Methods, apparatus, equipment and media for identifying electron microscopy images of metallocene catalysts

Abstract

This invention discloses a method, apparatus, device, and medium for recognizing metallocene catalysts from electron microscopy (TEM) images. The method includes the following steps: acquiring several scanning transmission electron microscopy (STEM) images of metallocene catalysts to construct a first dataset; annotating the bounding boxes of aggregation points or bright spots in each TEM image in the first dataset to generate a second dataset; training a YOLOv3 neural network model using the second dataset to construct a recognition model; and using the recognition model to identify the activation centers of the metallocene catalyst in the STEM image to be detected. This invention utilizes deep learning to recognize TEM images, solving the problems of poor generalization ability and difficulty in covering special scenarios in existing recognition methods.

Description

Technical Field

[0001] This invention relates to the field of scanning transmission electron microscopy (STEM) image analysis of metallocene catalysts, and particularly to methods, apparatus, equipment, and media for identifying STEM images of supported metallocene catalysts in heterogeneous polymerization systems. Background Technology

[0002] Metallocene catalysts refer to catalytic systems composed of group IVB transition metal complexes (such as Ti, Zr, and Hf) as the main catalyst and alkylaluminoxanes (such as MAO) or organoborides as co-catalysts. Metallocene catalysts are classified into homogeneous polymerization and heterogeneous polymerization; heterogeneous polymerization consists of metallocene compounds, co-catalysts, and supports. Due to the water and oxygen sensitivity of MAO and metallocene compounds, and the limitation that the main metallocene content should be below 0.5%, the dispersion of metallocene compounds on supports has not been systematically studied.

[0003] Chinese invention patent CN112132785B describes a method and system for identifying and analyzing transmission electron microscopy (TEM) images of two-dimensional materials. This method rapidly identifies the atomic positions of two materials from high-resolution (scanning) TEM images and accurately determines the morphology, angles, and bond lengths of the two-dimensional materials. However, this patent primarily targets two-dimensional materials and cannot identify aggregated points as atomic centers. Invention patent CN104820994A describes an analysis method suitable for continuous high-resolution TEM images. This method performs frame segmentation, conversion, and correlation processing on continuous high-resolution dynamic images, converting atomic or atomic cluster motion information into quantitatively expressible information. However, this method has poor generalization ability, cannot learn autonomously, has low recognition efficiency, and the TEM images are affected by sample information and imaging techniques, making it difficult to cover some special scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for recognizing electron micrographs of metallocene catalysts. By using deep learning methods to recognize electron micrographs, this method overcomes the deficiencies of existing recognition methods, such as poor generalization ability and difficulty in covering special scenarios.

[0005] To achieve the above objectives, the present invention provides a method for recognizing electron microscopy images of metallocene catalysts, comprising: acquiring several scanning transmission electron microscopy images of metallocene catalysts to construct a first dataset; annotating the aggregation points or bright spots of each electron microscopy image in the first dataset with bounding boxes to generate a second dataset; training a YOLOv3 neural network model using the second dataset to construct a recognition model; and using the recognition model to identify the activation centers of the scanning transmission electron microscopy images of the metallocene catalyst to be detected.

[0006] In some embodiments of the present invention, the step of acquiring a plurality of metallocene catalyst scanning transmission electron microscope (STEM) images and constructing a first dataset further includes the following steps: scaling the plurality of metallocene catalyst STEM images to a first size; normalizing the scaled plurality of metallocene catalyst STEM images; and converting the format of the normalized plurality of metallocene catalyst STEM images into the format used by the YOLOv3 neural network model.

[0007] In some embodiments of the present invention, the step of annotating the bounding boxes of the cluster points or bright spots in each electron microscopy image in the first dataset to generate a second dataset further includes the following steps: using the LabelImg annotation tool to annotate the bounding boxes of the cluster points, generating an XML file including cluster point category, cluster particle size, distance between cluster centers, and bounding box corner coordinate information; wherein, the second dataset includes the XML file.

[0008] In some embodiments of the present invention, the YOLOv3 neural network model uses the Darknet53 network for feature extraction.

[0009] In some embodiments of the present invention, the method for recognizing the electron microscopy image of the metallocene catalyst further includes: inputting the scanning transmission electron microscopy image of the metallocene catalyst to be detected into the recognition model to generate a target box for the active center; equating the area of ​​the irregular recognition object in the target box to a circle and calculating the diameter of the equivalent circle; extracting the diameter data of all equivalent circles to generate a size distribution map; and analyzing the performance of the metallocene catalyst to be detected based on the size distribution map.

[0010] In some embodiments of the present invention, the center of the equivalent circle is the center of the target box corresponding to the irregular identification object; the identification method of the metallocene catalyst electron microscopy image further includes: calculating the distance and average interval between all the target boxes based on the centers of all the equivalent circles.

[0011] In some embodiments of the present invention, the step of obtaining scanning transmission electron microscopy images of several metallocene catalysts further includes the following steps:

[0012] The preparation of a metallocene catalyst scanning transmission electron microscope (STEM) sample comprises: dispersing the metallocene catalyst sample in a glove box using an anhydrous solvent; adding the metallocene catalyst sample dropwise onto a micro-gland or copper grid for the STEM using a dropper; and transferring the metallocene catalyst sample to a vacuum transfer sample holder after heating or purging the sample for a certain period of time.

[0013] The metallocene catalyst sample on the vacuum transfer sample rod is transferred to an atomically resolved spherical aberration-corrected probe with a high-angle annular dark-field probe in the absence of air, and an atomically resolved electron microscope image of the metallocene catalyst sample is obtained by taking pictures.

[0014] In some embodiments of the present invention, the monitoring metrics for training the YOLOv3 neural network model using the second dataset include one or a combination of precision, recall, and mean precision.

[0015] On the other hand, the present invention also provides a device for recognizing electron micrographs of metallocene catalysts, employing the aforementioned method for recognizing electron micrographs of metallocene catalysts. This device comprises at least: a first dataset construction module for acquiring several scanning transmission electron micrographs of metallocene catalysts and constructing a first dataset; a second dataset construction module for annotating the bounding boxes of aggregation points or bright spots in each electron micrograph image of the first dataset to generate a second dataset; a training module for training a YOLOv3 neural network model using the second dataset to construct a recognition model; and a recognition module for using the recognition model to identify the activation centers of the scanning transmission electron micrographs of the metallocene catalyst to be detected.

[0016] In another aspect, the present invention also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the method for recognizing electron micrographs of metallocene catalysts as described above.

[0017] In another aspect, the present invention also provides a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, it implements the method for recognizing electron micrographs of metallocene catalysts as described above.

[0018] As can be seen from the above solutions, the advantages of the present invention are:

[0019] This invention utilizes the YOLOv3 neural network model to identify electron microscopy images of metallocene catalysts, quantitatively extract information about active centers, achieve quantitative characterization of dispersion, and improve generalization ability and identification accuracy. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method for recognizing electron microscopy images of metallocene catalysts according to the present invention;

[0021] Figure 2 for Figure 1 Flowchart of step S10;

[0022] Figure 3 for Figure 2 Flowchart of step S100;

[0023] Figure 4 for Figure 1 Another flowchart for step S10;

[0024] Figure 5 for Figure 1 Flowchart of step S20;

[0025] Figure 6 A schematic diagram of the structure of the YOLOv3 neural network model used in the electron microscopy image recognition method for metallocene catalysts of the present invention;

[0026] Figure 7 for Figure 1 Flowchart of step S40;

[0027] Figure 8a Image recognition results of agglomerated metallocene catalyst under 4M magnification (I);

[0028] Figure 8b Image recognition results of the agglomerated metallocene catalyst under 4M magnification (II);

[0029] Figure 8c Image recognition results of agglomerated metallocene catalyst under 4M magnification (Part 3);

[0030] Figure 8d Image recognition effect of scanning transmission electron microscopy of the aggregated metallocene catalyst under 4M magnification (IV);

[0031] Figure 8e Image recognition results of agglomerated metallocene catalyst under 4M magnification (V);

[0032] Figure 8f Image recognition results of agglomerated metallocene catalyst under 2M magnification (I);

[0033] Figure 8g Image recognition results of the agglomerated metallocene catalyst under 2M magnification (II);

[0034] Figure 8h Image recognition results of agglomerated metallocene catalyst under 2M magnification (Part 3);

[0035] Figure 8i This is a scanning transmission electron microscope (STEM) image of the aggregated metallocene catalyst under test at 500k magnification, representing the identification effect.

[0036] Figure 9aImage recognition results of non-agglomerated metallocene catalyst under 2M magnification (I);

[0037] Figure 9b Image recognition results of non-agglomerated metallocene catalyst under 2M magnification (II);

[0038] Figure 10 Size distribution diagrams of a new catalyst and three catalyst samples with different activities;

[0039] Figure 11 This is a schematic diagram of the structure of the electron microscope image recognition device for metallocene catalysts of the present invention;

[0040] Figure 12 for Figure 11 A schematic diagram of the structure of the first dataset construction module 20;

[0041] Figure 13 for Figure 12 A schematic diagram of the preprocessing module 202;

[0042] Figure 14 for Figure 11 A schematic diagram of the structure of the second dataset construction module 21;

[0043] Figure 15 for Figure 11 A schematic diagram of the structure of the identification module 23;

[0044] Figure 16 This is a schematic diagram of the structure of the electronic device of the present invention;

[0045] In the attached figures, the following labels are used:

[0046] Methods for identifying 1'-metallocene catalysts from electron microscopy images;

[0047] 10-YOLOv3 neural network model;

[0048] 100-Darknet53 network;

[0049] Recognition device for electron microscopy images of 2'-metallocene catalysts;

[0050] 20 - First Dataset Construction Module;

[0051] 200 - Preparation module;

[0052] 201-Shooting Module;

[0053] 202 - Preprocessing module;

[0054] 202a - Scaling Unit;

[0055] 202b - Normalized Unit;

[0056] 202c - Format Conversion Unit;

[0057] 21-Second Dataset Construction Module;

[0058] 210 - Labeling Unit;

[0059] 211-Divide into units;

[0060] 22-Training Module;

[0061] 23 - Recognition Module;

[0062] 230 - Input Unit;

[0063] 231 - First Calculation Unit;

[0064] 232 - Generating Unit;

[0065] 233 - Analysis Unit;

[0066] 234 - Second Calculation Unit;

[0067] 3'-Electronic devices;

[0068] 30-Processor;

[0069] 31-Memory;

[0070] 310 - Computer programs;

[0071] S10~S40, S100~S101, S100a~S100c, S110~S112, S200~S201, S400~S404 - Steps. Detailed Implementation

[0072] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments to further understand the purpose, solution and effect of the present invention, but it is not intended to limit the scope of protection of the appended claims.

[0073] References to "embodiment," "another embodiment," "this embodiment," etc., in the specification refer to embodiments that may include specific features, structures, or characteristics, but not every embodiment must include these specific features, structures, or characteristics. Furthermore, such expressions do not refer to the same embodiment. Moreover, when describing specific features, structures, or characteristics in conjunction with embodiments, whether or not explicitly described, it is indicated that incorporating such features, structures, or characteristics into other embodiments is within the knowledge of those skilled in the art.

[0074] In this invention, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. While certain terms are used in the specification and subsequent claims to refer to specific components or parts, those skilled in the art will understand that users or manufacturers may use different names or terms to refer to the same component or part. This specification and claims do not distinguish components or parts by differences in name, but by differences in function. The terms "comprising" and "including" as used throughout the specification and claims are open-ended and should therefore be interpreted as "including but not limited to."

[0075] This invention primarily addresses the problem that in heterogeneous polymerization systems, the active components of metallocene catalysts, due to their low loading and easy reaction upon air contact, cannot be characterized in terms of dispersion on the support surface. Furthermore, during electron microscopy image analysis, especially in atomically resolved images obtained using scanning transmission mode, data cannot be automatically extracted and quantified.

[0076] Figure 1 The flowchart below shows a method 1' for identifying electron micrographs of metallocene catalysts according to an embodiment of the present invention. The method 1' specifically includes the following steps:

[0077] S10: Obtain scanning transmission electron microscopy images of several metallocene catalysts and construct the first dataset;

[0078] S20: Label the clumping points or bright spots of each electron microscope image in the first dataset with bounding boxes to generate the second dataset;

[0079] S30: Train the YOLOv3 neural network model 10 using the second dataset to build a recognition model;

[0080] S40: Identify the activation centers in the scanning transmission electron microscopy image of the metallocene catalyst to be tested using a recognition model.

[0081] In step S10,

[0082] like Figure 2 and Figure 3 As shown, step S10 further includes the following steps:

[0083] S100: Preparation of scanning transmission electron microscopy samples of metallocene catalysts, including:

[0084] S100a: The metallocene catalyst sample was dispersed in a glove box using an anhydrous solvent;

[0085] S100b: The metallocene catalyst sample is dropped onto a micro-gauze or copper mesh specifically designed for transmission electron microscopy using a dropper;

[0086] S100c: After heating or purging the metallocene catalyst sample for a certain period of time, the metallocene catalyst sample is transferred to the vacuum transfer sample rod to obtain the metallocene catalyst scanning transmission electron microscope sample.

[0087] S101: To capture electron microscope images of metallocene catalyst scanning transmission electron microscope (STEM) samples, including: transferring the metallocene catalyst STEM sample to an atomically resolved spherical aberration correction device with a high-angle annular dark field probe in the absence of air, and capturing atomically resolved electron microscope images of the metallocene catalyst STEM sample.

[0088] Specifically, the original electron microscopy images of the metallocene catalyst are acquired as follows: The silica-supported metallocene catalyst is dispersed in a glove box using anhydrous solvents without hydroxyl groups and with low boiling points, such as anhydrous cyclohexane, anhydrous n-heptane, anhydrous n-hexane, and anhydrous n-pentane. Drying or purging methods are employed, including infrared lamps, incandescent lamps, and hot plates. Heating times can be 10 min, 20 min, 30 min, or 40 min. Purging atmospheres include N2, Ar, He, methane, and ethane, with purging times of 20 min, 30 min, 40 min, and 60 min. Residual organic matter in the sample is removed. The sample is dispersed on a micro-mesh or copper grid and then transferred to a vacuum transfer sample holder. After plasma cleaning, it is transferred under air-isolated conditions to an atomically resolved spherical aberration correction process using a high-angle annular dark-field probe. Atomic-resolution electron microscopy images are then obtained by imaging. Figure 4 As shown.

[0089] Preferably, the silica-supported metallocene catalyst is first dispersed in anhydrous n-pentane in a glove box, followed by cyclohexane; N2 purging is used for 40 min; the dispersion grid is preferably a microgrid; the aberration-corrected transmission electron microscope conditions are: 8C, CL light lanthanum 30 μm, HAADF mode, camera length 8 cm.

[0090] Specifically, the following process was used to prepare the supported metallocene catalyst scanning transmission electron microscopy (STEM) samples:

[0091] The silica-supported metallocene catalyst was dispersed in anhydrous hexane in a glove box, ground, and then the sample was dropped onto a transmission electron microscope (TEM) copper grid using a dropper. The copper grid with the sample was placed on a heating plate and heated at 80°C for 30 minutes. Finally, the sample was transferred to a vacuum transfer sample holder. Alternatively,

[0092] The silica-supported metallocene catalyst was dispersed in anhydrous cyclohexane in a glove box, ground, and then the sample was dropped onto a copper grid for transmission electron microscopy using a dropper. The copper grid with the sample was then heated over an incandescent lamp for 10 minutes. Finally, the sample was transferred to a vacuum transfer sample holder. Alternatively,

[0093] The silica-supported metallocene catalyst was dispersed in anhydrous n-pentane in a glove box, ground, and then the sample was dropped onto a transmission electron microscope (TEM) microsphere using a dropper. The microsphere with the sample was then heated over an infrared lamp for 20 minutes. Finally, the sample was transferred to a vacuum transfer sample holder. Alternatively,

[0094] The silica-supported metallocene catalyst was dispersed in anhydrous n-heptane in a glove box. After grinding, the sample was dropped onto a copper grid for transmission electron microscopy using a dropper. The sample was then purged with nitrogen for 20 minutes. Finally, the sample was transferred to a vacuum transfer sample holder. Alternatively,

[0095] The silica-supported metallocene catalyst was dispersed in anhydrous hexane in a glove box, ground, and then the sample was dropped onto a copper grid for transmission electron microscopy using a dropper. The sample was then purged with Ar for 30 minutes. Finally, the sample was transferred to a vacuum transfer sample holder. Alternatively,

[0096] The silica-supported metallocene catalyst was dispersed in anhydrous hexane in a glove box. After grinding, the sample was dropped onto a copper grid for transmission electron microscopy using a dropper. The sample was then purged with helium for 30 minutes. Finally, the sample was transferred to a vacuum transfer sample holder. Alternatively,

[0097] The silica-supported metallocene catalyst was dispersed in anhydrous hexane in a glove box. After grinding, the sample was dropped onto a copper grid for transmission electron microscopy using a dropper. The sample was then purged with methane for 40 minutes. Finally, the sample was transferred to a vacuum transfer sample holder. Alternatively,

[0098] The silica-supported metallocene catalyst was dispersed in anhydrous hexane in a glove box, ground, and then the sample was dropped onto a copper grid for transmission electron microscopy using a dropper. The sample was then purged with ethane for 40 minutes. Finally, the sample was transferred to a vacuum transfer sample holder.

[0099] In some embodiments, such as Figure 4 As shown, step S10 further includes the following steps:

[0100] S110: Scale several metallocene catalyst scanning transmission electron microscopy images to a first size;

[0101] S111: Normalize several scaled scanning transmission electron microscope images of metallocene catalysts.

[0102] S112: Convert the format of several normalized scanning transmission electron microscope images of metallocene catalysts into the format used by the YOLOv3 neural network model 10.

[0103] Specifically, before constructing the first dataset, the raw electron microscopy images of the metallocene catalysts were preprocessed, including:

[0104] Scaling: The original electron microscope image is scaled to a fixed size (e.g., first size: 416x416 pixels) to fit the input requirements of the YOLOv3 neural network model 10.

[0105] Normalization: Normalize the pixel values ​​of the scaled electron microscope image to between 0 and 1 for better numerical calculation.

[0106] Channel order: Convert the normalized electron microscope images from BGR format to RGB format for use with YOLOv3 neural network model 10.

[0107] The first dataset was constructed using the original electron microscopy images of the metallocene catalysts that had undergone the above pretreatment.

[0108] In step S20,

[0109] like Figure 5 As shown, step S20 further includes the following steps:

[0110] S200: Use the LabelImg annotation tool to annotate the bounding boxes of clusters or bright spots, and generate an XML file that includes the object category and bounding box corner coordinates;

[0111] S201: Divide the XML file into training and testing sets.

[0112] Specifically, the first dataset, following the VOC2007 data format, uses the LabelImg annotation tool on a Windows system to annotate each electron microscopy image, marking the bounding box and category of each target (e.g., aggregation points or bright spots). This generates an XML file containing aggregation point categories, aggregation particle sizes, distances between aggregation centers, and the coordinates of the four corner points of the bounding box. This XML file is defined as the second dataset, which is divided into training and testing sets for training and validation. Preferably, the training set accounts for 80% and the testing set for 20%. Aggregate point categories include catalyst particles, support particles, metal nanoparticles, or aggregates. Bright spots or aggregation points in the electron microscopy images of metallocene catalysts represent the active centers of the metallocene catalyst.

[0113] In step S30,

[0114] like Figure 6 As shown, the YOLOv3 neural network model 10 uses the Darknet53 network 100 for feature extraction. Training the YOLOv3 neural network model 10 using the second dataset to build the recognition model specifically includes the following steps:

[0115] First, modify the training path or the network configuration file for the YOLOv3 neural network model 10, including but not limited to parameters such as learning rate, batch size, and maximum number of iterations.

[0116] Next, load the YOLOv3 neural network model 10. For example, you can download the training weight file, such as YOLOv3.weights, from the YOLO website. This file includes a model based on the Darknet53 network 100. Use the instructions provided by Darknet to start the training process. During training, the Darknet53 network 100 is the feature extraction network used in the YOLOv3 neural network model 10. It borrows design ideas from ResNet and uses residual blocks to build the deep network. In the YOLOv3 neural network model 10, the feature maps extracted by the Darknet53 network 100 are used for multi-scale detection, a process involving feature concatenation.

[0117] The feature extraction process is implemented by the Darknet53 network 100, which has a total of 6 individual convolutional layers and 23 residual layers. Each residual layer contains 2 convolutional layers. The first 52 layers are used only for feature extraction, and the last layer is used to input the predicted value.

[0118] First, a pre-processed, fixed-size RGB image (e.g., 416*416 pixels) is input to the Darknet53 network. This image then passes through convolutional layers, a fundamental building block of the Darknet53 network. Each layer contains convolution operations, batch normalization (BN), and an activation function (typically Leaky ReLU). The convolution operations extract local features from the image, such as clumping size and distance. Finally, the image is output after passing through pooling layers.

[0119] Next, the YOLOv3 neural network model 10 was trained using the training set of the second dataset. The main metrics for monitoring the training process were one or a combination of precision, recall, and mean precision.

[0120] Finally, the model is evaluated using a test set. If the YOLOv3 neural network model 10 performs poorly in terms of precision, recall, and mean precision, parameters such as the learning rate and batch size can be adjusted to improve the overall generalization rate. The trained YOLOv3 neural network model 10 is defined as the recognition model. The test set can also be used periodically to evaluate the performance of the YOLOv3 neural network model 10 to check for overfitting or underfitting.

[0121] In step S40,

[0122] like Figure 7 As shown, the method of identifying the metallocene catalyst to be detected using a recognition model in a scanning transmission electron microscope image also includes the following steps:

[0123] S400: Input the scanning transmission electron microscope image of the metallocene catalyst to be detected into the recognition model to generate the target bounding box of the active center;

[0124] S401: Equivalently represent the area of ​​irregularly identified objects in the target bounding box as a circle, and calculate the diameter of the equivalent circle;

[0125] S402: Extract the diameter data of all equivalent circles and generate a size distribution map;

[0126] S403: Analyze the performance of the metallocene catalyst under test based on the size distribution diagram.

[0127] Step S40 also includes:

[0128] S404: Calculate the distance between all bounding boxes and the average spacing based on the centers of all equivalent circles.

[0129] Specifically, for scanning transmission electron microscopy (STEM) samples of aggregated metallocene catalysts, the area of ​​the irregularly shaped identification object is represented as an equivalent circle. The center of the target bounding box corresponding to the irregularly shaped identification object is taken as the center of the circle, and the diameter of the equivalent circle is taken as the diameter of the target bounding box. The identification effect is shown in the image below. Figures 8a to 8i As shown, the mutual distances and average spacing between agglomerated particles (i.e., agglomeration points in the target box) can be obtained simultaneously. The particle size of the active center is directly related to the catalyst activity. The smaller the distance between particles, the larger the surface area of ​​the catalyst may be, which may provide higher catalytic activity. By analyzing the distance between particles, the catalyst preparation process can be optimized to obtain a more uniform and efficient catalyst structure. Therefore, obtaining the mutual distances and average spacing between agglomerated particles is of great significance for catalyst design, performance optimization, reaction engineering, and fundamental research in materials science.

[0130] Figures 8a to 8e This is an image of the identification effect of a certain metallocene catalyst under scanning transmission electron microscopy at 4M magnification. Figures 8f to 8h This is an image of the identification effect of a certain metallocene catalyst under scanning transmission electron microscopy at 2M magnification. Figure 8i This is a scanning transmission electron microscope (STEM) image of a metallocene catalyst at 500K magnification, where "4M" refers to four million, "2M" to two million, and "500K" to 500,000. Figures 8a-8i It can be seen that, compared with conventional algorithms, the recognition model of the present invention can effectively recognize and detect electron microscope images at different magnifications.

[0131] For scanning transmission electron microscopy (STEM) images of non-agglomerated metallocene catalysts, the identification model primarily identifies bright spots in the images. These bright spots represent the active centers of the non-agglomerated metallocene catalyst, such as... Figure 9a and Figure 9b The image shown is a recognition result of scanning transmission electron microscopy (STEM) images of certain non-agglomerated metallocene catalysts at 4M magnification. Figure 9a and Figure 9b It can be seen that the recognition model of this invention is effective for particles smaller than 0.5 nm. Figure 9a and Figure 9b Samples with the same dimensions (of the bright spots) are also applicable.

[0132] To facilitate understanding of the above embodiments, a specific application scenario of the above embodiments will be used as an example for illustration below:

[0133] A new catalyst and three catalyst samples with different activities were selected. Atomic-resolution electron microscopy (EM) images were obtained, and then these images were imported into a recognition model to obtain recognized EEM images. The size distribution of active centers in the EEM images was extracted to obtain a size distribution map, such as... Figure 10As shown in the diagram, the size variation reflected in this size distribution map is related to the diameter of the catalytic reaction activity energy. Figure 10 In the graph, the horizontal axis represents the size distribution range of active centers, and the vertical axis represents the proportion of active centers in different samples within that size distribution. "1", "2", and "3" represent three catalyst samples with different activities. From "0-1nm" to "2-3nm", the proportions of active centers in "fresh agent", "sample 1", "sample 2", and "sample 3" are represented from bottom to top, respectively. From "0-1nm" to "2-3nm", the proportion of active centers in "fresh agent" and "sample 1" gradually decreases, reaching 0% in "3-4nm" and "4-5nm". In "3-4nm" and "4-5nm", "sample 2" and "sample 3" are represented from bottom to top, respectively. Figure 10 Statistical data from numerous images show that the size of the active sites on the catalyst increases gradually with increasing reaction time, rather than abruptly. This result is significant for optimizing the catalyst structure, stabilizing the size of the active sites, and ultimately improving the long-term stability of the catalyst.

[0134] This invention addresses the challenges posed by metallocene catalysts with loadings less than 0.5% and sensitivity to water and oxygen. First, it explores and implements vacuum transfer and sample dispersion methods. Considering the single active site and low content characteristics of metallocene catalysts, it employs an aberration-corrected transmission electron microscope (TEM) with a high-angle annular dark-field probe for characterizing these single active sites. This TEM is the only tool capable of directly observing atoms. Finally, image recognition is used to achieve automatic identification and analysis of the scanning transmission electron microscope images.

[0135] This invention employs vacuum transfer to preserve the intrinsic information of the sample, uses anhydrous and oxygen-free low-boiling-point reagents to ensure high sample dispersion and maintain intrinsic structural information, and uses an aberration-corrected transmission electron microscope with atomic number imaging and a high-angle annular dark-field probe to directly observe single atoms, thus obtaining the direct distribution of the metallocene main catalyst on the support.

[0136] This invention utilizes deep learning methods and a regression-based target recognition algorithm. It uses provided electron microscope images as a dataset, adjusts the dataset images for feature extraction, performs feature stitching, and finally achieves target detection with high recognition accuracy.

[0137] The following are apparatus embodiments corresponding to the above method embodiments. This embodiment can be implemented in conjunction with the above embodiments. The relevant technical details mentioned in the above embodiments remain valid in this embodiment, and will not be repeated here to reduce repetition. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.

[0138] like Figure 11 As shown in the diagram, another embodiment of the present invention provides a structural schematic diagram of a metallocene catalyst electron microscope image recognition device 2', employing the aforementioned metallocene catalyst electron microscope image recognition method 1'. The metallocene catalyst electron microscope image recognition device 2' includes at least:

[0139] The first dataset construction module 20 is used to acquire scanning transmission electron microscopy images of several metallocene catalysts and construct the first dataset.

[0140] The second dataset construction module 21 is used to annotate the bounding boxes of the cluster points or bright spots of each electron microscope image in the first dataset to generate the second dataset;

[0141] Training module 22 is used to train YOLOv3 neural network model 10 using the second dataset to build a recognition model;

[0142] The identification module 23 is used to identify the activation centers in the scanning transmission electron microscope image of the metallocene catalyst to be detected using an identification model.

[0143] In some embodiments, such as Figure 12 As shown, the first dataset construction module 20 also includes:

[0144] Preparation module 200: Used to prepare metallocene catalyst scanning transmission electron microscope (STEM) samples, including: dispersing the metallocene catalyst sample in a glove box using an anhydrous solvent; adding the metallocene catalyst sample dropwise onto a micro-gland or copper grid for STEM using a dropper; and transferring the metallocene catalyst sample to a vacuum transfer sample holder after heating or purging the sample for a certain period of time to obtain the metallocene catalyst STEM sample.

[0145] Image module 201: Used to capture electron microscope images of metallocene catalyst scanning transmission electron microscope samples, including: transferring the metallocene catalyst scanning transmission electron microscope sample to an atomically resolved spherical aberration correction process with a high-angle annular dark field probe in the absence of air, and capturing atomically resolved electron microscope images of the metallocene catalyst scanning transmission electron microscope sample.

[0146] Preprocessing module 202 is used to preprocess scanning transmission electron microscopy images of several metallocene catalysts, such as... Figure 13 As shown, the preprocessing module 201 includes:

[0147] Scaling unit 202a is used to scale several metallocene catalyst scanning transmission electron microscopy images to a first size;

[0148] Normalization unit 202b is used to normalize several scaled scanning transmission electron microscope images of metallocene catalysts.

[0149] The format conversion unit 202c is used to convert the format of several normalized scanning transmission electron microscope images of metallocene catalysts into the format used by the YOLOv3 neural network model.

[0150] In some embodiments, such as Figure 14 As shown, the second dataset construction module 21 also includes:

[0151] Annotation unit 210 is used to annotate the bounding boxes of cluster points or bright spots using the LabelImg annotation tool, and generate an XML file that includes object category and bounding box corner coordinate information;

[0152] Division unit 211 is used to divide the XML file into training and test sets.

[0153] In some embodiments, the YOLOv3 neural network model 10 in training module 22 includes Darknet53 network 100.

[0154] In some embodiments, such as Figure 15 As shown, the recognition module 23 includes:

[0155] The input unit 230 is used to input the scanning transmission electron microscope image of the metallocene catalyst to be detected into the recognition model to generate the target bounding box of the active center;

[0156] The first calculation unit 231 is used to convert the area of ​​the irregularly identified object in the target box into a circle and calculate the diameter of the equivalent circle.

[0157] Generation unit 232 is used to extract the diameter data of all equivalent circles and generate a size distribution map;

[0158] Analysis unit 233 is used to analyze the performance of the metallocene catalyst under test based on the size distribution diagram.

[0159] The second calculation unit 234 is used to calculate the distance between all target boxes and the average interval based on the centers of all equivalent circles.

[0160] Furthermore, those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the device 2' for recognizing electron microscope images of metallocene catalysts can be referred to the corresponding process in the aforementioned embodiment of the method 1' for recognizing electron microscope images of metallocene catalysts, and will not be repeated here.

[0161] Furthermore, another embodiment of the present invention also discloses an electronic device 3', such as... Figure 15 As shown, the content in the figure should not be considered as any limitation on the scope of use of the present invention.

[0162] Figure 16This is a schematic diagram of the structure of an electronic device 3' provided in an embodiment of the present invention. Specifically, the electronic device 3' may include at least one processor 30 and at least one memory 31. The memory 31 stores a computer program 310, which is loaded and executed by the processor 30 to implement the relevant steps in the method 1 for recognizing electron micrographs of metallocene catalysts disclosed in any of the foregoing embodiments, such as... Figure 1 The steps are shown.

[0163] In addition, the memory 31, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, and the storage method can be temporary storage or permanent storage.

[0164] In addition to including a computer program capable of performing the method 1' for identifying electron micrographs of metallocene catalysts executed by electronic device 3' as disclosed in any of the foregoing embodiments, computer program 310 may further include a computer program capable of performing other specific tasks.

[0165] Furthermore, this embodiment of the invention also discloses a computer storage medium storing computer-executable instructions. When these computer-executable instructions are loaded and executed by a processor, they implement step 1' of the method for recognizing electron micrographs of metallocene catalysts disclosed in any of the foregoing embodiments, such as... Figure 1 The steps are shown.

[0166] It should be understood that the storage medium in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0167] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms fall within the scope of protection of the present invention.

Claims

1. A method for recognizing electron microscopy images of metallocene catalysts, characterized in that, Includes the following steps: Acquire scanning transmission electron microscopy images of several metallocene catalysts and construct the first dataset; Boundary boxes are labeled for the clusters or bright spots in each electron microscopy image in the first dataset to generate the second dataset; The YOLOv3 neural network model was trained using the second dataset to construct a recognition model; The identification model is used to identify the active centers in the scanning transmission electron microscopy images of the metallocene catalyst to be tested.

2. The method for recognizing electron microscopy images of metallocene catalysts according to claim 1, characterized in that, The step of acquiring several scanning transmission electron microscopy images of metallocene catalysts and constructing the first dataset also includes the following steps: The scanning transmission electron microscope images of the aforementioned metallocene catalysts are scaled to a first size; The scaled scanning transmission electron microscope images of the aforementioned metallocene catalysts were normalized. The normalized scanning transmission electron microscope images of the metallocene catalysts are converted to the format used by the YOLOv3 neural network model.

3. The method for recognizing electron microscopy images of metallocene catalysts according to claim 1, characterized in that, The step of annotating the bounding boxes of the clusters or bright spots in each electron microscopy image in the first dataset to generate the second dataset further includes the following steps: The LableImg annotation tool is used to annotate the bounding boxes of the cluster points, generating an XML file that includes cluster point category, cluster particle size, distance between cluster centers, and bounding box corner coordinates; the second dataset includes the XML file.

4. The method for recognizing electron microscopy images of metallocene catalysts according to claim 1, characterized in that, The YOLOv3 neural network model uses the Darknet53 network for feature extraction.

5. The method for recognizing electron microscopy images of metallocene catalysts according to claim 1, characterized in that, Also includes: The scanning transmission electron microscope image of the metallocene catalyst to be detected is input into the recognition model to generate the target bounding box of the active center; The area of ​​the irregularly identified object in the target box is equivalent to a circle, and the diameter of the equivalent circle is calculated. Extract the diameter data of all equivalent circles and generate a size distribution map; The performance of the metallocene catalyst under test was analyzed based on the size distribution diagram.

6. The method for recognizing electron microscopy images of metallocene catalysts according to claim 5, characterized in that, The center of the equivalent circle is the center of the target box corresponding to the irregular identification object; the identification method of the metallocene catalyst electron microscopy image further includes: Calculate the distance between all the target boxes and the average interval based on the centers of all equivalent circles.

7. The method for recognizing electron microscopy images of metallocene catalysts according to claim 1, characterized in that, The process of obtaining scanning transmission electron microscopy images of several metallocene catalysts further includes the following steps: The preparation of a metallocene catalyst scanning transmission electron microscope (STEM) sample comprises: dispersing the metallocene catalyst sample in a glove box using an anhydrous solvent; adding the metallocene catalyst sample dropwise onto a micro-gland or copper grid for the STEM using a dropper; and transferring the metallocene catalyst sample to a vacuum transfer sample holder after heating or purging the sample for a certain period of time. The metallocene catalyst sample on the vacuum transfer sample rod is transferred to an atomically resolved spherical aberration-corrected probe with a high-angle annular dark-field probe in the absence of air, and an atomically resolved electron microscope image of the metallocene catalyst sample is obtained by taking pictures.

8. A device for recognizing electron micrographs of metallocene catalysts, characterized in that, The method for identifying electron micrographs of metallocene catalysts according to any one of claims 1 to 7, wherein the identification device for electron micrographs of metallocene catalysts comprises at least: The first dataset construction module is used to acquire scanning transmission electron microscopy images of several metallocene catalysts and construct the first dataset. The second dataset construction module is used to annotate the bounding boxes of the clusters or bright spots in each electron microscope image in the first dataset to generate the second dataset. The training module is used to train a YOLOv3 neural network model using the second dataset to build a recognition model; The identification module is used to identify the activation centers in the scanning transmission electron microscopy image of the metallocene catalyst to be detected using the identification model.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method for recognizing electron micrographs of metallocene catalysts as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; when executed by a processor, the computer programs implement the method for recognizing electron micrographs of metallocene catalysts as described in any one of claims 1 to 7.

Images