Method and equipment for detecting particulate matters in transformer oil, medium and product

Abstract

The invention discloses a method and device for detecting particulate matter in transformer oil, a medium and a product, and relates to the field of power transformer detection.The method comprises the steps that an image of the particulate matter in the transformer oil is obtained, and a multi-task segmentation data set is constructed; according to the multi-task segmentation data set, building a multi-task segmentation model by adopting a double-branch framework comprising segmentation branches and classification branches; the segmentation branch comprises a global attention up-sampling module, a mixing pool module, a feature pyramid attention module and a self-attention module; the classification branch comprises a full attention module; the multi-task segmentation model takes a particulate matter image as input and takes a detection result of the particulate matter image as output; obtaining a to-be-detected particulate matter image; and determining a detection result of the to-be-detected particulate matter image by adopting the trained multi-task segmentation model. According to the invention, accurate and efficient automatic identification and classification of particulate matters can be realized.

Description

Technical Field

[0001] The present application relates to the field of power transformer detection, and in particular to a method, equipment, medium and product for detecting particulate matter in transformer oil. Background Art

[0002] Transformer oil is a key insulating and cooling medium for power transformers. Its quality and purity have an important impact on the normal operation and service life of the equipment. During long-term operation, various tiny particles will be generated and accumulated in the insulating oil. These particles mainly come from the aging of insulating materials, wear of metal parts and the intrusion of external pollutants. Studies have shown that when particulate matter accumulates to a certain concentration in the oil, it may reduce the dielectric strength of the oil, increase the risk of partial discharge, and even accelerate the insulation aging of the equipment, eventually leading to transformer failure. Therefore, the accurate identification and quantitative detection of particulate matter in transformer oil has become an important task to ensure the stable operation of power equipment.

[0003] At present, particle detection technology mainly relies on traditional methods such as optical microscopy and laser particle size measurement. However, optical microscopy relies on manual operation, is highly subjective and has low processing efficiency, and is difficult to handle a large number of samples, affecting the accuracy and repeatability of detection. Although laser particle size measurement can provide particle size distribution information, it is not accurate enough when dealing with complex or composite particles. In addition, these traditional methods are difficult to meet the real-time requirements of transformer oil particle detection.

[0004] In recent years, microfluidic imaging technology has become an innovative means in the field of particle detection. It uses microfluidic systems to perform high-resolution imaging of liquid samples to achieve high-throughput, non-destructive real-time detection, and has shown broad application prospects in the fields of biomedicine and environmental monitoring. However, the application of this technology in transformer oil particle detection is still in the exploratory stage, and it still faces problems such as fuzzy image segmentation boundaries, complex and diverse particle morphology, and insufficient detection accuracy.

[0005] In order to solve the above problems, it is urgent to provide a method suitable for transformer oil particle detection, which can overcome the limitations of traditional technology in dealing with particle morphology diversity, size variation and segmentation edge fuzziness, improve detection accuracy and efficiency and reduce reliance on manual operation. Summary of the invention

[0006] The purpose of this application is to provide a method, device, medium and product for detecting particulate matter in transformer oil, which can realize accurate and efficient automatic identification and classification of particulate matter.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a method for detecting particulate matter in transformer oil, the method comprising:

[0009] Obtain images of particulate matter in transformer oil and construct a multi-task segmentation dataset;

[0010] According to the multi-task segmentation data set, a dual-branch framework including a segmentation branch and a classification branch is adopted to build a multi-task segmentation model; the segmentation branch is used to segment the particle image and identify the shape and boundary of the particle; the segmentation branch includes: a global attention upsampling module, a mixing pool module, a feature pyramid attention module and a self-attention module; the classification branch is used to classify the particle type based on the image segmentation; the classification branch includes: a full attention module; the multi-task segmentation model takes the particle image as input and takes the detection result of the particle image as output;

[0011] Acquire an image of the particle to be detected; and use the trained multi-task segmentation model to determine the detection result of the image of the particle to be detected.

[0012] Optionally, the step of acquiring an image of particulate matter in transformer oil and constructing a multi-task segmentation dataset specifically includes:

[0013] Using microfluidics system to obtain images of particles in transformer oil;

[0014] Preprocessing the image of particles in transformer oil; the preprocessing includes: data enhancement processing and fuzzy processing;

[0015] Based on the preprocessed images, a multi-task segmentation dataset is constructed.

[0016] Optionally, the skeleton network of the multi-task segmentation model adopts a resnet50 deep residual network; in the resnet50 deep residual network, convolutions in five stages adopt hollow convolutions.

[0017] Optionally, in the multi-task segmentation model, the low-order features output by stage one, stage two and stage three are input into the global attention upsampling module, and the high-order features output by stage four are input into the mixing pool module to extract local information, and are respectively input into the global attention upsampling module with the low-order features output by stage one, stage two and stage three, and the local information is output to the feature pyramid attention module at the same time; the high-order features output by stage five are passed through the full attention module to extract global information, and the classification results are output, and a global classification judgment is performed based on the classification results; the features output by the self-attention module and the global information output by the full attention module are upsampled layer by layer and fused to generate the detection results of the particle image.

[0018] Optionally, the full attention module adopts a dual attention mechanism.

[0019] Optionally, the multi-task segmentation model is trained using a stochastic gradient descent algorithm with momentum.

[0020] Optionally, the loss function of the multi-task segmentation model is a boundary-sensitive balanced loss function;

[0021] The boundary sensitive balance loss function BABL is determined using the formula BABL = α·DPCE Loss + β·Sensitivity-Specificity Loss;

[0022] Among them, DPCE Loss = -∑ i (y i log(p i )+(1-y i )log(1-p i ))·Distance(i), DPCE Loss is the distance graph penalty cross entropy loss function, Sensitivity-Specificity Loss is the sensitivity-specificity loss function, y i is the true label of the i-th pixel, p i is the predicted probability of the i-th pixel, Distance(i) is the distance from the i-th pixel to the target boundary; TP, FN and FP are the true positive, false positive and false positive numbers of the multi-task segmentation model respectively; α and β are numbers ranging from 0 to 1.

[0023] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for detecting particulate matter in transformer oil.

[0024] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the method for detecting particulate matter in transformer oil when executed by a processor.

[0025] In a fourth aspect, the present application provides a computer program product, including a computer program, which implements the method for detecting particulate matter in transformer oil when executed by a processor.

[0026] According to the specific embodiments provided in this application, this application has the following technical effects:

[0027] The present application provides a method, device, medium and product for detecting particulate matter in transformer oil. By acquiring images of particulate matter in transformer oil and constructing a multi-task segmentation data set, a unified and effective training data set of particles in oil is constructed, and the semantic segmentation technology in the multi-task segmentation model is used to perform pixel-level classification on microflow imaging images, thereby realizing automatic identification and classification of particulate matter. The present application can overcome the limitations of traditional technologies in dealing with the diversity of particle morphology, size changes and blurred segmentation edges, and can also improve the accuracy and efficiency of detection and reduce reliance on manual operations. The present application can provide a more intelligent and efficient solution for online monitoring and fault warning of power equipment, which helps to improve the long-term stability and safety of transformers. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 This is a flow chart of a method for detecting particulate matter in transformer oil in one embodiment of the present application;

[0030] Figure 2 This is an overall flow chart of a method for detecting particulate matter in transformer oil in one embodiment of the present application;

[0031] Figure 3 This is a structural diagram of a multi-task segmentation model in a method for detecting particulate matter in transformer oil in one embodiment of the present application;

[0032] Figure 4 Schematic diagram of the mixed pool module in the multi-task segmentation model;

[0033] Figure 5 Schematic diagram of the global attention upsampling module in the multi-task segmentation model;

[0034] Figure 6 Schematic diagram of the feature pyramid attention module in the multi-task segmentation model;

[0035] Figure 7 Schematic diagram of the full attention module in the multi-task segmentation model;

[0036] Figure 8 Schematic diagram of the self-attention module in the multi-task segmentation model. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0038] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0039] In an exemplary embodiment, Figure 1 and Figure 2 As shown, a method for detecting particulate matter in transformer oil is provided, the method comprising the following S101 to S103. Wherein:

[0040] S101, obtaining images of particles in transformer oil and constructing a multi-task segmentation dataset;

[0041] S101 specifically includes:

[0042] S1, using the microfluidic system to obtain images of particles in transformer oil; the microfluidic system is applied to the detection of particles in transformer oil based on the existing microfluidic imaging technology to achieve real-time and non-destructive detection of particles;

[0043] S2, preprocessing the image of the particles in the transformer oil; the preprocessing includes: data enhancement processing and fuzzy processing;

[0044] Data enhancement processing is performed on the particle image in transformer oil, including horizontal flipping, vertical flipping, rotation, translation, scaling, cropping, erasing, adjusting contrast, saturation and color jittering, and a synthetic dataset is created by combining the real and simulated particle features.

[0045] When adding difficult samples, we first target particles with similar morphology by increasing the proportion of samples and marking them as Similar_Particles_ <type>(e.g. Similar_Particles_Carbon_Metal) to enhance the multi-task segmentation model’s ability to identify subtle differences between similar particles. Secondly, some small-sized particle images (a few microns in size) are randomly distributed in the image and marked as Tiny_Particles_ <size>(such as Tiny_Particles_1um) to improve the detection sensitivity of the multi-task segmentation model for tiny particles. In addition, the Gaussian blur technology is used to blur the boundaries of some particles to generate samples with unclear edges, named Blurred_Edge_ <degree>(such as Blurred_Edge_High), which can improve the accuracy of segmentation when dealing with particles with blurred boundaries.

[0046] S3, constructs a multi-task segmentation dataset based on the preprocessed images.

[0047] The generated synthetic granular data and difficult samples are added to the multi-task segmentation dataset for further data enhancement and annotation, and the dataset format is unified to ensure the high consistency between the synthetic and difficult samples and the real samples, so as to construct the final multi-task segmentation dataset.

[0048] S102, based on the multi-task segmentation dataset, a dual-branch framework including a segmentation branch and a classification branch is used to build a multi-task segmentation model, and Figure 3 As shown; the segmentation branch is used to segment the particle image and identify the shape and boundary of the particles; the segmentation branch includes: a global attention upsampling module, a mixing pool module, a feature pyramid attention module and a self-attention module; the classification branch is used to classify the particle type based on the segmentation to distinguish different types of particles; the classification branch includes: a full attention module; the multi-task segmentation model takes the particle image as input and takes the detection result of the particle image as output;

[0049] In the segmentation branch, the image features output by the skeleton network are used to guide the segmentation feature learning, and the image features include high-order features and low-order features; in the classification branch, the input high-order features are used as auxiliary discrimination information to classify and solve the problem of confusion items. The classification task is completed through the network framework of the segmentation and classification dual branches, and finally the identification of particles in insulating oil is realized.

[0050] The skeleton network of the multi-task segmentation model adopts the resnet50 deep residual network; in the resnet50 deep residual network, the convolutions of the five stages adopt hollow convolutions.

[0051] In the multi-task segmentation model, the low-order features output from stage one, stage two and stage three are input into the global attention upsampling module, and the high-order features output from stage four are input into the mixed pool module to extract local information. They are respectively input into the global attention upsampling module with the low-order features output from stage one, stage two and stage three, and the local information is output to the feature pyramid attention module at the same time. The high-order features output from stage five are passed through the full attention module to extract global information, and the classification results are output. Global classification judgment is performed based on the classification results. The features output from the self-attention module and the global information output from the full attention module are upsampled layer by layer and fused to generate the detection results of the particle image.

[0052] The global attention upsampling module achieves global context weighting of features by combining high-order and low-order features. First, the low-order features extract local information through a 3×3 convolution layer, and the high-order features are directly introduced into the global attention upsampling module through jump connections. Then, the high-order features extract global context information through global pooling and are weighted with the low-order features after 1×1 convolution through point-by-point multiplication. By guiding the weighted calculation of low-order features through high-order features, local details can not only be interpreted in a broader context, but also effectively filter out possible background noise. Finally, the weighted low-order features are added to the high-order features to integrate global and local useful information, thereby enhancing the feature representation ability, improving the expression of target features and reducing the interference of background noise on the segmentation task.

[0053] like Figure 4 As shown in Figure 2, the hybrid pool module combines multi-scale features to enhance the expressiveness of the multi-task segmentation model by processing short-term and long-term dependencies at the same time. The hybrid pool module is divided into two branches, which process Figure 4 The short-term dependence shown in part (a) and Figure 4 The long-term dependency shown in part (b) of the network; the short-term dependency input feature x1 is extracted through two 2D convolutions of different scales to extract fine-grained local features and generate feature maps of different resolutions. The low-resolution features are then upsampled to the same size as the input features through an upsampling (UP) operation and further processed through 2D convolution to capture local dependency information, thereby accurately distinguishing local details and solving the morphological differences of similar objects. The long-term dependency input feature x2 extracts long-term dependency features of different dimensions through 1D convolution, processing the rows and columns of the space separately. Subsequently, the original feature size is restored through an upsampling operation, and the fused long-term dependency features are further integrated through 2D convolution to ensure that the global morphological differences are captured. Finally, the features generated by the two branches are extracted through multi-scale convolutions to integrate the short-term and long-term dependency information in the space. Through this structure, short-term dependency retains local details, and long-term dependency enhances the capture of global morphological features. Finally, the multi-scale features generated by the two are integrated to solve the problem of inter-class similarity. For example, although carbon particles, metal particles, and bubbles are similar in morphology, through this local and global multi-scale fusion, the multi-task segmentation model can better distinguish their subtle differences, enhance the ability to distinguish various types of objects, and reduce the occurrence of misclassification.

[0054] like Figure 6 As shown in the figure, the feature pyramid attention module enhances the feature extraction capability of the multi-task segmentation model by fusing global information and multi-scale local features. The input features are first reduced in dimension by 1×1 convolution, and then processed in two paths: in the global branch, the input is subjected to global pooling to extract global context information and capture the overall semantic structure, and then the dimension is reduced by 1×1 convolution and upsampled back to the original resolution; in the multi-scale branch, the input is subjected to 7×7, 5×5 and 3×3 convolution to extract local features of different scales (16×16, 8×8, 4×4), and generate multi-scale feature maps to capture local morphology and size changes. Next, the local features of each scale are point-by-point multiplied with the global features (i.e., weighted fusion), so that local details can be re-weighed in the global context. In this way, the global context information provides guidance for the detailed expression of local features, helping the multi-task segmentation model to better understand the diversity of local morphology. Finally, all fused multi-scale features are added to the global features through an addition operation, and a feature map containing both global dependencies and multi-scale details is output. This combination enables the multi-task segmentation model to effectively deal with the differences in morphology and size of the same particles, ensuring that even if the same type of particles have great variations in shape, size, etc., the multi-task segmentation model can still accurately identify and classify through global and local multi-scale information integration, thereby improving the robustness of classification and segmentation.

[0055] like Figure 5 As shown in the figure, the global attention upsampling module first receives the feature maps from each stage of ResNet50, extracts the global context information, and restores the feature maps of each stage to the same resolution as the previous stage through upsampling operations. Multi-scale features are fused together to enhance the global dependency in the feature map, and these fused features are input to the next layer through point-by-point addition. The feature map of a specific stage is also processed by the pyramid attention module, which extracts context information of different spatial scales through multi-scale convolution and weightedly fuses these multi-scale features with global information to enhance the ability to capture different spatial details. The features output by the global attention upsampling module and the pyramid attention module are fused together through addition operations to form a feature map containing global context information and multi-scale features. The fused feature map is input into the self-attention module as input. In the self-attention module, high-order semantic information is further extracted to capture key details in the input features, especially the significant features of the image boundary, and strengthen the attention to image details. The features processed by the self-attention module combine global, local and high-order semantic information to improve the clarity and accuracy of the segmentation edge. Ultimately, the multi-level and multi-scale fusion processing enables the multi-task segmentation model to effectively solve the problem of blurred edges and achieve more accurate boundary segmentation.

[0056] like Figure 8 As shown, the self-attention module generates the final output feature 4 by integrating information from the global attention upsampling module and the pyramid attention module (features 1, 2, and 3). First, features 1, 2, and 3 are respectively subjected to 1×1 convolution for dimensionality reduction and feature extraction, and then the features are converted into a shape suitable for matrix multiplication through dimensional transformation. Next, the attention weights are calculated between features 1 and 2 through matrix multiplication (i.e., point-by-point multiplication) to obtain the correlation between the two and generate a spatial weight matrix. This weight matrix is ​​multiplied point-by-point with feature 3 to form a weighted feature representation. After that, the result of the dimensional transformation operation is weighted again with feature 3, and finally the output result is fused with features 1, 2, and 3, and the final feature 4 is obtained through addition operation. Through the self-attention mechanism, a weight matrix of spatial dependency is generated based on the correlation between the input features, so that different features are weighted and fused, the correlation between features and the global perception ability are enhanced, and finally an output feature with more semantic expression ability is generated.

[0057] like Figure 7 As shown in the figure, the full attention module adopts a dual attention mechanism. The full attention module includes a position attention submodule and a channel attention submodule to enhance the feature expression capability. First, the input image is subjected to preliminary feature extraction through ResNet to generate a feature map of size C×H×W, which is input into the position attention and channel attention modules respectively. In the position attention module, the feature map is reshaped, and the correlation between different spatial positions is calculated by matrix multiplication to generate a spatial attention matrix, thereby capturing the dependency of each spatial position in the image. This further helps the multi-task segmentation model to better understand the spatial distribution differences of particles and perform weighted fusion with the original features to enhance local feature information. In the channel attention module, the feature map is also reshaped, and the correlation between different spatial positions is calculated by matrix multiplication to generate a spatial attention matrix, thereby capturing the dependency of each spatial position in the image. The dependency between channels is calculated by matrix multiplication to generate a channel attention matrix, which is then weighted fused with the original feature map to output enhanced channel features. Through the dual attention mechanism, the multi-task segmentation model can capture the differences in space and channels at the same time, especially in particles with similar morphology, and more accurately identify small local feature differences. Finally, the feature maps of the position attention and channel attention modules are added element by element to generate a feature map containing spatial context and channel dependency information. The fused feature map enhances the multi-task segmentation model's ability to perceive subtle differences between particles, enabling the multi-task segmentation model to more accurately distinguish different types of particles, thereby improving the accuracy of recognition and classification.

[0058] The features output by the self-attention module and the global information output by the full attention module are fused through a layer-by-layer upsampling process to finally generate the segmentation result. First, the features output by the self-attention module have obtained enhanced representation of global and local semantic information through the position attention and channel attention mechanisms, while the full attention module enhances the representation of the overall features by capturing the global context information of the input image. Next, the output features of the two modules are upsampled before fusion to gradually restore them to the same resolution as the input image. This is usually achieved through upsampling techniques such as bilinear interpolation or deconvolution, so that the size of the feature map increases layer by layer. During the upsampling process, the resolution of the feature map is continuously improved while retaining the semantic information of the features. The upsampled self-attention features and full-attention features are combined by element-by-element addition or cascade fusion to ensure that the global semantics and local detail information of both can be effectively integrated. The fused feature map is then processed by a 1×1 convolution layer and a sigmoid activation function to output the final segmentation result.

[0059] The multi-task segmentation model is trained using a stochastic gradient descent algorithm with momentum. The training process specifically includes:

[0060] Use the images in the preprocessed semantic segmentation image dataset as input images and unify the size of the input images;

[0061] The semantic segmentation network multi-task segmentation model is trained using a stochastic gradient descent algorithm with momentum, wherein the weight initialization method in the semantic segmentation network multi-task segmentation model is Kaiming and the activation function is ReLu;

[0062] The loss function is used to predict the difference between the actual data, and the loss function is a boundary-aware balanced loss function (BABL).

[0063] The boundary sensitive balance loss function BABL is determined using the formula BABL = α·DPCE Loss + β·Sensitivity-Specificity Loss;

[0064] Among them, DPCE Loss = -∑ i (y i log(p i )+(1-y i )log(1-p i ))·Distance(i), DPCE Loss is the distance graph penalty cross entropy loss function, Sensitivity-Specificity Loss is the sensitivity-specificity loss function, y i is the true label of the i-th pixel, p i is the predicted probability of the i-th pixel, Distance(i) is the distance from the i-th pixel to the target boundary; TP, FN and FP are the true positive, false positive and false positive numbers of the multi-task segmentation model, respectively, and are shown in Table 1; α and β are numbers ranging from 0 to 1.

[0065] Table 1

[0066] Predict the positive class (Positive) Predict negative class (Negative) Positive True Positive (TP) False Negative (FN) Actual negative class (Negative) False Positive (FP) True Negative (TN)

[0067] S103, obtaining an image of the particle to be detected; and using the trained multi-task segmentation model to determine a detection result of the image of the particle to be detected.

[0068] The present application uses a multi-task segmentation model to combine segmentation branches and classification branches, and uses high-order and low-order features to accurately segment and classify particle morphology, thereby solving the problem of poor recognition accuracy in the prior art. In particular, through the application of the global attention upsampling module, the mixed pool module and the feature pyramid attention module, the capture of local details and global information is enhanced, ensuring high-precision recognition and classification of particles. The capture and optimization of image boundaries are enhanced through the self-attention module and the dual attention mechanism. Combining global, local and multi-scale features, the clarity and accuracy of image boundaries are improved, and the problem of blurred particle boundaries can be effectively handled, thereby generating more accurate segmentation results. By combining global context information and multi-scale fusion of local features, the multi-task segmentation model can solve the confusion problem caused by the similarity of particle morphology and improve the classification ability of different types of particles. At the same time, the method effectively copes with the variability of the same type of particles in morphology and size through the feature pyramid attention module and the mixed pool module, ensuring that even if the particle shape and size are quite different, accurate classification and segmentation can be achieved. By constructing a unified and effective training data set for particles in oil, the accuracy and robustness of particle recognition are significantly improved. The dataset uses microfluidic imaging technology to collect real particle images, and expands its diversity through synthetic data and difficult samples. Combined with data enhancement technology, the multi-task segmentation model can adapt to changes in particles of various shapes and sizes. Through the unified and standardized dataset design, the problems of data scarcity and low recognition accuracy in the existing technology are solved, and the multi-task segmentation model is enhanced in the ability to distinguish between similarities between classes and differences within classes, ensuring more accurate particle recognition and segmentation effects. By using the boundary-sensitive balanced loss function, the training process not only focuses on the overall segmentation accuracy, but also finely optimizes the segmentation boundaries. The loss function introduces distance graph penalty cross entropy and sensitivity-specificity loss, which further improves the sensitivity of the multi-task segmentation model to particle boundaries and reduces mis-segmentation and misclassification.

[0069] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for detecting particulate matter in transformer oil is implemented.

[0070] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0071] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0072] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0073] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0074] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0075] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0076] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.< / degree> < / size> < / type>

Claims

1. A method for detecting particulate matter in transformer oil, characterized in that: The method for detecting particulate matter in transformer oil comprises: Obtain images of particulate matter in transformer oil and construct a multi-task segmentation dataset; According to the multi-task segmentation data set, a dual-branch framework including a segmentation branch and a classification branch is adopted to build a multi-task segmentation model; the segmentation branch is used to segment the particle image and identify the shape and boundary of the particle; the segmentation branch includes: a global attention upsampling module, a mixing pool module, a feature pyramid attention module and a self-attention module; the classification branch is used to classify the particle type based on the image segmentation; the classification branch includes: a full attention module; the multi-task segmentation model takes the particle image as input and takes the detection result of the particle image as output; Acquire an image of the particle to be detected; and use the trained multi-task segmentation model to determine the detection result of the image of the particle to be detected.

2. The method for detecting particulate matter in transformer oil according to claim 1, characterized in that: The method of obtaining the image of particulate matter in transformer oil and constructing a multi-task segmentation dataset specifically includes: Using microfluidics system to obtain images of particles in transformer oil; Preprocessing the image of particles in transformer oil; the preprocessing includes: data enhancement processing and fuzzy processing; Based on the preprocessed images, a multi-task segmentation dataset is constructed.

3. The method for detecting particulate matter in transformer oil according to claim 1, characterized in that: The skeleton network of the multi-task segmentation model adopts the resnet50 deep residual network; in the resnet50 deep residual network, the convolutions of the five stages adopt hollow convolutions.

4. The method for detecting particulate matter in transformer oil according to claim 3, characterized in that: In the multi-task segmentation model, the low-order features output from stage one, stage two and stage three are input into the global attention upsampling module, and the high-order features output from stage four are input into the mixed pool module to extract local information. They are respectively input into the global attention upsampling module with the low-order features output from stage one, stage two and stage three, and the local information is output to the feature pyramid attention module at the same time. The high-order features output from stage five are passed through the full attention module to extract global information, and the classification results are output. Global classification judgment is performed based on the classification results. The features output from the self-attention module and the global information output from the full attention module are upsampled layer by layer and fused to generate the detection results of the particle image.

5. The method for detecting particulate matter in transformer oil according to claim 1 or claim 3, characterized in that: The full attention module adopts a dual attention mechanism.

6. The method for detecting particulate matter in transformer oil according to claim 1, characterized in that: The multi-task segmentation model is trained using a stochastic gradient descent algorithm with momentum.

7. The method for detecting particulate matter in transformer oil according to claim 1, characterized in that: The loss function of the multi-task segmentation model is a boundary sensitive balanced loss function; The boundary sensitive balance loss function BABL is determined using the formula BABL = α·DPCE Loss + β·Sensitivity-Specificity Loss; Among them, DPCE Loss = -∑ i (y i log(p i )+(1-y i )log(1-p i ))·Distance(i), DPCE Loss is the distance graph penalty cross entropy loss function, Sensitivity-SpecificityLoss is the sensitivity-specificity loss function, y i is the true label of the i-th pixel, p i is the predicted probability of the i-th pixel, Distance(i) is the distance from the i-th pixel to the target boundary; TP, FN and FP are the true positive, false positive and false positive numbers of the multi-task segmentation model respectively; α and β are numbers ranging from 0 to 1.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for detecting particulate matter in transformer oil according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting particulate matter in transformer oil according to any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for detecting particulate matter in transformer oil according to any one of claims 1 to 7 is implemented.

Images