Graphite column synthetic block quality detection method and system based on image processing

Abstract

The invention discloses a quality detection method and system for a graphite column synthetic block based on image processing. The method comprises the following steps: acquiring section images of the graphite column synthetic block under various production parameters for constructing a training set; the section image comprises a section complete image; the production parameters comprise equipment parameters, environment parameters and material parameters; the equipment parameters further comprise power parameters and pressure parameters; manually marking the crystal distribution grade and the crystal form grade in the cross-section image of the graphite column synthesis block; taking the marked cross-section image of the graphite column synthesis block as an input training model so as to obtain a quality detection model; and acquiring a to-be-detected image of the section of the graphite column synthetic block from a production line in real time, inputting the to-be-detected image into the quality detection model so as to obtain the crystal distribution grade and the crystal form grade of the graphite column synthetic block, and matching corresponding production parameters. The method is used for real-time automatic detection of diamond crystal distribution and crystal form grades in the production process, and a producer can conveniently and timely adjust parameters according to the grades.

Description

Technical Field

[0001] The present invention relates to the technical field of diamond quality detection, and in particular to a quality detection method and system for a graphite column synthetic block based on image processing. Background Art

[0002] Industrial diamond presses operate under high temperature and high pressure. In the process of synthesizing diamonds, precise control of equipment pressure and power is a key factor in ensuring the quality of diamond synthesis. Traditional methods mainly adjust pressure and power based on the operator's personal experience. However, the complexity and variability of the production environment and the differences between equipment make the original standard adjustment rules no longer applicable. Manual judgment is not only easily affected by workers' subjective factors, but also fails to consider the performance differences between equipment, resulting in fluctuations in the quality of the synthesized blocks.

[0003] In order to be able to adjust the parameters in the production environment in real time to obtain high-quality diamond synthetic blocks, that is, graphite column synthetic blocks, which have multiple diamond crystals, it is necessary to detect the quality grade of the diamond synthetic blocks currently produced. For the diamond synthetic blocks in the production process, their quality needs to be described from two dimensions: crystal distribution and crystal form. The crystal distribution is divided into 7 levels, and the crystal form is divided into 12 levels. During the production process, the operator adjusts the pressure and power parameters of the press according to the quality grade of the synthetic block produced by the previous block, so that the next synthetic block can obtain the optimal crystal distribution state and shape.

[0004] Prior art Chinese patent application CN116883394A discloses a diamond quality detection method based on image data processing, including: image acquisition stage, preprocessing stage, detection extraction stage, calculation processing stage, model training stage, classification stage, output display stage. Diamond quality detection is performed through visual technology and artificial intelligence models, thereby improving detection efficiency and consistency.

[0005] However, the above detection method is a finished product quality detection method, which is mainly used to detect the quality grade of independent diamond crystals after they come off the production line. It is not suitable for the production process, so that the operator cannot obtain the quality grade of the graphite column synthetic block with multiple diamonds, and thus cannot make targeted adjustments to the synthetic press equipment. Summary of the invention

[0006] The object of the present invention is to provide a quality detection method and system for graphite column synthetic blocks based on image processing, which partially solves or alleviates the above-mentioned deficiencies in the prior art and is used for real-time detection of diamond crystal distribution and crystal grade in graphite column synthetic blocks during the production process, so as to facilitate producers to adjust equipment parameters in time according to the grade.

[0007] In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions: a quality detection method of a graphite column synthetic block based on image processing, comprising:

[0008] Collecting cross-sectional images of graphite column synthetic blocks under various production parameters for constructing a training set; the cross-sectional images include complete cross-sectional images; the production parameters include equipment parameters, environmental parameters, and material parameters; the equipment parameters include power parameters and pressure parameters;

[0009] Annotate the crystal distribution grade and crystal form grade in the cross-section image of the graphite column synthetic block;

[0010] The annotated graphite column synthetic block cross-section image is input into the model for training, thereby obtaining a quality inspection model;

[0011] The image to be inspected of the cross section of the graphite column synthetic block is collected in real time from the production line and input into the quality inspection model to obtain the crystal distribution grade and crystal type grade of the graphite column synthetic block.

[0012] As an improvement, when constructing the training set, background images of graphite column synthetic blocks are collected and annotated as background as negative samples during annotation.

[0013] As an improvement, the model is based on the fasterrcnn_resnet50_fpn model, and its classifier is FastRCNNPredictor; the output layer of the FastRCNNPredictor classifier is configured into 20 categories, including 12 crystal form levels, 7 crystal distribution levels and 1 background.

[0014] As an improvement, after obtaining the crystal distribution grade and the crystal form grade of the graphite column synthetic block, if the crystal distribution grade is lower than the threshold grade, the crystal form grade is re-inspected in different zones.

[0015] As an improvement, the step of the partition re-inspection specifically includes:

[0016] Partitioning the acquired complete cross-sectional image to obtain a partitioned image;

[0017] Performing image enhancement processing on the partition image;

[0018] The enhanced partitioned image is input into the pre-trained crystal quality detection sub-model to obtain the crystal grade after re-inspection.

[0019] As an improvement, the step of partitioning the acquired complete cross-sectional image specifically includes:

[0020] Perform instance segmentation on the crystals in the complete cross-sectional image to obtain the position of each crystal in the complete cross-sectional image;

[0021] The image within the threshold range around each crystal was framed and cropped to obtain a partitioned image.

[0022] As an improvement, the threshold range is a range surrounded by points whose distance from the crystal edge is the threshold distance.

[0023] As an improvement, when an image within a certain crystal threshold range contains another crystal, the distance between the closest points of the two crystals is calculated, and half of the distance is used as the threshold distance for frame selection.

[0024] As an improvement, the steps of image enhancement specifically include:

[0025] Resize partition images to the same resolution;

[0026] Convert the partition image to grayscale;

[0027] Increase the number of samples by randomly rotating, mirroring, and scaling.

[0028] The present invention also provides a quality detection system for graphite column synthesis blocks based on image processing, comprising:

[0029] A training set construction module is used to collect cross-sectional images of graphite column synthesis blocks under various production parameters for constructing a training set; the cross-sectional images include complete cross-sectional images; the production parameters include equipment parameters, environmental parameters, and material parameters; the equipment parameters further include power parameters and pressure parameters;

[0030] Annotation module, used to annotate the crystal distribution grade and crystal form grade in the cross-section image of the graphite column synthesis block;

[0031] A model building module is used to input the annotated graphite column synthetic block cross-section image into the model for training, thereby obtaining a quality inspection model;

[0032] The detection module is used to collect the image to be detected of the cross section of the graphite column synthesis block from the production line in real time and input it into the quality detection model to obtain the crystal distribution grade and crystal type grade of the graphite column synthesis block.

[0033] The present invention is beneficial in that:

[0034] In order to improve the accuracy and adaptability of model detection, the present invention covers as many graphite column synthesis blocks generated under various production parameters as possible when constructing the training set. In addition, the generation of graphite column synthesis blocks under various parameter environments can also provide better reference and guidance for the adjustment of subsequent production parameters, avoiding the situation where parameters are missing at a certain graphite column synthesis block level.

[0035] In order to eliminate the influence of graphite reflection on the prediction results, the present invention also adds background images as negative samples when constructing the training set, and annotates the background during annotation.

[0036] When building the model, the fasterrcnn_resnet50_fpn model was selected as the basic architecture, and 20 output layers were configured to match the quality inspection in the production process, namely 12 crystal form grades, 7 crystal distribution grades and 1 background. This allows the model to detect not only the crystal form grade, but also the crystal distribution grade.

[0037] In order to improve the accuracy of detection and eliminate the influence of dense crystals on crystal form detection. In the present invention, images with crystal distribution levels lower than the threshold level are re-inspected in different regions. Each crystal is framed and cropped from the image and then input into the sub-model for detection, thereby improving the accuracy of detection. For crystal distribution above the threshold level, the main model can be directly used for prediction. On the one hand, in order to increase the speed of the entire system and reduce the computing power requirements, on the other hand, the credibility of the detection is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without paying creative labor.

[0039] Figure 1 is a flow chart of the present invention;

[0040] Figure 2 It is a structural diagram of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0042] Herein, suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention, and have no specific meanings by themselves. Therefore, "module", "component" or "unit" can be used mixedly.

[0043] In this document, the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0044] In this document, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0045] Herein "and / or" includes any and all combinations of one or more of the associated listed items.

[0046] Herein, "plurality" means two or more than two, ie, it includes two, three, four, five, etc.

[0047] Embodiment 1

[0048] In order to provide guidance on equipment parameter adjustment during production, such as Figure 1 As shown, the present invention provides a quality detection method for a graphite column synthetic block based on image processing, comprising:

[0049] S1 collects cross-sectional images of graphite column synthetic blocks under various production parameters for constructing a training set; the cross-sectional images include complete cross-sectional images; the production parameters include equipment parameters, environmental parameters, and material parameters; the equipment parameters further include power parameters and pressure parameters.

[0050] In order to better collect the cross-sectional images of the graphite column synthetic block, a high-resolution camera or microscope (on which a smart mobile terminal, such as a mobile phone, is mounted through a mounting bracket to collect images) can be deployed in an industrial environment to collect the cross-sectional images of the graphite column synthetic block with high precision. Of course, in order to improve the robustness of the later model, the images can be collected at multiple angles and distances.

[0051] In addition, when constructing the training set, graphite column synthetic blocks generated under various production parameters are covered as much as possible. In this embodiment, the so-called various production parameters include equipment parameters, environmental parameters and material parameters.

[0052] The most important parameters of the equipment are pressure and power parameters, so it is necessary to cover as much as possible the images of the graphite column synthetic blocks produced under all the pressures and powers that the equipment can achieve.

[0053] Environmental parameters mainly include the temperature and humidity of the production environment and other parameters that can affect the generation of graphite column synthetic blocks. Of course, under permitted conditions, images of graphite column synthetic blocks generated under various environmental temperatures and humidities are collected as much as possible.

[0054] Material parameters mainly include the composition of the material. Raw materials may also fluctuate in content and other aspects, so it is necessary to collect images of graphite column synthetic blocks generated within the material fluctuation range.

[0055] In an ideal situation, all images of graphite column composite blocks generated under various possible arrangements of equipment parameters, environmental parameters, and material parameters need to be collected. Of course, in actual production, only as many images as possible can be collected.

[0056] S2 marks the crystal distribution grade and crystal type grade in the cross-sectional image of the graphite column synthetic block.

[0057] In this embodiment, professionals annotate the crystal distribution level and crystal type level in the cross-sectional image of the graphite column synthesis block to ensure the accuracy and consistency of the annotation as much as possible. Of course, in other embodiments, pre-trained annotation tools can also be used for machine annotation.

[0058] Because some backgrounds in the graphite column synthetic block are easily mistaken for crystals under light, in order to eliminate misjudgment, in this embodiment, when constructing the training set, background images of the graphite column synthetic block are collected and annotated as background as negative samples.

[0059] S3 inputs the annotated graphite column synthetic block cross-section image into the model for training, thereby obtaining a quality inspection model.

[0060] In this embodiment, before the image is input into the model, the image may be enhanced, specifically including:

[0061] 1. Resizing: Resize all images to a uniform resolution to ensure consistency of input data.

[0062] 2. Color conversion: Convert images to grayscale format to reduce the complexity of model training and speed up calculations.

[0063] 3. Sample expansion: Apply techniques such as random rotation, flipping, and scaling to enhance the model’s robustness to image deformations.

[0064] In this embodiment, the image classification model selected is based on fasterrcnn_resnet50_fpn, which combines ResNet-50 and Feature Graph Network (FPN) and is suitable for complex target detection tasks. ResNet-50 provides effective feature extraction capabilities, while FPN enhances the model's detection performance at multiple scales.

[0065] In order to cater to the quality inspection in the production process, that is, to perform grade quality inspection on both crystal distribution and crystal form, the classification head of the model is redefined for the above model in this embodiment. In fasterrcnn_resnet50_fpn, the original classifier is replaced with a new FastRCNNPredictor, whose output layer is configured to recognize 20 categories, namely 12 crystal form grades, 7 crystal distribution grades, and 1 background.

[0066] Specifically, the structure of the model is as follows:

[0067] Convolutional layer: A total of 50 convolutional networks, each layer contains multiple residual blocks to enhance learning ability and increase training depth.

[0068] Feature Pyramid Network (FPN): The structure contains a top-down path and lateral connections, which optimizes the fusion of multi-scale features.

[0069] ROI head: includes the region proposal network (RPN) and ROI alignment layer, followed by customized classification and regression layers.

[0070] Classifier: Based on the original Fast R-CNN predictor, it is expanded to 20 outputs (12 polymorph classes, 7 crystal distribution classes, and 1 background).

[0071] Activation function: Use the ReLU activation function for nonlinear transformation to ensure stability as the network depth increases.

[0072] The training process of the model in this implementation is as follows:

[0073] a. The model is deployed on appropriate hardware (GPU is preferred to accelerate computation). Images and objects (including bounding boxes and category labels) are loaded onto the same device for processing;

[0074] b. During the training process, batch processing is implemented. Each batch of images is forward propagated through the model, the loss between the predicted results and the actual labels is calculated, and then the weights are updated through backpropagation;

[0075] c. Use SGD optimizer for parameter optimization, set the initial learning rate, momentum and weight decay, and adjust the learning rate regularly to optimize the training process;

[0076] d. The loss function uses cross entropy loss, which is particularly suitable for multi-class classification problems.

[0077] e. Training parameters:

[0078] Training cycle: The model was optimized for 200 training cycles, and each cycle traversed the entire training set; Optimizer: The SGD optimizer was used, the initial learning rate was set to 0.005, and the learning rate was dynamically adjusted to optimize the convergence speed; Batch size: Each batch of data contained 16 images, which balanced the training speed and memory usage efficiency;

[0079] The deployment process of the model in this embodiment is as follows:

[0080] a. Performance evaluation:

[0081] During training, continuously monitor the model's loss and classification accuracy on the training set and validation set, and use indicators such as precision, recall, and F1 score to evaluate model performance;

[0082] b. Model optimization:

[0083] Adjust model parameters and training strategies based on performance feedback, such as increasing or decreasing the number of network layers, adjusting batch size or optimizer parameters when necessary;

[0084] Adjust model parameters and training strategies based on test results, such as modifying learning rate scheduling or adding data augmentation methods;

[0085] c. Actual deployment:

[0086] Deploy trained models in production environments to process images from production line cameras in real time.

[0087] Set up regular model maintenance and update cycles to accommodate changes that may occur during production.

[0088] The actual effects of this model are as follows:

[0089] a.Accuracy;

[0090] Overall Accuracy: The model achieved 93.5% accuracy on the independent test set.

[0091] Crystalline grade accuracy: The average accuracy was 95%, with the best category accuracy reaching 98% and the lowest at 92%.

[0092] Distribution level accuracy: The average accuracy is 90%, with the highest being 94% and the lowest being 85%.

[0093] b. Recall rate and F1 score;

[0094] Overall recall: 92%.

[0095] The overall F1 score is 92.5%, which shows that the model has a good balance between precision and recall.

[0096] c. Real-time performance;

[0097] In actual production, the model demonstrated efficient processing capabilities, processing and analyzing about 20 frames of images per second on average.

[0098] S4 collects the image to be inspected of the cross section of the graphite column synthetic block from the production line in real time and inputs it into the quality inspection model to obtain the crystal distribution grade and crystal type grade of the graphite column synthetic block.

[0099] Furthermore, after predicting the crystal distribution grade and crystal form grade based on the image, the model will also match the corresponding production parameters (including equipment adjustment parameters, such as power parameters and pressure parameters) for the graphite column synthesis block in the database based on the crystal distribution grade and crystal form grade, so that the two can be combined to guide the subsequent parameter adjustment. For example, after matching the corresponding production parameters, the equipment parameters of the production equipment can be automatically or manually adjusted according to the production parameters, so that the quality of the next graphite column synthesis block is improved.

[0100] In the distribution level of crystals, the better the level, the more uniform the crystal distribution. However, the distribution of crystals with poorer levels will be locally clustered.

[0101] For images with densely distributed crystals, the effect of crystal grade detection is not very good. The reason is that when the crystals are too dense, they will interfere with each other during the image recognition and detection process, thus affecting the accuracy of the prediction.

[0102] Therefore, in order to improve the accuracy of crystal grade detection under low crystal distribution levels, in this implementation, if the crystal distribution level is lower than the threshold level, the crystal grade is re-inspected in different areas.

[0103] Generally speaking, in the detection above the crystal distribution threshold level (level 3 in this embodiment), the accuracy is acceptable, but once it is lower than level 3, the accuracy of the detection result will be unsatisfactory due to the above reasons.

[0104] In order to increase the speed of the entire system and reduce computing power requirements on the one hand, and to improve the reliability of detection on the other hand, in this embodiment, only images with crystal distribution levels below the threshold level are rechecked in partitions.

[0105] Specifically, the steps of the partition re-inspection specifically include:

[0106] S401 partitions the acquired complete cross-sectional image to obtain partitioned images.

[0107] Specifically, the crystals in the complete cross-sectional image are segmented by instance to obtain the position of each crystal in the complete cross-sectional image; the image within the threshold range around each crystal is framed and cropped to obtain a partitioned image.

[0108] Since the detection of the crystal grade needs to avoid interference as much as possible, the complete crystal edge contour needs to be retained during the frame selection. In this embodiment, a threshold range is set for the crystal, and the threshold range includes the complete crystal image and a part of the background.

[0109] More specifically, the threshold range is a range enclosed by points that are at a threshold distance from the crystal edge. In simple terms, each pixel point at the crystal edge is extended outward by the threshold distance as the edge of the frame selection.

[0110] With this setting, the integrity of the crystal image can be ensured while the selection of adjacent crystals can be avoided as much as possible.

[0111] Of course, there may still be other crystals within the threshold range. In order to deal with this problem, in this embodiment, the distance between the closest points of the two crystals is calculated, and half of the distance is used as the threshold distance for frame selection. For example, the original threshold distance is 20 pixels. However, when framing, it is found that other crystals have invaded the image, so the distance between the closest points of the two crystals is calculated. If the distance is 10 pixels, then half of the 10 pixels, that is, 5 pixels, is used as the new threshold distance between the two crystals for re-frame selection and cropping, so that there is only one crystal in each partition image, which is convenient for detecting the crystal grade of each crystal.

[0112] S402 performs image enhancement processing on the partition image.

[0113] The image enhancement in this step is consistent with the image enhancement step in the main model and will not be repeated here.

[0114] S403 inputs the enhanced partition image into a pre-trained crystal quality detection sub-model to obtain the crystal grade after re-inspection.

[0115] Example 2

[0116] like Figure 2 As shown, the present invention also provides a quality detection system for graphite column synthesis blocks based on image processing, comprising:

[0117] A training set construction module is used to collect cross-sectional images of graphite column synthesis blocks under various production parameters for constructing a training set; the cross-sectional images include complete cross-sectional images; the production parameters include equipment parameters, environmental parameters, and material parameters; the equipment parameters further include power parameters and pressure parameters;

[0118] Annotation module, used to annotate the crystal distribution grade and crystal form grade in the cross-section image of the graphite column synthesis block;

[0119] A model building module is used to input the annotated graphite column synthetic block cross-section image into the model for training, thereby obtaining a quality inspection model;

[0120] The detection module is used to collect the images to be detected of the cross section of the graphite column synthetic block from the production line in real time and input them into the quality detection model to obtain the crystal distribution grade and crystal type grade of the graphite column synthetic block, and match the corresponding production parameters.

[0121] In this step, the architecture of the crystal quality detection sub-model is different from the above quality detection model only in the output layer of the classifier, because it only detects the crystal grade, so the output layer is configured as 12 layers, corresponding to 12 crystal grades. The training set used during training is also the crystal partition image, rather than the complete image of the graphite synthesis block cross section.

[0122] The present invention also performs real-time detection of the quality grade of the graphite column synthesis block through a pre-constructed quality detection model, thereby avoiding the inconsistency of quality grade classification caused by different people performing the quality grade classification when performing the quality grade classification manually, and the inconsistency caused by different quality grade classification standards of different manufacturers, while also avoiding dependence on manual experience and improving efficiency.

[0123] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0124] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a computer terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0125] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A quality detection method for graphite column synthetic blocks based on image processing, characterized in that include: Collecting cross-sectional images of graphite column synthetic blocks under various production parameters for constructing a training set; the cross-sectional images include complete cross-sectional images; the production parameters include equipment parameters, environmental parameters, and material parameters; the equipment parameters include power parameters and pressure parameters; Annotate the crystal distribution grade and crystal form grade in the cross-section image of the graphite column synthetic block; The annotated graphite column synthetic block cross-section image is input into the model for training, thereby obtaining a quality inspection model; The images to be inspected of the cross section of the graphite column synthetic block are collected in real time from the production line and input into the quality inspection model to obtain the crystal distribution grade and crystal type grade of the graphite column synthetic block and match the corresponding production parameters.

2. The quality detection method of graphite column synthetic block based on image processing according to claim 1 is characterized in that: When constructing the training set, background images of graphite column synthetic blocks are collected and annotated as background as negative samples during annotation.

3. The quality detection method of graphite column synthetic block based on image processing according to claim 1 is characterized in that: The model is based on the fasterrcnn_resnet50_fpn model, and its classifier is FastRCNNPredictor; the output layer of the FastRCNNPredictor classifier is configured into 20 categories, including 12 crystal form levels, 7 crystal distribution levels and 1 background.

4. The quality detection method of graphite column synthetic block based on image processing according to claim 1 is characterized in that: After obtaining the crystal distribution grade and the crystal form grade of the graphite column synthetic block, if the crystal distribution grade is lower than the threshold grade, the crystal form grade is re-inspected in different areas.

5. The quality detection method of graphite column synthetic block based on image processing according to claim 1 is characterized in that The steps of the partition re-inspection specifically include: Partitioning the acquired complete cross-sectional image to obtain a partitioned image; Performing image enhancement processing on the partition image; The enhanced partitioned image is input into the pre-trained crystal quality detection sub-model to obtain the crystal grade after re-inspection.

6. The quality detection method of graphite column synthetic block based on image processing according to claim 5 is characterized in that The steps of partitioning the acquired complete cross-sectional image specifically include: Perform instance segmentation on the crystals in the complete cross-sectional image to obtain the position of each crystal in the complete cross-sectional image; The image within the threshold range around each crystal was framed and cropped to obtain a partitioned image.

7. The quality detection method of graphite column synthetic block based on image processing according to claim 6 is characterized in that: The threshold range is a range enclosed by points whose distance from the crystal edge is the threshold distance.

8. The method for detecting the amount of graphite column synthetic blocks based on image processing according to claim 7 is characterized in that: When an image within a certain crystal threshold range contains another crystal, the distance between the closest points of the two crystals is calculated, and half of the distance is selected as the threshold distance.

9. The method for detecting the amount of graphite column synthetic blocks based on image processing according to claim 5, characterized in that The steps of image enhancement specifically include: Resize partition images to the same resolution; Convert the partition image to grayscale; Increase the number of samples by randomly rotating, mirroring, and scaling.

10. A quality inspection system for graphite column synthesis blocks based on image processing, characterized in that include: A training set construction module is used to collect cross-sectional images of graphite column synthesis blocks under various production parameters for constructing a training set; the cross-sectional images include complete cross-sectional images; the production parameters include equipment parameters, environmental parameters, and material parameters; the equipment parameters include power parameters and pressure parameters; Annotation module, used to annotate the crystal distribution grade and crystal form grade in the cross-section image of the graphite column synthesis block; A model building module is used to input the annotated graphite column synthetic block cross-section image into the model for training, thereby obtaining a quality inspection model; The detection module is used to collect the image to be detected of the cross section of the graphite column synthetic block from the production line in real time and input it into the quality detection model, so as to obtain the crystal distribution grade and crystal type grade of the graphite column synthetic block.

Images