Method, device and system for classifying and identifying impurity particles in transformer oil and medium

Abstract

The invention discloses a method, device and system for classifying and recognizing impurity particles in transformer oil and a medium, and the method comprises the steps: obtaining an image of the impurity particles in the transformer oil in real time, taking the image as an original image, and forming a data set based on the original image; preprocessing the original image in the data set to obtain a preprocessed image; a K-means + + algorithm is adopted to generate a new prior frame; according to the new prior frame, training a YOLO target detection algorithm to obtain a trained YOLO target detection model; and performing target detection on the preprocessed images based on the trained YOLO target detection model, automatically identifying different types of impurity particles, automatically counting the number of prediction frames on all the images, and outputting the number of different types of particles. According to the invention, the accuracy and efficiency of monitoring the particle impurities in the transformer oil can be effectively improved, and important reference is provided for state evaluation and fault detection of the transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of detecting impurity particles in power transformer oil, and specifically relates to a method, device, system and medium for classifying and identifying impurity particles in transformer oil. Background Art

[0002] At present, the detection methods for impurity particles in power transformer oil stipulated by the power industry standards in China are the automatic particle counting method and the microscope method. Among them, the automatic particle counting method uses an automatic particle counter to automatically count the number of particles with different particle sizes; the microscope method grades the particle size of the oil product by visually comparing the oil sample and the oil particle size grading template, and qualitatively classifies the observed particles based on experience.

[0003] However, the automatic particle counting method cannot classify and identify the impurities in the oil, while the microscope method relies on naked-eye observation, with poor credibility and accuracy. Moreover, the microscope method requires off-line observation and cannot obtain complete particle information in the oil sample. These two existing methods cannot achieve comprehensive and on-line classification and identification of impurities in transformer oil.

[0004] In view of this, this application is specifically proposed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the existing methods cannot achieve comprehensive and on-line classification and identification of impurities in transformer oil, and there are problems such as low recognition accuracy, low classification types and detection efficiency. The purpose of the present invention is to provide a method, device, system and medium for classifying and identifying impurity particles in transformer oil. By using the liquid flow dynamic imaging technology to capture the particle images in the transformer oil in real time, and using the YOLO target detection model to classify and identify different types of impurity particles, the automatic and rapid identification and counting of the types and quantities of impurity particles in the transformer oil can be realized. The present invention can effectively improve the accuracy and efficiency of monitoring particle impurities in transformer oil, and provide an important reference for the condition assessment and fault detection of transformers.

[0006] The present invention is realized through the following technical solutions:

[0007] In the first aspect, the present invention provides a method for classifying and identifying impurity particles in transformer oil, and the method includes:

[0008] Obtain the impurity particle images in the transformer oil in real time and use them as the original images, and form a data set based on the original images;

[0009] Preprocess the original images in the data set to obtain the preprocessed images;

[0010] The K-means++ algorithm is adopted to generate new prior boxes; based on the new prior boxes, the YOLO object detection algorithm is trained to obtain a trained YOLO object detection model; the YOLO object detection algorithm adopts the YOLOv9 network structure;

[0011] Based on the trained YOLO object detection model, target detection is performed on the preprocessed image, different types of impurity particles are automatically identified, and the number of prediction boxes on all images is automatically counted, and the number of different types of particles is output.

[0012] Furthermore, impurity particle images in transformer oil are obtained in real time through a liquid flow dynamic imaging device, including:

[0013] The flow of transformer oil is controlled by a liquid flow pump, the area through which the transformer oil flows is illuminated by pulsed light, and a high-definition digital photosensitive element is used to perform dynamic imaging on the particles in the flowing transformer oil.

[0014] Furthermore, the original images in the dataset are preprocessed to obtain preprocessed images, including:

[0015] The original image is adjusted to a preset fixed size to obtain an adjusted image;

[0016] The adjusted image is normalized, and each pixel value is normalized to the interval [0,1] to obtain a normalized image;

[0017] And the normalized image is subjected to enhancement and image cropping processing to obtain a preprocessed image.

[0018] Furthermore, the K-means++ algorithm is adopted to generate new prior boxes, including:

[0019] A1, randomly select an anchor box from the dataset as the initial clustering center c, and calculate the distance D(x) between the remaining anchor boxes in the dataset and the initial clustering center c;

[0020] A2, based on the distance D(x), calculate the probability P(x) that each anchor box is selected as the clustering center, and select the anchor box farther from the current clustering center as the new clustering center;

[0021] A3, repeat steps A1 and A2 until K initial clustering centers are selected;

[0022] A4, calculate the distance between each anchor box and all K initial clustering centers, and assign the anchor box to the category of the clustering center with the closest distance to it;

[0023] A5, update and calculate the clustering center of each category.

[0024] A6. Repeat steps A4 and A5 until the cluster centers no longer change, and obtain K optimized cluster centers as the new prior boxes.

[0025] Further, the update calculation formula in step A5 is:

[0026]

[0027] where C i is the set of anchor boxes in the i-th clustering category, c i is the cluster center of this category, and x is the anchor box.

[0028] In a second aspect, the present invention further provides a device for classifying and identifying impurity particles in transformer oil, and the device includes:

[0029] A flow dynamic imaging device, configured to acquire images of impurity particles in transformer oil in real time and use them as original images, and form a data set based on the original images;

[0030] An image processing module, configured to preprocess the original images in the data set to obtain preprocessed images;

[0031] A deep learning identification module, configured to use the K-means++ algorithm to generate new prior boxes; according to the new prior boxes, train the YOLO object detection algorithm to obtain a trained YOLO object detection model; perform object detection on the preprocessed images based on the trained YOLO object detection model, automatically identify different types of impurity particles, and automatically count the number of prediction boxes on all images, and output the number of different types of particles.

[0032] Further, the flow dynamic imaging device includes a liquid flow pump, a pulsed light, and a photosensitive element;

[0033] The liquid flow pump is configured to control the flow of transformer oil;

[0034] The pulsed light is configured to illuminate the area through which the transformer oil flows;

[0035] The photosensitive element is configured to perform dynamic imaging on the particles in the flowing transformer oil.

[0036] In a third aspect, the present invention further provides a system for classifying and identifying impurity particles in transformer oil, and the system includes:

[0037] A flow dynamic imaging system, using the flow dynamic imaging device in the above-mentioned device for classifying and identifying impurity particles in transformer oil, configured to acquire images of impurity particles in transformer oil in real time;

[0038] An image post - processing computer is internally provided with the image - processing module and the deep - learning recognition module in the above - mentioned impurity particle classification and recognition device for transformer oil. It is used to input the processed data through the image - processing module into the deep - learning recognition module for classification and recognition, so as to obtain the impurity particle classification and quantity information.

[0039] A transformer oil sample container, which is used to provide transformer oil samples for imaging by the flow - type dynamic imaging system.

[0040] A waste oil container, which is used to hold the transformer oil samples after being used by the flow - type dynamic imaging system.

[0041] Furthermore, a report generation module is also set in the image post - processing computer.

[0042] The report generation module is used to generate a report on the impurity particle classification and quantity information identified by the deep - learning recognition module, so as to assist in evaluating the operating state of the transformer.

[0043] In a fourth aspect, the present invention further provides a computer - readable storage medium. The computer - readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above - mentioned method for classifying and recognizing impurity particles in transformer oil.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0045] The method, device, system and medium for classifying and recognizing impurity particles in transformer oil according to the present invention use the liquid - flow dynamic imaging technology to capture the particulate matter images in the transformer oil in real - time, and use the YOLO object detection model to classify and recognize different types of impurity particles, covering impurities such as fiber particles, carbon particles, copper particles and bubbles. Furthermore, it realizes the automatic, rapid recognition and counting of the types and quantities of impurity particles in the transformer oil. The present invention can effectively improve the accuracy and efficiency of monitoring particulate impurities in transformer oil, and provide an important reference for the condition assessment and fault detection of transformers. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0047] Figure 1 is the flow chart of the method for classifying and recognizing impurity particles in transformer oil according to the present invention;

[0048] Figure 2 is the pre - processing flow chart of the present invention;

[0049] Figure 3 is the schematic diagram of the network structure of YOLOv9 of the present invention;

[0050] Figure 4 This is the structural block diagram of the device for classifying and identifying impurity particles in transformer oil of the present invention;

[0051] Figure 5 This is the operation flow chart of the device for classifying and identifying impurity particles in transformer oil of the present invention;

[0052] Figure 6 This is the system block diagram of the device for classifying and identifying impurity particles in transformer oil of the present invention;

[0053] Figure 7 This is the schematic diagram of flow imaging of the present invention.

[0054] Reference numerals and corresponding component names:

[0055] 1 - Flow dynamic imaging system, 2 - Image post - processing computer, 3 - Transformer oil sample container, 30 - Test sample, 4 - Waste oil container, 5 - Pulse light source, 6 - CMOS camera, 7 - Zoom lens. Specific embodiments

[0056] In the following, the term "comprising" or "may comprise" that can be used in various embodiments of the present invention indicates the presence of the functions, operations or elements of the present invention, and does not limit the addition of one or more functions, operations or elements. In addition, as used in various embodiments of the present invention, the terms "comprising", "having" and their cognates are only intended to represent specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be construed as precluding the existence or addition of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items first.

[0057] In various embodiments of the present invention, the expression "or" or "at least one of A or / and B" includes any combination or all combinations of the recited words. For example, the expression "A or B" or "at least one of A or / and B" may include A, may include B, or may include both A and B.

[0058] In various embodiments of the present invention, expressions (such as "first", "second", etc.) used may modify various components in various embodiments, but do not limit the corresponding components. For example, the above expressions do not limit the order and / or importance of the components. The above expressions are only used for the purpose of distinguishing one component from other components. For example, the first user device and the second user device indicate different user devices, although both are user devices. For example, without departing from the scope of various embodiments of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component.

[0059] It should be noted that: If a description "connects" one component to another component, the first component can be directly connected to the second component, and a third component can be "connected" between the first component and the second component. Conversely, when a component is "directly connected" to another component, it can be understood that there is no third component between the first component and the second component.

[0060] The terms used in various embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the various embodiments of the present invention. As used herein, the singular form is intended to also include the plural form unless the context clearly indicates otherwise. Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the various embodiments of the present invention pertain. The terms (such as those defined in a commonly used dictionary) will be interpreted as having the same meaning as the contextual meaning in the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning unless clearly defined in the various embodiments of the present invention.

[0061] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with embodiments and drawings. The illustrative embodiments and descriptions of the present invention are only for explaining the present invention and do not constitute a limitation to the present invention.

[0062] The present invention is a method, device, system, and medium for classifying and identifying impurity particles by combining a liquid flow dynamic imaging system and a target detection model. By using liquid flow dynamic imaging technology to capture particulate matter images in transformer oil in real time, and using the YOLO target detection model to classify and identify different types of impurity particles, including fiber particles, carbon particles, copper particles, and bubbles, etc., thereby realizing the automatic and rapid identification and counting of the types and quantities of impurity particles in transformer oil. The present invention can effectively improve the accuracy and efficiency of monitoring particulate impurities in transformer oil, and provide an important reference for the condition assessment and fault detection of transformers.

[0063] Embodiment 1

[0064] As Figure 1 shown, the method for classifying and identifying impurity particles in transformer oil of the present invention includes:

[0065] Step 1, obtain the impurity particle images in transformer oil in real time and use them as original images, and form a data set based on the original images;

[0066] In this embodiment, the impurity particle images in transformer oil are obtained in real time through a liquid flow dynamic imaging device, including:

[0067] Control the flow of transformer oil through a liquid flow pump, illuminate the area through which the transformer oil flows using pulsed light, and use a high-definition digital photosensitive element to dynamically image the particles in the flowing transformer oil.

[0068] Step 2, preprocess the original images in the dataset to obtain preprocessed images;

[0069] In this embodiment, as Figure 2 shown, Step 2 specifically includes:

[0070] Adjust the original images to a preset fixed size (such as 640x640 size) to obtain adjusted images;

[0071] Perform normalization processing on the adjusted images, normalize each pixel value to the interval [0,1] to obtain normalized images;

[0072] And perform enhancement and image cropping processing on the normalized images to obtain preprocessed images.

[0073] Among them, enhancement includes flipping, rotating, cropping, color perturbation, etc., and the diversity of training data is increased through enhancement processing.

[0074] Step 3, adopt the K-means++ algorithm to generate new prior boxes; according to the new prior boxes, train the YOLO object detection algorithm to obtain a trained YOLO object detection model;

[0075] In this embodiment, the present invention adopts the YOLO object detection algorithm, which is a real-time object detection algorithm based on deep learning, has high efficiency and accuracy, and is improved on the basis version to improve the detection accuracy and speed.

[0076] The prior box refers to a set of anchor boxes with different sizes and dimensions set for the YOLO object detection algorithm before training. These anchor boxes can help the model optimize the object localization and classification performance during training and improve the matching degree with the real boxes. The original prior boxes in the YOLO object detection algorithm are generated by the K-means algorithm on the COCO dataset. However, due to the wide variety of object types and sizes in the COCO dataset, there are significant differences from the dataset of the method of the present invention; in addition, the aspect ratios of some objects (such as fiber particles, etc.) in the dataset adopted by the method of the present invention have a large deviation from the anchor boxes of the standard dataset. Therefore, it is necessary to re-optimize the prior boxes.

[0077] Therefore, on the existing basis, the present invention adopts the K-means++ algorithm to generate new prior boxes, and the specific steps are as follows:

[0078] A1, randomly initialize the clustering centers

[0079] Randomly select an anchor box from the dataset as the initial clustering center c, and calculate the distance D(x) between the remaining anchor boxes in the dataset and the initial clustering center c;

[0080] D(x) = 1 - IoU(x, c)

[0081] where IoU represents the intersection over union of the anchor box x and the clustering center c.

[0082] A2. Calculation of the probability of selecting the next clustering center

[0083] Based on the distance D(x), calculate the probability P(x) of each anchor box being selected as the clustering center, and select the anchor box farther from the current clustering center as the new clustering center; the expression of P(x) is:

[0084]

[0085] where X is the set of all anchor boxes, and D(x) is the distance between the anchor box x and the selected clustering center.

[0086] A3. Repeat the initialization process

[0087] Repeat steps A1 and A2 to select repeatedly until K initial clustering centers are selected;

[0088] A4. Assign anchor box categories

[0089] Calculate the distance between each anchor box and all K initial clustering centers, and assign the anchor box to the category of the clustering center with the closest distance;

[0090] A5. Update the clustering center

[0091] Use the following formula to update and calculate the clustering center of each category;

[0092]

[0093] where C i is the set of anchor boxes in the i-th clustering category, c i is the clustering center of this category, and x is the anchor box.

[0094] A6. Iterative optimization

[0095] Repeat steps A4 and A5 until the clustering center no longer changes, and obtain K optimized clustering centers as the new prior boxes.

[0096] Through the above optimization method, the newly generated new prior boxes are more in line with the characteristics of the dataset of the present invention, thereby further improving the detection performance of the model for the target.

[0097] In this embodiment, the YOLO object detection algorithm adopts the YOLOv9 network structure, uses a convolutional neural network (CNN) for feature extraction to capture semantic information in the image; then, a multi-scale feature fusion mechanism is adopted to combine low-level and high-level features to enhance the detection ability for targets of different sizes. The algorithm predicts the object category, bounding box, and confidence for each grid cell. The loss function calculation combines bounding box loss, classification loss, and confidence loss to optimize the model training process. Subsequently, overlapping boxes are removed through the non-maximum suppression (NMS) algorithm, and the optimal bounding boxes are retained, and finally the category and location of the object are output. Combining the weight file obtained through pre-training, it can efficiently display various impurity particles such as identified fiber particles, carbon particles, metal particles, and bubbles on the input picture, automatically count the number of prediction boxes on all images, and output the number of different types of particles.

[0098] As Figure 3 shown, the network structure of YOLOv9 consists of three parts: Backbone (backbone network), Neck (feature fusion network), and Head (detection head), aiming to achieve efficient object detection.

[0099] The role of the Backbone is to extract multi-level features of the input image. It mainly includes an input layer, a convolutional module, a RepNCSPELAN4 module, and an SPLPELAN module. The input layer starts from "0:Silence", and the input enters the network after preliminary preprocessing. The convolutional (Conv) module, such as nodes 1 and 2, performs basic operations for feature extraction. The RepNCSPELAN4 module: starts from nodes 3, 5, 7, 9, etc. This module is the core improvement module of YOLOv9, enhancing the multi-scale feature extraction ability while taking into account computational efficiency. The SPLPELAN module is at node 10, further compressing the feature dimension by combining pooling operations to enhance the feature extraction effect.

[0100] The role of the Neck is to fuse multi-level feature information to enhance the detection ability for targets of different sizes. It mainly includes an upsampling (Upsample) module, a Concat (feature concatenation) module, a RepNCSPELAN4 module, and convolutional operations. The upsampling module, such as nodes 11 and 14, upsamples the low-resolution feature map to a higher resolution for easy fusion with the shallow feature map. The feature concatenation module, such as nodes 12 and 15, concatenates feature maps at different levels to integrate high-level semantic information and low-level detail features. The RepNCSPELAN4 module, such as nodes 13 and 16, continues to process the concatenated feature map to extract more advanced features. The convolutional operations, such as nodes 17 and 20, adjust the number of channels to adapt to subsequent modules.

[0101] The role of Head is to complete the object detection task, including bounding box regression and object classification, and output the multi-scale structure of the feature map: the high-resolution branch (P3) outputs from the 33rd node and is used to detect small objects; the medium-resolution branch (P4) outputs from the 36th node and is used to detect medium-sized objects; the low-resolution branch (P5) outputs from the 37th node and is used to detect large objects. The core modules include CBLinear, CBFuse, conv-reg, conv-cls, and Concat. CBLinear and CBFuse, such as the 23rd - 30th nodes, are used to adjust the expressive ability of features in different branches and fuse information; conv-reg and conv-cls are used for bounding box regression (locating the object) and classification tasks (determining the category) respectively; Concat is used to integrate the final multi-scale detection results.

[0102] YOLOv9 adopts the efficient RepNCSPELAN4 module in the Backbone, enhancing the feature extraction ability; realizes multi-scale feature fusion through upsampling and concatenation in the Neck; and outputs multi-resolution feature maps in the Head to process objects of different sizes. In addition, YOLOv9 further optimizes the feature extraction and fusion process compared with previous models, improving the detection accuracy and speed, and is applicable to the classification and recognition scenario of impurity particles in power transformer oil.

[0103] Step 4, based on the trained YOLO object detection model, perform object detection on the preprocessed images, automatically identify different types of impurity particles, and automatically count the number of prediction boxes on all images, and output the number of particles of different types.

[0104] Combined with the weight file obtained by pre-training, the present invention can efficiently display various impurity particles such as fiber particles, carbon particles, metal particles, and bubbles recognized on the input pictures, automatically count the number of prediction boxes on all images, and output the number of particles of different types. In addition, the present invention can generate a detailed evaluation report according to the recognition results, providing an important reference for the condition monitoring and maintenance strategy of the transformer.

[0105] The present invention realizes the automated, precise, and real-time detection of impurity particles in transformer oil. By using the K-means++ algorithm to generate new prior boxes to meet the large deviation in the aspect ratio of some objects (such as fiber particles, etc.) in the dataset of the present invention from the anchor boxes of the standard dataset; then, through the object detection algorithm of YOLOv9, the accuracy of impurity particle classification is improved. The present invention solves the limitations of existing detection means in terms of classification types and detection efficiency, and thus can realize the evaluation of the operating state of the transformer and the retrospection of the cause of transformer faults, further improving the operating reliability of the power system.

[0106] Example 2

[0107] As Figure 4 shown, the difference between this embodiment and Embodiment 1 is that this embodiment provides a device for classifying and identifying impurity particles in transformer oil, and the device includes:

[0108] A flow-type dynamic imaging device, which is used to obtain impurity particle images in transformer oil in real time and use them as original images, and form a data set based on the original images;

[0109] An image processing module, which is used to preprocess the original images in the data set, filter out noise and optimize the image quality to obtain preprocessed images;

[0110] A deep learning recognition module, which is used to adopt the K-means++ algorithm to generate new prior boxes; according to the new prior boxes, train the YOLO object detection algorithm to obtain a trained YOLO object detection model; based on the trained YOLO object detection model, perform object detection on the preprocessed images, automatically identify different types of impurity particles, and automatically count the number of prediction boxes on all images, and output the number of different types of particles.

[0111] In this embodiment, the flow-type dynamic imaging device includes a liquid flow pump, pulsed light and a photosensitive element;

[0112] The liquid flow pump is used to control the flow of transformer oil;

[0113] The pulsed light is used to illuminate the area where the transformer oil flows through;

[0114] The photosensitive element is used to perform dynamic imaging on the particles in the flowing transformer oil.

[0115] The flow-type imaging device of the device of the present invention uses a precision syringe pump to regulate the oil flow, ensures the uniform flow of transformer oil in the microfluidic channel, and cooperates with a high-definition digital photosensitive element to achieve high-definition shooting of particles. The image processing module will augment the captured images and convert them into standard image data suitable for input to the deep learning model through processing steps such as image rotation, segmentation, denoising, and enhancement, thereby improving the recognition accuracy.

[0116] The operation flow chart of the device of the present invention is as Figure 5 shown.

[0117] Embodiment 3

[0118] As Figure 6 shown, the difference between this embodiment and Embodiment 1 is that the present invention further provides a system for classifying and identifying impurity particles in transformer oil, and the system includes:

[0119] A flow-type dynamic imaging system 1, which adopts the flow-type dynamic imaging device of Embodiment 1 and is used to obtain high-definition images of micron-sized impurity particles in transformer oil in real time;

[0120] An image post - processing computer 2, which is internally provided with the image - processing module and the deep - learning recognition module of Embodiment 1, is used to input the processed image through the image - processing module into the deep - learning recognition module for classification and recognition, so as to obtain impurity particle classification and quantity information.

[0121] A transformer oil sample container 3, which is used to provide a transformer oil sample for imaging by the flow - type dynamic imaging system;

[0122] A waste oil container 4, which is used to hold the transformer oil sample after being used by the flow - type dynamic imaging system.

[0123] In the above - mentioned technical solution, the transformer oil sample in the transformer oil sample container 3 of the present invention is pumped into the flow - type dynamic imaging system 1 by a liquid flow pump and then sent to the waste oil container 4. The images obtained by this system are sent to the image post - processing computer 2, and after being processed by the image - processing module, they are input into the target detection algorithm for classification and recognition.

[0124] As a further implementation, a report generation module is also provided in the image post - processing computer, which is used to generate a report on the impurity particle classification and quantity information recognized by the deep - learning recognition module to assist in evaluating the operating state of the transformer.

[0125] After the classification and recognition of the present invention are completed, a report containing impurity particle classification and quantity information is generated through the report generation module, which is used to evaluate the operating state of the transformer and provide an important reference for the condition monitoring and maintenance strategy of the transformer.

[0126] The flow - type dynamic imaging system 1 is as Figure 7 shown. The flow of the transformer oil is controlled by a liquid flow pump. A pulsed light source 5 is used to illuminate the area through which the transformer oil flows. A high - definition digital photosensitive element (such as a high - definition CMOS camera 6) is used to perform dynamic imaging on the particles (i.e., the test sample 30) in the flowing transformer oil through a zoom lens 7. The lens adopts an automatic zoom technology to ensure that particles within different particle size ranges can be clearly imaged. The flow - type dynamic imaging system 1 can obtain high - definition images of all particles in the transformer oil.

[0127] A large number of high - definition images obtained from the flow - type dynamic imaging system 1 are transmitted to the image post - processing computer 2 for pre - processing, including rotating and cutting the images to achieve augmentation, denoising the images, and mosaic - ing to achieve enhancement.

[0128] Meanwhile, the present invention also provides a computer - readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method for classifying and recognizing impurity particles in transformer oil of Embodiment 1.

[0129] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0130] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0131] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0132] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0133] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for classifying and identifying impurity particles in transformer oil, characterized in that: The method includes: Acquire the impurity particle images in the transformer oil in real time and use them as the original images, and form a data set based on the original images; Preprocessing the original images in the data set to obtain preprocessed images; A K-means++ algorithm is used to generate a new priori frame; a YOLO target detection algorithm is trained according to the new priori frame to obtain a trained YOLO target detection model; the YOLO target detection algorithm uses a YOLOv9 network structure; Based on the trained YOLO target detection model, the preprocessed images are subjected to target detection, different types of impurity particles are automatically identified, the number of prediction boxes on all images is automatically counted, and the number of different types of particles is output.

2. The method for classifying and identifying foreign particles in transformer oil according to claim 1, characterized in that: The images of impurity particles in transformer oil are obtained in real time through the liquid flow dynamic imaging device, including: The flow of transformer oil is controlled by a liquid flow pump, pulse light is used to illuminate the area through which the transformer oil flows, and a photosensitive element is used to dynamically image the particles in the flowing transformer oil.

3. The method for classifying and identifying foreign particles in transformer oil according to claim 1, characterized in that: Preprocessing the original images in the data set to obtain preprocessed images includes: Adjusting the original image to a preset fixed size to obtain an adjusted image; The adjusted image is subjected to standardization processing, and each pixel value is normalized to the interval [0, 1] to obtain a standardized image; The standardized image is enhanced and cropped to obtain a preprocessed image.

4. The method for classifying and identifying foreign particles in transformer oil according to claim 1, characterized in that: Use K-means++ algorithm to generate new prior frames, including: A1, randomly select an anchor box from the data set as the initial cluster center c, and calculate the distance D(x) between the remaining anchor boxes in the data set and the initial cluster center c; A2, based on the distance D(x), calculate the probability P(x) of each anchor box being selected as the cluster center, and select the anchor box farther away from the current cluster center as the new cluster center; A3, repeat steps A1 and A2 until K initial cluster centers are selected; A4, calculate the distance between each anchor box and all K initial cluster centers, and assign the anchor box to the cluster center category with the closest distance to it; A5, update and calculate the cluster center of each category; A6: Repeat steps A4 and A5 until the cluster center no longer changes, and obtain K optimized cluster centers as new prior boxes.

5. The method for classifying and identifying foreign particles in transformer oil according to claim 4, characterized in that: The updated calculation formula in step A5 is: Among them, C i is the set of anchor boxes in the i-th cluster category, c i is the cluster center of this category, and x is the anchor box.

6. A device for classifying and identifying impurity particles in transformer oil, characterized in that: The device includes: A streaming dynamic imaging device is used to obtain images of impurity particles in transformer oil in real time and use them as original images to form a data set based on the original images; An image processing module, used for preprocessing the original images in the data set to obtain preprocessed images; The deep learning recognition module is used to generate a new priori frame by using the K-means++ algorithm; train the YOLO target detection algorithm according to the new priori frame to obtain a trained YOLO target detection model; perform target detection on the preprocessed image based on the trained YOLO target detection model, automatically identify different types of impurity particles, and automatically count the number of prediction frames on all images, and output the number of different types of particles.

7. The device for classifying and identifying foreign particles in transformer oil according to claim 6, characterized in that: The flow dynamic imaging device comprises a liquid flow pump, a pulse light and a photosensitive element; The liquid flow pump is used to control the flow of transformer oil; The pulse light is used to illuminate the area through which the transformer oil flows; The photosensitive element is used to dynamically image particles in the flowing transformer oil.

8. The classification and identification system of impurity particles in transformer oil is characterized by: The system includes: A streaming dynamic imaging system, using the streaming dynamic imaging device in the device for classifying and identifying impurity particles in transformer oil as claimed in any one of claims 6 to 7, for acquiring images of impurity particles in transformer oil in real time; An image post-processing computer, which is internally provided with an image processing module and a deep learning recognition module in the device for classifying and identifying impurity particles in transformer oil as described in any one of claims 6 to 7, and is used for inputting the image processing module into the deep learning recognition module for classification and recognition to obtain the classification and quantity information of the impurity particles; Transformer oil sample container, used to provide transformer oil sample for imaging by flow dynamic imaging system; Waste oil container, used to hold transformer oil samples after being used by the flow dynamic imaging system.

9. The system for classifying and identifying foreign particles in transformer oil according to claim 8, characterized in that: The image post-processing computer is also provided with a report generation module; The report generation module is used to generate a report based on the classification and quantity information of the impurity particles identified by the deep learning recognition module to assist in evaluating the operating status of the transformer.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for classifying and identifying foreign particles in transformer oil according to any one of claims 1 to 5 is implemented.

Images