Oil leakage monitoring method and system for oil-immersed transformer equipment based on image recognition

Abstract

The present invention relates to the field of image processing, and more specifically, to an oil-immersed transformer equipment oil leakage monitoring method and system based on image recognition, the method comprising: obtaining multiple consecutive grayscale frames of the transformer surface; calculating the LBP value of each pixel in all grayscale images, and obtaining the difference value of the LBP value at the same position in two consecutive grayscale frames to generate a difference map; calculating the edge performance of each difference map; selecting any one of the difference maps as a target map, and calculating the abnormal performance of the target map; calculating the defect degree of the difference map based on the edge performance and the abnormal performance, and triggering an oil leakage warning mechanism when the defect degree is greater than a preset defect threshold. The present invention accurately evaluates the oil leakage on the transformer surface, which not only improves the efficiency and accuracy of monitoring, but also provides a strong technical guarantee for the safe operation of the transformer.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and more specifically, to an oil leakage monitoring method and system for oil-immersed transformer equipment based on image recognition. Background Art

[0002] Oil-immersed transformers are core equipment in power transmission and distribution systems and are used in power plants, substations, and industrial power systems. The insulating oil filled inside plays two main roles: insulation and heat dissipation. However, during long-term use, some faults may occur, among which oil leakage is a common problem. Oil leakage not only causes the loss of transformer oil, but also may affect the insulation performance and cooling effect of the transformer, thereby threatening the safe and stable operation of the power system. Therefore, it is particularly important to monitor oil leakage in oil-immersed transformers.

[0003] Although traditional oil leakage monitoring technologies such as infrared imaging technology, contact sensors or manual inspections can monitor oil leakage to a certain extent, they still have some limitations, such as limited ability to detect small leaks, cost issues and low efficiency. With the development of image processing and artificial intelligence technology, oil leakage monitoring technology based on image recognition has emerged. It uses a camera to collect images of the transformer surface in real time, and uses advanced image processing algorithms (such as local binary patterns, LBP) to analyze image texture changes, thereby achieving accurate monitoring of oil leakage.

[0004] However, since oil-immersed transformers are usually deployed in outdoor environments, image acquisition of their surfaces is easily affected by external lighting conditions. As a result, during the oil leakage monitoring process, texture changes caused by lighting changes are misjudged as oil leaks (false alarms) or oil leaks are mistaken for normal lighting changes (missed alarms), which makes it impossible to detect and handle them in time, resulting in safety failures. Summary of the invention

[0005] In order to solve the above-mentioned technical problem of how to accurately extract the surface texture features of the transformer and improve the safety of oil leakage monitoring, the present invention provides solutions in the following aspects.

[0006] In a first aspect, a method for monitoring oil leakage of an oil-immersed transformer device based on image recognition comprises:

[0007] Acquire a continuous multi-frame grayscale image of the transformer surface;

[0008] Calculate the LBP value of each pixel in all grayscale images, and obtain the difference value of the LBP value at the same position in two consecutive grayscale images to generate a difference image; calculate the edge performance of each difference image;

[0009] Select any difference image as the target image, calculate the abnormal performance of the target image, and the abnormal performance is positively correlated with the difference degree of the target image relative to the target set of the target image as a whole, and negatively correlated with the texture uniformity inside the target image; wherein, starting from the target image, a set number of difference images are obtained forward as the target set of the target image;

[0010] The defect degree of the difference image is calculated based on the edge performance and the abnormal performance, and when the defect degree is greater than a preset defect threshold, an oil leakage early warning mechanism is triggered.

[0011] Since illumination changes are usually relatively stable in continuous frames, and oil leakage can cause dynamic texture changes, the present invention first calculates the LBP value of continuous multiple frames of grayscale images and their difference maps, which can effectively capture the texture changes and dynamic features of the transformer surface, and then filter out static illumination changes. Due to the generation of the difference map, the changes in pixels in the image are highlighted, making tiny traces of oil leakage or abnormal textures more obvious; further, when analyzing the difference map, a comprehensive judgment is made in combination with edge performance and abnormal performance, wherein the edge performance can highlight the contour features of the oil leakage area, while the abnormal performance is compared with the overall difference and internal texture uniformity of the target set to avoid misjudging the texture caused by illumination changes as oil leakage. Therefore, the defect degree calculated based on the edge performance and abnormal performance can accurately evaluate the oil leakage on the transformer surface, which not only improves the efficiency and accuracy of monitoring, but also provides a strong technical guarantee for the safe operation of the transformer.

[0012] Preferably, the process of obtaining the edge representation includes:

[0013] The frequency of occurrence of each LBP difference value in the difference graph is counted, and the frequency of occurrence of all LBP difference values ​​multiplied by the negative value of their logarithm is summed to obtain the information entropy of the difference graph;

[0014] The LBP difference mean in the difference map is calculated, and the product of the information entropy of the difference map and the LBP difference mean is normalized to obtain the edge representation of the difference map.

[0015] By counting the frequency of each LBP difference value in the difference map and calculating the information entropy, the complexity and unevenness of the image texture can be quantified. The higher the information entropy, the more complex the image texture and the richer the edge information. Combining the information entropy with the LBP difference mean can further highlight the edge features, making them easier to detect and identify in subsequent processing.

[0016] Preferably, the process of obtaining the degree of difference includes:

[0017] Calculate the LBP difference mean of all difference maps in the target set; calculate the sum of the absolute difference values ​​of the LBP difference mean of the target map and the LBP difference mean of all other difference maps in the target set, and then divide it by the number of difference maps in the target set and normalize it to obtain the degree of difference of the target map relative to the target set as a whole.

[0018] By calculating the difference in the LBP differential mean, the difference between the target image and other images in the target set is quantified into a specific value. This operation can evaluate the "uniqueness" or "abnormality" of the target image in the entire target set. If the degree of difference is high, it means that the target image is very different from other images in the target set. Otherwise, if the degree of difference is low, it means that the target image is more similar to other images in the target set. At the same time, it can also avoid misjudgment caused by noise or abnormal features of individual images, and improve the overall robustness and reliability of the algorithm.

[0019] Preferably, the process of acquiring the texture uniformity includes:

[0020] The target image is divided into multiple windows according to the preset division parameters, and the sum of the squares of the differences between the LBP difference mean in all windows and the LBP difference mean of the target image is calculated, and then divided by the total number of windows to obtain the average square difference value, and the average square difference value is normalized to obtain the texture uniformity inside the target image.

[0021] Since noise usually causes mutations in LBP features, and this mutation will be amplified when calculating the sum of squared differences, therefore, by performing square and average operations on the differences, the influence of noise in the image can be suppressed to a certain extent, making the texture features more obvious.

[0022] Preferably, the abnormal performance satisfies the relationship:

[0023] ; In the formula, is the abnormal performance of the target graph, is the difference between the target graph and the target set as a whole, is the texture uniformity inside the target image; where is the total number of difference graphs in the target set except the target graph, is the LBP difference mean of the target image, For the The LBP difference mean of the difference graphs, represents normalization processing, is the total number of windows after the target graph is divided, For the The LBP difference mean within the window, Represents an exponential function with the natural constant e as the base.

[0024] By combining the difference between the target image and the target set as a whole and the uniformity of the texture inside the target image, the formula can more comprehensively evaluate the abnormal performance of the target image. If the target image is significantly different from the target set as a whole and the internal texture is uneven, the abnormal performance value will be larger, indicating that the target image is more likely to contain abnormal features. On the contrary, if the difference is small and the texture is uniform, the abnormal performance value will be smaller, indicating that the target image is relatively normal. By comprehensively considering external differences and internal uniformity, this method can detect anomalies more robustly and avoid misjudgment due to interference from a single feature.

[0025] Preferably, the process of obtaining the defect degree includes:

[0026] The defect degree is obtained by normalizing the product of the edge performance and the abnormal performance.

[0027] By combining edge and abnormal performance, the advantages of both can be fully utilized to improve the accuracy of defect detection. For example, in some cases, edge performance may be more likely to capture tiny defect edges, while abnormal performance may be more likely to identify larger defect areas. The combination of the two can make the detection process more comprehensive and accurate.

[0028] Preferably, the LBP difference means of all other difference maps in the target set also include:

[0029] Calculate the similarity scores between other difference graphs in the target set and the target graph, assign weights to the difference graphs according to the similarity scores, and obtain the weighted average LBP difference mean of other difference graphs in the target set.

[0030] By calculating the similarity scores between other difference graphs in the target set and the target graph, and assigning weights according to the similarity scores, the weighted average LBP difference mean of other difference graphs in the target set can be obtained. This weighting method makes the difference graphs that are more similar to the target graph occupy a greater weight in the calculation, making the final features more representative.

[0031] Preferably, after acquiring the grayscale image, smoothing and image enhancement are performed on the grayscale image.

[0032] Preferably, the process of obtaining the edge representation of each difference image includes:

[0033] The Canny edge detection algorithm is used to extract the edge pixels of the difference image, and the ratio of the number of edge pixels to the total number of pixels in the difference image is calculated as the edge representation of the difference image.

[0034] In a second aspect, an oil-immersed transformer equipment oil leakage monitoring system based on image recognition includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, any one of the oil-immersed transformer equipment oil leakage monitoring methods based on image recognition is implemented.

[0035] The beneficial effects of the present invention are:

[0036] The present invention combines edge performance and abnormal performance to evaluate the defect degree of the differential image. Edge performance quantifies the edge features of the image through information entropy and LBP differential mean, which helps to distinguish edge changes caused by oil leakage and edge changes caused by illumination changes; abnormal performance takes into account the difference degree of the target image relative to the target set as a whole and the texture uniformity within the target image, further improving the accuracy of oil leakage monitoring, and effectively solving the problem of false alarms and missed alarms caused by illumination changes during oil leakage monitoring, so as to timely discover and handle oil leakage and prevent the occurrence of safety failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a method flow chart of steps S1 to S4 in the oil leakage monitoring method for oil-immersed transformer equipment based on image recognition in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0039] Reference Figure 1 The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition includes steps S1 to S4, which are as follows:

[0040] S1: Acquire multiple continuous grayscale frames of the transformer surface.

[0041] In one embodiment, a camera with high resolution and good light adaptability is selected, and a suitable installation position is selected according to the structure and layout of the transformer to ensure that the camera can clearly capture the image of the target part.

[0042] The preset acquisition frequency is 1 Hz, that is, one image is collected per second. The acquisition locations include but are not limited to the transformer's oil tank, welds, flange joints, oil pumps, oil pipe joints, valves, transformer top and bottom areas and other oil leakage-prone locations.

[0043] Convert the RGB image captured by the camera into a grayscale image.

[0044] In addition, a filter (such as a Gaussian filter) can be used to smooth the grayscale image to reduce random noise introduced by the camera sensor, light changes, or transmission process.

[0045] Other image enhancement techniques such as adaptive histogram equalization, contrast stretching, etc. can also be applied to further improve the image quality.

[0046] S2: Calculate the LBP value of each pixel in all grayscale images, and obtain the difference value of the LBP value at the same position in two consecutive grayscale images to generate a difference image; calculate the edge performance of each difference image.

[0047] Local Binary Pattern (LBP) is a visual descriptor used for texture classification. Its basic idea is to compare each pixel with the pixels in its neighborhood, convert these comparison results into a binary number, and then convert this binary number into a decimal number as the LBP value of the pixel. This value reflects a local structural information of the texture around the pixel.

[0048] In one embodiment, each grayscale image is traversed to calculate the LBP values ​​of all pixels. For two consecutive frames of images, for example, the first frame of grayscale image and the second frame of grayscale image, the difference value of the LBP value of the pixel at the same position is calculated, and then all the difference values ​​of the first frame of grayscale image and the second frame of grayscale image are combined into a new image, that is, a difference image is generated.

[0049] The continuous frame images such as the second frame grayscale image and the third frame grayscale image are subjected to the same operation as the first frame grayscale image and the second frame grayscale image, thereby obtaining a plurality of differential images.

[0050] Since oil leakage in oil-immersed transformers usually produces obvious edges in the image, by measuring the distribution of all LBP difference values ​​in the difference map and the difference between these difference values ​​and the difference mean, if the LBP difference values ​​in a difference map are widely distributed and have a large difference from the average difference value, then the edge performance of the difference map will be strong. Conversely, if the LBP difference values ​​in the difference map are concentrated and close to the average difference value, then the edge performance of the difference map will be weak. Based on the edge performance, it is possible to evaluate whether the edge features in the grayscale image are significant.

[0051] Specifically, all the difference values ​​and their occurrence probabilities in the difference graph are counted, and the edge performance of the difference graph is calculated, that is, the relationship is satisfied:

[0052]

[0053] In the formula, For the The edge representation of the difference graph, For the In the difference map The probability of occurrence of LBP difference values, For the The total number of LBP difference values ​​in the difference map, For the The mean LBP difference in the difference map, Indicates normalization processing.

[0054] In the above formula, As the calculation formula of information entropy, it measures the distribution of LBP difference values. When the distribution is more uniform, the information entropy of the difference image is larger, which means that the texture or edge information in the grayscale image is more complex; conversely, when the distribution is more concentrated, the information entropy is smaller, which means that the texture or edge information in the grayscale image is simpler. The larger it is, the more obvious the grayscale change in the grayscale image is and the stronger the edge performance is. The larger the value of, the stronger the edge performance. Strong edge performance means that there is obvious texture or edge change in the image, which may be related to oil leakage.

[0055] According to the above The edge performance calculation formula of a difference graph can be used to calculate the edge performance of all other difference graphs in the same way.

[0056] It should be noted that when calculating the LBP value, a window needs to be determined, and this window defines the neighborhood range around the central pixel. The window is exemplarily set to a size of 3×3 or 4×4.

[0057] In another embodiment, after all the difference images are obtained, edge detection can be performed directly, such as using the Canny edge detection algorithm to extract edge pixels of the difference image, and calculating the ratio of the number of edge pixels to the total number of pixels in the difference image as the edge representation of the difference image.

[0058] S3: Select any differential image as the target image, and calculate the abnormal performance of the target image. The abnormal performance is positively correlated with the degree of difference of the target image relative to the overall target set of the target image, and negatively correlated with the texture uniformity within the target image; wherein, taking the target image as the starting point, a set number of differential images are obtained forward as the target set of the target image.

[0059] Taking into account the particularity of the deployment environment of oil-immersed transformers and the impact of lighting conditions on image acquisition, firstly, by comparing the current frame with its previous frames, the changing trend in the image that gradually evolves over time can be captured. Secondly, the difference map is divided into multiple windows of fixed size, and the impact of lighting changes on image analysis is reduced through local feature extraction, so as to more accurately determine whether the changes in the image are caused by oil leakage or lighting changes, thereby reducing the occurrence of false alarms and missed alarms.

[0060] In one embodiment, first, starting from the target graph, five difference graphs are forwardly acquired as a target set of target graphs, that is, a total of six difference graphs in the target set, and then the LBP difference means of all difference graphs in the target set are calculated, as well as the sum of the absolute difference values ​​between the LBP difference mean of the target graph and the LBP difference means of all other difference graphs in the target set. Next, the sum of the absolute difference values ​​is divided by the total number of difference graphs in the target set to obtain an average difference value, and the average difference value is normalized to obtain the degree of difference of the target graph relative to the target set as a whole.

[0061] Then the difference between the above target graph and the target set as a whole satisfies the relationship:

[0062]

[0063] In the formula, is the difference between the target graph and the target set as a whole, is the total number of difference graphs in the target set except the target graph, is the LBP difference mean of the target image, For the The LBP difference mean of the difference graphs, Indicates normalization processing.

[0064] In one embodiment, any differential image is selected as the target image, and the target image is divided into a plurality of square windows of a fixed size. For example, the side length of the window is set to 4 pixels, which means that the differential image is divided into 4×4 square areas.

[0065] In addition, for the processing of the boundary part, if the side length of the target image is not equal to a multiple of the side length of the window, the remaining part can usually be processed by filling or cropping to ensure that each window can completely cover a part of the target image.

[0066] Furthermore, for each divided window, the mean of all LBP difference values ​​inside it is calculated, and this mean reflects the local texture features inside the window area; then the difference between the LBP difference mean of each window and the LBP difference mean of the target image as a whole is calculated, and these differences are squared, and then all square differences are accumulated to obtain a total sum of squared differences, which reflects the degree of difference between each window and the overall texture features in the target image; in order to obtain an average difference measure, the sum of squared differences obtained in the previous step is divided by the total number of windows to obtain an average squared difference value, which represents the consistency of the texture features inside the target image. The smaller the value, the more uniform the internal texture, and vice versa, the larger the value, the greater the internal texture difference; finally, this average squared difference value is normalized to obtain the texture uniformity inside the target image.

[0067] Then the texture uniformity inside the target image satisfies the relationship:

[0068]

[0069] In the formula, is the texture uniformity inside the target image, is the total number of windows after the target graph is divided, For the The LBP difference mean within the window, is the LBP difference mean of the target image as a whole, Represents an exponential function with the natural constant e as the base.

[0070] After obtaining the difference between the target image and the target set as a whole and the texture uniformity inside the target image, the product of the difference between the target image and the target set as a whole and the texture uniformity inside the target image is taken as the abnormal performance of the target image. That is, the relationship is satisfied:

[0071]

[0072] In the formula, is the abnormal performance of the target graph, is the difference between the target graph and the target set as a whole, is the texture uniformity inside the target image; where is the total number of difference graphs in the target set except the target graph, is the LBP difference mean of the target image, For the The LBP difference mean of the difference graphs, represents normalization processing, is the total number of windows after the target graph is divided, For the The LBP difference mean within the window, Represents an exponential function with the natural constant e as the base.

[0073] By dividing the difference map into multiple fixed-size windows and extracting local texture features (i.e., LBP differential mean), the impact of illumination changes on overall image analysis can be effectively reduced. This is because local features are less sensitive to illumination changes, thereby more accurately determining that the changes are caused by oil leaks rather than illumination changes. By comparing the current frame (target map) with its previous frames, the trend of changes in the image that gradually evolve over time can be captured, because oil leaks usually do not occur suddenly, but gradually appear over time. By combining time series analysis and local texture feature extraction, the impact of illumination changes can be effectively reduced, the evolution trend of anomalies can be captured, and the accuracy and reliability of anomaly detection can be improved.

[0074] According to the above operation of calculating the abnormal performance of the target graph, the abnormal performance of all other difference graphs can be obtained in the same way.

[0075] In another embodiment, in the operation of calculating the difference degree of the target image relative to the target image of the target set as a whole, an image similarity measurement method (such as SSIM, MSE, etc.) is used to calculate the similarity between other difference images in the target set and the target image, and the difference images are divided into different levels according to the similarity score, and a weight is assigned to each level. For example, a difference image with a higher similarity score is assigned a greater weight. After determining the weight of each difference image, the weighted average LBP difference mean of all other difference images in the target set is calculated, that is, the relationship is satisfied:

[0076]

[0077] In the formula, is the first The weighted average LBP difference mean of the difference maps, is the total number of difference graphs in the target set except the target graph, For the The LBP difference mean of the difference graphs, For the The weight of the difference graph.

[0078] By assigning a weight to each difference map, we can more flexibly control which difference maps have a greater influence when calculating the LBP difference mean.

[0079] S4: Calculating the defect degree of the difference image based on the edge performance and the abnormal performance, and triggering the oil leakage warning mechanism when the defect degree is greater than a preset defect threshold.

[0080] In one embodiment, the product of the edge performance and the abnormal performance of the difference image calculated by S2 and S3 is normalized to obtain the defect degree of the difference image, that is, the relationship is satisfied:

[0081]

[0082] In the formula, Indicates The defect degree of a difference map; Indicates The edge representation of the difference graph, Indicates The abnormal performance of the difference graph, Represents the standard normalization function.

[0083] The above operation combines two key indicators, edge performance and abnormal performance, to more comprehensively and accurately evaluate the degree of defects in the differential image. Edge performance usually reflects the structural information in the image, while abnormal performance reveals changes in the image that are different from the normal pattern. By combining these two indicators, more subtle defect features can be captured and the accuracy of the evaluation can be improved.

[0084] Furthermore, the defect threshold is set to 0.8. When the defect degree of the differential image is greater than the defect threshold, the oil leakage warning mechanism is triggered, so that the staff can take necessary measures to prevent the oil leakage from further deteriorating.

[0085] The system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to the first aspect of the present invention is implemented.

[0086] The system also includes other components familiar to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0087] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these modifications and improvements all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An oil-immersed transformer equipment oil leakage monitoring method based on image recognition, characterized in that: include: Acquire a continuous multi-frame grayscale image of the transformer surface; Calculate the LBP value of each pixel in all grayscale images, and obtain the difference value of the LBP value at the same position in two consecutive grayscale images to generate a difference map; Calculate the edge performance of each difference map; Select any difference image as the target image, calculate the abnormal performance of the target image, and the abnormal performance is positively correlated with the difference degree of the target image relative to the target set of the target image as a whole, and negatively correlated with the texture uniformity inside the target image; wherein, starting from the target image, a set number of difference images are obtained forward as the target set of the target image; Calculating the defect degree of the difference image based on the edge performance and the abnormal performance, and triggering the oil leakage early warning mechanism when the defect degree is greater than a preset defect threshold; The process of obtaining the edge representation includes: counting the frequency of occurrence of each LBP differential value in the differential map, multiplying the frequency of occurrence of all LBP differential values ​​by the negative value of their logarithms and summing them to obtain the information entropy of the differential map; calculating the LBP differential mean in the differential map, and normalizing the product of the information entropy of the differential map and the LBP differential mean to obtain the edge representation of the differential map.

2. The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to claim 1 is characterized in that: The process of obtaining the degree of difference includes: Calculate the LBP difference mean of all difference maps in the target set; calculate the sum of the absolute difference values ​​of the LBP difference mean of the target map and the LBP difference mean of all other difference maps in the target set, and then divide it by the number of difference maps in the target set and normalize it to obtain the degree of difference of the target map relative to the target set as a whole.

3. The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to claim 2 is characterized in that: The process of obtaining the texture uniformity includes: The target image is divided into multiple windows according to the preset division parameters, and the sum of the squares of the differences between the LBP difference mean in all windows and the LBP difference mean of the target image is calculated, and then divided by the total number of windows to obtain the average square difference value, and the average square difference value is normalized to obtain the texture uniformity inside the target image.

4. The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to claim 3 is characterized in that: The abnormal performance satisfies the relationship: ; In the formula, is the abnormal performance of the target graph, is the difference between the target graph and the target set as a whole, is the texture uniformity inside the target image; where is the total number of difference graphs in the target set except the target graph, is the LBP difference mean of the target image, For the The LBP difference mean of the difference graphs, represents normalization processing, is the total number of windows after the target graph is divided, For the The LBP difference mean within the window, Represents an exponential function with the natural constant e as the base.

5. The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to claim 4 is characterized in that: The process of obtaining the defect degree includes: The defect degree is obtained by normalizing the product of the edge performance and the abnormal performance.

6. The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to claim 2 is characterized in that: The LBP difference means of all other difference maps in the target set also include: Calculate the similarity scores between other difference graphs in the target set and the target graph, assign weights to the difference graphs according to the similarity scores, and obtain the weighted average LBP difference mean of other difference graphs in the target set.

7. The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to claim 1 is characterized in that: After the grayscale image is acquired, smoothing and image enhancement are performed on the grayscale image.

8. The oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to claim 1 is characterized in that: The process of obtaining the edge representation of each difference image includes: The Canny edge detection algorithm is used to extract the edge pixels of the difference image, and the ratio of the number of edge pixels to the total number of pixels in the difference image is calculated as the edge representation of the difference image.

9. Oil-immersed transformer equipment oil leakage monitoring system based on image recognition, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the oil leakage monitoring method for oil-immersed transformer equipment based on image recognition according to any one of claims 1 to 8 is implemented.

Images