharper than the normal distribution, that is, it has a higher peak and a thinner tail; <3, it means that the brightness distribution is flatter than the normal distribution, that is, it has a lower peak and a thicker tail; =3, it means that the brightness distribution is close to the normal distribution.

[0053] S35: Use Fourier transform analysis method to evaluate the energy distribution of the flame and calculate the spectrum characteristics of the flame, such as frequency peak and spectrum width. The detailed formula is as follows: Where (u, v) is the frequency domain coordinate, and F(u, v) is the two-dimensional Fourier transform of the image of the flame area; S36: Unify 24 characteristic values ​​including flame area, flame boundary length, circularity, rectangularity, eccentricity, flame energy, flame contrast, flame uniformity, flame directionality, flame brightness center, flame brightness variance, flame height, flame width, brightness mean, brightness median, brightness maximum, brightness minimum, brightness variance, brightness standard deviation, brightness skewness, brightness kurtosis, flame frequency peak and flame frequency width to the same frequency, and then normalize them to a unified range (such as 0 to 1) to eliminate the influence of different characteristic dimensions; the normalized values ​​are as follows: where i = 1,2,3,...,24, There are 24 eigenvalues; min( ) is the minimum value of the i-th eigenvalue; max ( ) is the maximum value of the i-th eigenvalue; S37: Use statistical tests and principal component analysis (PCA) methods to select the most representative features; The pseudo code is as follows: # Pseudocode: Select features using principal component analysis #Input: Normalized feature matrix X', where each row represents a sample and each column represents a feature #Output: selected feature subset X_selected, principal component coefficient matrix principal_components #Step 1: Calculate the mean vector mu of the feature matrix mu = calculate mean(X') #Step 2: Center the feature matrix and subtract the mean vector X_centered=X'-mu #Step 3: Calculate the covariance matrix covariance_matrix = X_centered transpose * X_centered / (number of rows of X' - 1) #Step 4: Calculate the eigenvalues ​​and eigenvectors of the covariance matrix Eigenvalues, Eigenvectors = Calculate eigenvalues ​​and eigenvectors (covariance matrix) #Step 5: Sort by eigenvalue in descending order and select the first k eigenvectors as principal components sorted_feature_values ​​= Sort descending (feature values) principal_components = Select the first k (eigenvectors) #Step 6: Construct the transformation matrix W to transform the original features into the principal component space W = construction matrix (principal_components) #Step 7: Convert feature matrix to principal component space X_pca=W*X_centered #Step 8: Select features corresponding to the main components X_selected = Select Features(X', Principal Component Index) #Step 9: Return the selected features and principal component coefficient matrix Returns X_selected, W S38: Multiple representative features are merged into a unified feature vector to prepare for the subsequent input into the prediction model; Step 4: Dataset segmentation: To further evaluate the prediction performance of the model, the data after feature engineering is segmented. According to the ratio of 8:2, 80% of the data is allocated as the training set to train the model, and the remaining 20% ​​constitutes the validation set to evaluate the accuracy of the model. Step 5. Analysis of the correlation between flame characteristics and spheroidized powder quality: Combining the flame shape characteristics and the quality evaluation data of spheroidized powder, using statistical methods, explore the potential relationship between different flame morphological characteristics and spheroidized powder quality; through feature selection technology, identify the key features that significantly affect the quality of spheroidized powder, laying the foundation for building a prediction model.

[0054] Specifically, according to the data set of the present invention, the flame features have a total of 24 eigenvalues, and the spheroidized powder quality has a total of 3 categories (good quality, average quality, and poor quality). Each time series data of each input eigenvalue will correspond to a type of spheroidized powder quality. Combined with the training set and the validation set, through multiple iterative training, the model with the best performance is finally obtained so that the accuracy of the model on the validation set is the best.

[0055] Step 6: Model construction and performance optimization: Using the extracted flame characteristics and spheroidized powder quality data, a convolutional neural network (CNN) model based on deep learning is constructed, and combined with machine learning algorithms such as decision trees, random forests, and support vector machines to form a comprehensive prediction model; through cross-validation and hyperparameter tuning, the model is iteratively trained until the model has the best accuracy on the validation set: Specifically, based on the extracted flame characteristics and historical production quality data, the CNN model is first used to predict the correlation between different flame characteristic shapes and the quality of spheroidized powder, and then the traditional machine learning model is used for prediction, and finally the one with the highest prediction accuracy is selected; this model can learn the nonlinear relationship between flame shape and spheroidized powder quality.

[0056] Among them, the convolutional neural network model (CNN) model: 1. The neural network starts with a pair of Conv2D layers, each with 32 filters of size 3x3, followed by a batch normalization layer, which speeds up training and provides a degree of regularization that helps prevent overfitting.

[0057] 2. The Conv2D layer pair is followed by a MaxPooling2D layer, which reduces the spatial dimensions (height and width), effectively providing a translation invariance and reducing computational complexity, followed by a Dropout layer that randomly sets a portion of the input units (0.2 for the first dropout layer) to 0 at each update during training, helping to prevent overfitting.

[0058] 3. This pattern consisting of two Conv2D layers, a batch normalization layer, a MaxPooling2D layer, and a dropout layer is repeated three more times. The number of filters in the Conv2D layer doubles at each repetition, starting from 32, increasing to 64, 128, and then 256. This increasing pattern helps the network learn more complex features at each level. The dropout rate also increases from 0.2 to 0.5 in each step.

[0059] After the convolution and pooling layers, a flatten layer is used to convert the 2D output of the previous layer into a 1D vector.

[0060] 4. Finally, a dense (or fully connected) layer is used for classification. It has 10 units, each unit represents one of the 10 categories in the CIFAR-10 dataset, and a softmax activation function is used to convert the output into a probability score for each category.

[0061] This architecture takes advantage of deep CNNs to learn hierarchical features from CIFAR-10 images, and also uses regularization techniques such as L2 regularization, dropout, and batch normalization to combat overfitting; although inspired by VGG16, the model remains simple and does not adopt more advanced architectural features, but instead focuses on efficiency and simplicity.

[0062] Based on the obtained real-time collected original flame image, the model parameters and the established estimation model are fitted after preprocessing and feature extraction to predict the powder yield of the spheroidization process; Specifically, the flame images collected in real time are processed and feature extracted, and then input into the trained machine learning model to predict the yield of the current spheroidization process, and the prediction efficiency of the model is evaluated using accuracy, precision, recall, F1 score, mean square error, and cross entropy loss.

[0063] The commonly used performance evaluation indicators of different models are: Accuracy - the proportion of correctly predicted samples to the total number of samples; Precision - the proportion of truly positive samples among all predicted positive samples; Recall - the proportion of correctly predicted positive samples among all truly positive samples; F1 score - the harmonic mean of precision and recall, which is a balance indicator between the two; Mean Square Error (MSE) - the average of the squares of the difference between the predicted value and the actual value; Cross-Entropy Loss - measures the difference between the model's predicted probability distribution and the actual distribution.

[0064] When comparing different models, K-fold cross validation was used to evaluate the performance of different models and reduce the risk of overfitting. Then the above evaluation indicators were calculated on the training set and test set to compare the performance of different models.

[0065] Step 7: Feature Importance Assessment: Use the permutation importance method and the built-in feature importance assessment tool of the model to score the importance of the features, so as to identify the features that have the greatest impact on the quality of the spheroidized powder. This step mainly has three stages: Phase 1, permutation importance method: After the model training is completed, use models such as CNN, decision tree, random forest or support vector machine to make a preliminary score on the importance of features in the validation set. Then, randomly rearrange the value of a feature in the validation set, and perform model prediction again. By comparing the change in prediction accuracy before and after the rearrangement, the importance of the feature is evaluated. This method reflects the importance of the feature by the decrease in accuracy, and the score is recorded as score0.

[0066] Phase 2: Weighting of model prediction accuracy: Select the best performing convolutional neural network model, decision tree, random forest, and support vector machine model on the validation set, and record their prediction accuracy (acc1, acc2, acc3, acc4 respectively). Based on these accuracy rates, assign weights (weight1, weight2, weight3, weight4) to each model, calculated by dividing the sum of their respective accuracy rates by the total accuracy rate. Using these weights, combined with the model's feature importance function (such as feature_importances_), calculate the weighted importance score of each feature, recorded as score1, score2, score3, score4 respectively.

[0067] Stage 3, determination of the final feature score: combine the score0 obtained in the above steps with the feature importance scores (score1, score2, score3, score4) obtained through different models to calculate the final score of each feature. This score reflects the comprehensive contribution of the feature to the model's predictive ability. Features with high scores will be considered key factors affecting the quality of spheroidized powder.

[0068] Step 8: Automatic adjustment of process parameters: Based on the model prediction results, develop an intelligent feedback control system. The system can analyze the prediction output in real time and automatically adjust the key and relevant process parameters of the plasma melt-blown equipment, such as gas flow, power and other key variables. Through precise control, the flame shape is optimized, the production quality of spheroidized powder is improved, and the process flow is intelligent and automated.

[0069] The present invention uses machine learning and image recognition technology to define the numerical range of each indicator when the output and the qualified rate are optimal; the output, qualified rate and safety factor are predicted by each numerical value; under supervised learning and unsupervised learning, an algorithm for automatically selecting characteristic values ​​is used to link multiple characteristic values ​​and the output with the qualified rate.

[0070] Example 2 Figure 2 An embodiment of the plasma melt-blown powder spheroidization processing device based on image recognition of the present invention is shown.

[0071] In this optional embodiment, the plasma melt-blown powder spheroidization processing device based on image recognition includes: a circulating water system, a powder feeder 1, a cyclone separator 2, a vacuum pump 3, a powder drop bin 4, a collector 5, an air compressor 6 and a host 7, a valve is adaptively arranged on the pipeline, and an industrial camera for capturing a dynamic image of the flame in real time is arranged in the powder drop bin 4; The specific steps include: Step 1, equipment inspection and preparation: Check whether all equipment is working properly, including valves, circulating water system, powder feeder, cyclone separator, vacuum pump, air compressor, etc.; at the same time, confirm that the power connection is correct, the host power supply is 380V, and the vacuum pump and air compressor are 220V; finally, confirm whether the on-site public auxiliaries are complete. After confirmation, turn on the cooling water.

[0072] Step 2: Put the metal powder into the powder feeder 1 and close the powder drop bin 4 furnace body.

[0073] Step 3: Open the valve of vacuum pump 3, close the Ar gas valve, the valves of collector 5 and powder feeder 1, start the vacuum system, and evacuate the air at room temperature. When the pressure in the furnace drops below 10 Pa, maintain it for 5 minutes.

[0074] Step 4: Stop the vacuum system, open the Ar gas valve, close the vacuum pump 3 valve, and introduce Ar gas into the powder drop bin 4 furnace to make the pressure in the powder drop bin 4 furnace reach about +2000Pa. Stop introducing Ar gas and let it stand for 5 minutes.

[0075] Step 5: Open the valve of powder feeder 1 and the valve of collector 5 and start the machine.

[0076] Step six, adjust the Ar gas flow rate and ignite. During the process, monitor the pressure in the furnace at all times and release the pressure in time if it is too high.

[0077] Step seven, start the powder feeder 1 and start plasma spheroidization, the process takes about 1 hour.

[0078] Step 8. After the spheroidization is completed, turn off the powder feeder 1 and its airflow, turn off the plasma power supply (to prevent the liquid in the collector from being sucked back during the process), and then stop introducing Ar gas.

[0079] Step nine, after cooling to room temperature, turn off the cooling water of the powder dropping bin 4 furnace body, disassemble the powder dropping bin 4 furnace body and collector 5, collect the spheroidized product, and store the product in a sealed bag.

[0080] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in the above method embodiment are implemented.

[0081] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0082] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiment when executing the computer program.

[0083] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiment are implemented.

[0084] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0085] The present invention is not limited to the structures which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A plasma melt-blown powder spheroidization processing method based on image recognition, characterized in that: The following steps are involved: S1: Equipment inspection and preparation: Determine the process parameters required for plasma melt-blown powder spheroidization and turn on the cooling water; S2: Put the metal powder into the powder feeder and close the powder bin; S3: Open the vacuum pump valve, close the Ar gas valve, collector and powder feeder valve, start the vacuum system, and evacuate at room temperature. When the air pressure in the powder drop bin drops below 10Pa, maintain it; S4: Stop the vacuum system, open the Ar gas valve, close the vacuum pump valve, and introduce Ar gas into the powder dropping bin to make the pressure in the powder dropping bin reach +2000Pa. Stop introducing Ar gas and let it stand; S5: Open the powder feeder valve and collector valve; S6: Adjust the Ar gas flow rate and ignite, and monitor the pressure in the powder drop chamber; S7: Turn on the powder feeder and start plasma spheroidization; S8: Online intelligent diagnosis and control of the spheroidization process; S9: After the spheroidization is completed, the powder feeder and its airflow are turned off, the plasma power supply is turned off, and then the Ar gas is stopped; S10: After cooling to room temperature, turn off the cooling water in the powder dropping bin, disassemble the powder dropping bin and collector, collect the spheroidized product, and store the product in a sealed bag.

2. According to the plasma melt-blown powder spheroidization processing method based on image recognition according to claim 1, it is characterized in that: The step S8 of performing online intelligent diagnosis and control on the spheroidization process specifically includes: Step 1: original flame image data acquisition; Step 2: flame image data preprocessing: preprocessing is performed based on the collected original flame image data to obtain a clear flame image with a clear flame outline; Step 3: Flame shape feature extraction: Use image analysis algorithms to identify the edge, area, and color of the flame and quantify the morphological characteristics of the flame; at the same time, record the stability and energy distribution of the flame; Step 4: Dataset segmentation: The data obtained in step 3 after feature engineering of the flame shape features is segmented. According to the ratio of 8:2, 80% of the data is allocated as a training set to train the model, and the remaining 20% ​​constitutes a validation set. Step 5: Analysis of the correlation between flame characteristics and spheroidized powder quality: Combining the flame shape characteristics and the quality evaluation data of spheroidized powder, using statistical methods, predict the correlation between different flame morphological characteristics and spheroidized powder quality; using feature selection technology, identify key features that significantly affect the quality of spheroidized powder; Step 6: Model construction and performance optimization: Using the extracted flame characteristics and spheroidized powder quality data, a convolutional neural network model based on deep learning is constructed, and combined with machine learning algorithms, including decision trees, random forests, and support vector machines, to form a comprehensive prediction model; through cross-validation and hyperparameter tuning, the model is iteratively trained until the model has the best accuracy on the validation set; Step 7, feature importance assessment: The feature importance score calculated by the Permutation importance method, and the feature importance score of each model is weighted according to the prediction accuracy of the convolutional neural network model, decision tree, random forest, and support vector machine model in step 6 to obtain the final importance score of the feature, and the key features with the highest scores are selected based on this, which is used as the basis for analyzing the factors affecting the quality of the spheroidized powder finished product; Step 8: Automatic adjustment of process parameters: Based on the model prediction results, develop an intelligent feedback control system, analyze the prediction output in real time, and automatically adjust the key process parameters of the plasma melt-blowing equipment. Through precise control, optimize the flame shape and improve the production quality of spheroidized powder.

3. According to the plasma melt-blown powder spheroidization processing method based on image recognition according to claim 2, it is characterized in that: The third step of flame shape feature extraction is to use image analysis algorithm to identify the edge, area, and color of the flame and quantify the morphological characteristics of the flame; at the same time, record the stability and energy distribution state of the flame, which specifically includes: S31: Feature calculation: Calculate the number of pixels in the segmented flame area, calculate the length of the flame boundary, and calculate multiple parameters to describe the flame shape, as follows: When calculating the flame interview, first traverse each pixel of the image, and add 1 unit for each pixel that exists. The final value of the counter is the number of pixels in the flame area. Where m is the number of image rows; n is the number of image columns; I(i,j) is the pixel value at position (i,j); When calculating the boundary length of the flame area, first select a boundary pixel of the flame area and move pixel by pixel along the edge of the flame; then for each boundary pixel, check its 8 neighbors and determine the next boundary pixel; then record the positions of all boundary pixels; finally, calculate the Euclidean distance, which is calculated as follows: Where N is the total number of boundary pixels; ( , ) is the coordinate of the kth boundary pixel; ( , ) is the coordinate of the k+1th boundary pixel; The shape of the flame is calculated using circularity, rectangularity and eccentricity. The specific formula is as follows: Circle shape: ; Rectangularity: ; Eccentricity: ; S32: The gray level co-occurrence matrix is ​​used to calculate the texture features of the flame area, such as energy, contrast, uniformity and directionality. The calculation formula is as follows: Where L is the number of gray levels, P(i,j) is the frequency of co-occurrence of gray levels i and j in the gray level co-occurrence matrix; Where |ij| represents the difference between gray levels i and j, and P(i,j) is the frequency of co-occurrence of gray levels i and j; Directivity D(θ)= ; in, is at an angle The frequency of gray levels i and j appearing together; is the change in angle from i to j; S33: The brightness center and brightness variance are calculated through the flame area brightness composition histogram to describe the brightness distribution of the flame. The specific formula is as follows: Where, the brightness center is the average brightness; b is the brightness value, ranging from 0 to L-1 (L is the maximum brightness, usually 255); H(b) is the number of pixels with brightness value i; S34: Calculate the height of the flame by determining the baseline and top of the flame; calculate the width of the flame by determining the widest part of the flame; calculate the morphological features based on the geometric shape of the flame in the image: in, ; in, ; Brightness standard deviation = ; is the brightness value of the kth interval; when >3, it means that the brightness distribution is sharper than the normal distribution; when <3, it means that the brightness distribution is flatter than the normal distribution; when =3, indicating that the brightness distribution is close to normal distribution; S35: Use Fourier transform analysis to evaluate the energy distribution of the flame and calculate the spectrum characteristics of the flame, such as frequency peak and spectrum width, as follows: Where (u, v) is the frequency domain coordinate, and F(u, v) is the two-dimensional Fourier transform of the image of the flame area; S36: The characteristic values ​​of flame area, flame boundary length, circularity, rectangularity, eccentricity, flame energy, flame contrast, flame uniformity, flame directionality, flame brightness center, flame brightness variance, flame height, flame width, brightness mean, brightness median, brightness maximum, brightness minimum, brightness variance, brightness standard deviation, brightness skewness, brightness kurtosis, flame frequency peak and flame frequency width are unified to the same frequency, and then normalized to a unified range to eliminate the influence of different characteristic dimensions; the normalized values ​​are obtained as follows: Where i=1,2,3,...,24, There are 24 eigenvalues; min( ) is the minimum value of the i-th eigenvalue; max ( ) is the maximum value of the i-th eigenvalue; S37: Use statistical tests and principal component analysis to select the most representative features; S38: Merge multiple representative features into a unified feature vector.

4. According to the plasma melt-blown powder spheroidization processing method based on image recognition according to claim 2, it is characterized in that: The seventh step is feature importance assessment: the feature importance score calculated by the permutation importance method, and the feature importance score of each model is weighted according to the prediction accuracy of the convolutional neural network model, decision tree, random forest, and support vector machine model in step six, to obtain the final importance score of the feature, and the key features with the highest score are selected based on this, which is used as the basis for analyzing the factors affecting the quality of the spheroidized powder finished product, specifically including: Phase 1, permutation importance method: After the model training is completed, the importance of the features in the validation set is initially scored using a convolutional neural network model, decision tree, random forest or support vector machine model; then, the value of a feature in the validation set is randomly rearranged, and the model is predicted again. The importance of the feature is evaluated by comparing the change in prediction accuracy before and after the rearrangement; this method reflects the importance of the feature by the decrease in accuracy, and the score is recorded as score0; Phase 2: Weighting of model prediction accuracy: Select the best performing convolutional neural network model, decision tree, random forest, and support vector machine model on the validation set, record their prediction accuracy, and assign weights to each model based on these accuracies, calculated by dividing the sum of their respective accuracies by the total accuracy; use these weights, combined with the model’s feature importance function, to calculate the weighted importance score of each feature; Stage 3, determination of the final feature score: The score0 obtained in the above steps is combined with the feature importance scores obtained through different models to calculate the final score of each feature. This score reflects the comprehensive contribution of the feature to the model's predictive ability. Features with high scores will be considered as key factors affecting the quality of spheroidized powder.

5. According to the plasma melt-blown powder spheroidization processing method based on image recognition according to claim 2, it is characterized in that: The second step, flame image data preprocessing: preprocessing is performed based on the collected original flame image data to obtain a clear flame image with a clear flame outline, specifically including: S21: selecting a global threshold or multiple local thresholds according to the brightness characteristics of the flame and by analyzing the histogram of the image; S22: all pixels with values ​​higher than the threshold are set as foreground, and the rest are set as background; S23: manually selecting or automatically detecting one or more pixel points in the flame area as seeds, and determining which pixels can be merged into the area where the seed points are located according to the gray value, color and texture of the pixels; S24: Starting from the seed point, check the pixels around it, and if they meet the similarity criteria, merge them into the region; repeat this process until there are no more pixels to merge; S25: Select Sobel and Canny as operators, and apply the selected edge detection operators to the image to highlight the edges; S26: On the one hand, reduce the width of the edge to make it a single pixel wide line, and on the other hand, remove weak edges by setting a threshold and only keep strong edges; S27: Morphological methods are used to further smooth the flame edge and remove small noise points.

6. The plasma melt-blown powder spheroidization processing method based on image recognition according to claim 5 is characterized in that: The step S27 further smoothes the flame edge and removes small noise points by using a morphological method, specifically comprising: (1) Read the flame image; (2) Convert to grayscale image; (3) Binarize the image; (4) Define the structural element; (5) Morphological dilation; (6) Morphological erosion; (7) Opening operation: erosion first and then dilation; (8) Closing operation: dilation first and then erosion; (9) Morphological gradient: dilation image-erosion image; (10) Top hat transformation: original image-opening operation result; (11) Black hat transformation: closing operation result-original image; (12) Output result.

7. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, it implements the plasma melt-blown powder spheroidization processing method based on image recognition as described in any one of claims 2 to 6.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 2 to 6 are implemented.

Images