A Transformer Monitoring System Based on Image Recognition

Abstract

The present invention belongs to the field of monitoring, and discloses a transformer monitoring system based on image recognition, including a drone and an image recognition terminal; the drone includes an environmental data acquisition module, a calculation module, a shooting module and a communication module; the environmental data acquisition module is used to obtain the current wind speed and light intensity after the drone arrives at the position where the appearance of the transformer is photographed; the calculation module is used to calculate the number of shots based on the wind speed and the light intensity; the shooting module is used to shoot the appearance of the transformer based on the number of shots to obtain the appearance image of the transformer; the communication module is used to send the appearance image of the transformer to the image recognition terminal; the image recognition terminal is used to recognize the appearance image to obtain the appearance detection result of the transformer. The present invention can make the number of shots adapt to the spatial environment where the transformer is located, and reduce the probability of events that affect the recognition efficiency and the events that cannot accurately obtain the recognition results.

Description

Technical Field

[0001] The present invention relates to the field of monitoring, and in particular to a transformer monitoring system based on image recognition. Background Art

[0002] Transformer monitoring refers to the regular inspection of the internal and external appearance of the transformer, so as to continuously monitor whether the operating status of the transformer is normal, which is conducive to technical maintenance of transformers in abnormal conditions.

[0003] When inspecting the appearance of a transformer, image recognition is usually used. Compared with the traditional method of relying on human eyes for inspection, it requires less experience from the operation and maintenance personnel, and the inspection standards are more unified, which is conducive to obtaining more efficient and accurate transformer appearance inspection results.

[0004] In the prior art, when a drone is used to photograph the appearance of a transformer, a single image is usually taken or a fixed number of images is set to be taken, and then the single image taken or the image with the highest quality selected from the multiple images is identified to obtain the appearance status recognition result of the transformer.

[0005] However, this shooting method does not take into account the environmental conditions of the space where the transformer is located. When the wind is low and the light is sufficient in the space where the transformer is located, it is obviously unnecessary to shoot multiple images. In this case, shooting multiple images will result in too many images that need to be screened, which will obviously affect the efficiency of image-based recognition of the appearance status of the transformer. A single image can obtain an accurate recognition result of the appearance status; when the wind is strong and / or the light is insufficient in the space where the transformer is located, shooting a single image may result in quality defects in the image, such as insufficient clarity, ghosting, etc., which will result in the inability to correctly obtain the recognition result of the appearance status of the transformer. Summary of the invention

[0006] The purpose of the present invention is to disclose a transformer monitoring system based on image recognition, so as to solve the problem of how to reasonably determine the number of shots so that the number of shots is adapted to the spatial environment in which the transformer is located, and to reduce the probability of occurrence of events that affect the recognition efficiency and events that cannot accurately obtain recognition results.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] The present invention provides a transformer monitoring system based on image recognition, including a drone and an image recognition terminal;

[0009] The drone includes an environmental data acquisition module, a computing module, a shooting module, and a communication module;

[0010] The environmental data acquisition module is used to obtain the current wind speed and light intensity after the UAV arrives at the location where the transformer appearance is photographed;

[0011] The calculation module is used to calculate the number of shots based on wind speed and light intensity;

[0012] The formula for calculating the number of shots is:

[0013]

[0014] numpho indicates the number of shots, winspd indicates the wind speed, winspdst indicates the set wind speed comparison value, ltig indicates the light intensity, ltigst indicates the set light intensity comparison value, α indicates the reference weight for wind speed, β indicates the reference weight for light intensity, and refqua indicates the maximum number of shots set;

[0015] The shooting module is used to shoot the appearance of the transformer based on the number of shots to obtain an appearance image of the transformer;

[0016] The communication module is used to send the appearance image of the transformer to the image recognition terminal;

[0017] The image recognition terminal is used to identify the appearance image and obtain the appearance detection result of the transformer.

[0018] Preferably, photographing the appearance of the transformer based on the number of photographs to obtain an appearance image of the transformer includes:

[0019] If numpho is 1, a single shot is taken of the appearance of the transformer to obtain the appearance image of the transformer;

[0020] If numpho is greater than 1, the appearance of the transformer is photographed numpho times to obtain numpho appearance images of the transformer.

[0021] Preferably, the appearance image of the transformer is sent to an image recognition terminal, including

[0022] The appearance image of the transformer is sent to the image recognition terminal via the 4G network or 5G network.

[0023] Preferably, the image recognition terminal includes a screening module, a preprocessing module, a cutting module and a recognition module;

[0024] The screening module is used to screen the appearance image transmitted by the communication module to obtain an image to be identified for subsequent identification;

[0025] The preprocessing module is used to optimize the image to be recognized to obtain the image to be cut;

[0026] The cutting module is used to cut the image to be cut, and obtain a partial image of the appearance part of the transformer in the image to be cut;

[0027] The recognition module is used to recognize the local image and obtain the appearance detection result of the transformer.

[0028] Preferably, the appearance image transmitted by the communication module is screened to obtain an image to be identified for subsequent identification, including:

[0029] If numpho is 1, the appearance image transmitted by the communication module is directly used as the image to be identified for subsequent identification;

[0030] If numpho is greater than 1, numpho appearance images of the transformer are screened to obtain the image to be identified for subsequent identification.

[0031] Preferably, optimizing the image to be recognized to obtain the image to be cut includes:

[0032] Performing grayscale transformation on the image to be identified to obtain a first processed image;

[0033] Denoising the first processed image to obtain a second processed image;

[0034] The second processed image is subjected to grayscale optimization processing to obtain an image to be cut.

[0035] Preferably, the image to be cut is cut to obtain a partial image of the appearance part of the transformer in the image to be cut, including:

[0036] Calculate the segmentation threshold of the image to be cut;

[0037] Obtain multiple connected domains consisting of pixel points greater than the segmentation threshold in the image to be segmented;

[0038] The local image is composed of the pixels in the connected domain with the largest area in the image to be cut.

[0039] Preferably, calculating the segmentation threshold of the image to be segmented includes:

[0040] The Otsu method is used to calculate the segmentation threshold of the image to be cut.

[0041] Preferably, the appearance inspection result is qualified quality or unqualified quality.

[0042] Preferably, it also includes an early warning terminal;

[0043] The image recognition terminal is also used to send the appearance detection results to the early warning terminal;

[0044] The early warning terminal is used to issue an early warning to the transformer operation and maintenance management personnel when the appearance inspection result shows that the quality is unqualified.

[0045] Beneficial effects:

[0046] When the present invention uses a drone to shoot the transformer, the number of shots is calculated based on the wind speed and light intensity of the environment in which the transformer is located, so that the number of shots can adaptively change with the change of the environmental conditions in which the drone is located. When the environmental conditions for shooting are good, only one image needs to be shot, and when the environmental conditions for shooting are poor, multiple images are shot. In this way, it is possible to avoid the need to obtain the appearance detection result of the transformer based on multiple images when the environmental conditions for shooting are good, effectively improving the overall detection efficiency, and when the environmental conditions for shooting are poor, the appearance detection result can be obtained based on multiple images, improving the accuracy of the detection. The present invention can make the number of shots adapt to the spatial environment in which the transformer is located, and reduce the probability of occurrence of events that affect the recognition efficiency and events that cannot accurately obtain the recognition results. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 A schematic diagram of a transformer monitoring system based on image recognition according to the present invention.

[0049] Figure 2 Another schematic diagram of a transformer monitoring system based on image recognition according to the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope parameters of the protection of the present invention.

[0051] like Figure 1 In one embodiment shown, the present invention provides a transformer monitoring system based on image recognition, including a drone and an image recognition terminal;

[0052] The drone includes an environmental data acquisition module, a computing module, a shooting module, and a communication module;

[0053] The environmental data acquisition module is used to obtain the current wind speed and light intensity after the UAV arrives at the location where the transformer appearance is photographed;

[0054] The calculation module is used to calculate the number of shots based on wind speed and light intensity;

[0055] The formula for calculating the number of shots is:

[0056]

[0057] numpho indicates the number of shots, winspd indicates the wind speed, winspdst indicates the set wind speed comparison value, ltig indicates the light intensity, ltigst indicates the set light intensity comparison value, α indicates the reference weight for wind speed, β indicates the reference weight for light intensity, and refqua indicates the maximum number of shots set;

[0058] The shooting module is used to shoot the appearance of the transformer based on the number of shots to obtain an appearance image of the transformer;

[0059] The communication module is used to send the appearance image of the transformer to the image recognition terminal;

[0060] The image recognition terminal is used to identify the appearance image and obtain the appearance detection result of the transformer.

[0061] In the above embodiment, when the transformer is photographed by a drone, the number of shots is calculated based on the wind speed and light intensity of the transformer environment. The greater the wind speed and the smaller the light intensity, the more shots are taken, so that the number of shots can adaptively change with the changes in the environmental conditions of the drone. When the shooting environment conditions are good, only one image needs to be taken. When the shooting environment conditions are poor, multiple images are taken. In this way, it is possible to avoid the need to obtain the appearance inspection result of the transformer based on multiple images when the shooting environment conditions are good, thereby effectively improving the overall inspection efficiency. When the shooting environment conditions are poor, the appearance inspection result can be obtained based on multiple images, thereby improving the accuracy of the inspection.

[0062] Specifically, the position for photographing the appearance of the transformer is a position at a height of the middle of the transformer, and the shortest distance from the transformer is greater than the safe flight distance of the drone.

[0063] Preferably, the wind speed comparison value is set to 14 m / s. The light intensity comparison value is set to 100,000 lux.

[0064] Preferably, the maximum value of the set number of shots is 50.

[0065] Preferably, the reference weight for wind speed is 0.5, and the reference weight for light intensity is 0.5.

[0066] Preferably, the data acquisition module can be fixed directly above the drone body, thereby effectively reducing the impact of the airflow of the rotor on the wind speed measurement result.

[0067] Preferably, the data acquisition module can also be away from the drone body and set in advance in the space where the transformer is located, and the obtained wind speed and light intensity are sent to the calculation module by wireless communication.

[0068] Preferably, photographing the appearance of the transformer based on the number of photographs to obtain an appearance image of the transformer includes:

[0069] If numpho is 1, a single shot is taken of the appearance of the transformer to obtain the appearance image of the transformer;

[0070] If numpho is greater than 1, the appearance of the transformer is photographed numpho times to obtain numpho appearance images of the transformer.

[0071] Specifically, when performing multiple shots, the shooting module adopts a continuous shooting mode, and the time interval between two adjacent shots is the same.

[0072] Preferably, the appearance image of the transformer is sent to an image recognition terminal, including

[0073] The appearance image of the transformer is sent to the image recognition terminal via the 4G network or 5G network.

[0074] Specifically, the appearance image of the transformer may be sent to the image recognition terminal by means of satellite communication.

[0075] Preferably, the image recognition terminal includes a screening module, a preprocessing module, a cutting module and a recognition module;

[0076] The screening module is used to screen the appearance image transmitted by the communication module to obtain an image to be identified for subsequent identification;

[0077] The preprocessing module is used to optimize the image to be recognized to obtain the image to be cut;

[0078] The cutting module is used to cut the image to be cut, and obtain a partial image of the appearance part of the transformer in the image to be cut;

[0079] The recognition module is used to recognize the local image and obtain the appearance detection result of the transformer.

[0080] When there are multiple appearance images, it is necessary to screen the appearance images to avoid complete image recognition on all images, thereby improving recognition efficiency.

[0081] Preferably, the appearance image transmitted by the communication module is screened to obtain an image to be identified for subsequent identification, including:

[0082] If numpho is 1, the appearance image transmitted by the communication module is directly used as the image to be identified for subsequent identification;

[0083] If numpho is greater than 1, numpho appearance images of the transformer are screened to obtain the image to be identified for subsequent identification.

[0084] When there is only one appearance image, it can be directly used as the image to be recognized.

[0085] Preferably, screening numpho appearance images of the transformer to obtain an image to be identified for subsequent identification includes:

[0086] Calculate the screening value for each appearance image separately;

[0087] The appearance image with the highest screening value is used as the image to be recognized.

[0088] The filter value allows you to select the best quality appearance image.

[0089] Preferably, the screening value of each appearance image is calculated separately, including:

[0090] Dispow is used to represent the distance between the camera module and the transformer, and the calculation formula of the screening radius R is:

[0091]

[0092] disref indicates the maximum shooting distance, nfrow and nfcol indicate the number of rows and columns of the appearance image, respectively, and min indicates obtaining the smaller value in the brackets;

[0093] The screening value of each appearance image is calculated using the following formula:

[0094]

[0095] filval represents the screening value, cirpix represents the set of pixels in the circular area with a radius of R and the center of the appearance image, grayval i Represents the grayscale value of pixel i, and ncir represents the number of pixels in cirpix.

[0096] In the present invention, the screening value is not calculated based on all the pixels in the appearance image, because when the image captured by the shooting module has a blurred background, if the calculation is performed for all the pixels, when the blurred background area is much larger than the foreground area, that is, the area where the transformer appearance is located, even if the clarity of the foreground part is greatly different, the difference in the final calculated screening value will not be significant. This also makes it impossible to significantly increase the difference in screening values ​​between different appearance images, which is not conducive to selecting the appearance image with the best quality.

[0097] Therefore, the present invention calculates the screening radius based on the distance between the shooting module and the transformer, so that when the shooting module is farther away from the transformer, the screening radius is smaller, so that the area used to calculate the screening value is more concentrated towards the center of the image, thus effectively reducing the influence of the blurred background on the calculated screening value; and when the shooting module is closer to the transformer, the screening radius is larger, thus, the screening value can be calculated based on more pixel points belonging to the appearance part of the transformer, so that the screening value can more fully represent the quality of the appearance image.

[0098] Preferably, optimizing the image to be recognized to obtain the image to be cut includes:

[0099] Performing grayscale transformation on the image to be identified to obtain a first processed image;

[0100] Denoising the first processed image to obtain a second processed image;

[0101] The second processed image is subjected to grayscale optimization processing to obtain an image to be cut.

[0102] Noise reduction and grayscale optimization processing can make the image outline clearer, which is conducive to the subsequent accurate segmentation of the local image belonging to the appearance of the transformer.

[0103] Specifically, when grayscale transformation is performed on the image to be identified, each pixel in the image to be identified may be processed by using a maximum value method, an average value method or a weighted average method, so as to obtain a first processed image.

[0104] Preferably, performing noise reduction on the first processed image to obtain the second processed image includes:

[0105] S1, using the first processed image as a denoised image;

[0106] S2, save all the pixels in the first processed image into a set pixs;

[0107] S3, calculating the noise reduction coefficient of each pixel in pixs in the noise reduction image;

[0108] S4, performing denoising on the pixel point with the largest denoising coefficient in the denoised image to obtain the grayscale value of the pixel point after denoising;

[0109] S5, in the denoised image, the grayscale value of the pixel point denoised in S4 is modified to the grayscale value obtained in S4, thereby obtaining a new denoised image;

[0110] S6, judging whether the number of pixels in pixs is greater than 0, if so, deleting the pixels whose noise is reduced in S4 from pixs and entering S3; if not, taking the new noise-reduced image obtained in S5 as the second processed image.

[0111] The noise reduction process of the present invention is different from the prior art. If the prior art is used to reduce the noise of the first image, each pixel is denoised separately based on the grayscale value in the first processed image. Obviously, this noise reduction method does not distinguish the priority of the noise reduction of the pixel points. Therefore, it is impossible to reduce the noise of the subsequent pixel points based on the previous noise reduction results. However, when the present invention performs noise reduction, a new noise-reduced image is obtained after each pixel point is denoised. The next noise reduction is performed on the basis of the noise-reduced image obtained by the previous noise reduction. For the pixel points, the later the order of noise reduction is, the more accurate the result obtained by filtering will be because the number of surrounding pixel points that have been noise-reduced is greater. The noise reduction method of the present invention can further improve the filtering effect, so that the degree of detail restoration in the image obtained after filtering is higher.

[0112] Preferably, the calculation formula of the noise reduction coefficient is:

[0113]

[0114] noiredcoef j Indicates the noise reduction coefficient of pixel j, edgjud j Indicates the edge judgment value of pixel j, edgjudse indicates the set radius, numalr j It represents the number of pixels that have been denoised in the circular area A with a radius of edgjudse and a pixel point i as the center. numfir represents the total number of pixels in A. gray j Represents the gray value of pixel j, vegray j represents the mean gray value of the 8-neighborhood of pixel j; λ1 represents the edge judgment weighting coefficient, λ2 represents the quantity weighting coefficient, and λ3 represents the gray weighting coefficient.

[0115] In the present invention, if the probability that a pixel point is at the edge of an image is smaller, that is, the edge judgment value is smaller, the number of pixels that have undergone noise reduction in the area around the pixel point is greater, and the difference in grayscale value with the surrounding pixels is greater, then the noise reduction priority of the pixel point is higher. In this way, the noise reduction order of the pixels belonging to the edge of the image can be made later, and filtering can be performed based on the grayscale values ​​of more pixels that have undergone noise reduction, thereby effectively improving the clarity of the edge of the image in the second processed image.

[0116] Preferably, edgjud j The acquisition process includes:

[0117] With the horizontal right direction as 0° and the pixel point j as the endpoint of the ray, rays are drawn along the directions of 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315° respectively;

[0118] Save the pixel points in A that are passed by the rays drawn along the directions of 0° and 180° into the set B0;

[0119] The pixel points in A that are passed by the rays drawn along the directions of 45° and 225° are saved in the set B45;

[0120] Save the pixel points in A that are passed by the rays drawn along the directions of 90° and 270° into the set B90;

[0121] The pixel points in A that are passed by the rays drawn along the directions of 135° and 315° are saved in the set B135;

[0122] Calculate the absolute values ​​ABS0, ABS45, ABS90, ABS135 of the differences between the mean grayscale values ​​of the pixels in B0, B45, B90, B135 and the mean grayscale values ​​of the pixels in A respectively;

[0123] The maximum value among ABS0, ABS45, ABS90 and ABS13 is represented as ABSmax;

[0124] The corresponding set of ABSmax in B0, B45, B90, and B135 is denoted as Bmax;

[0125] The number of pixels in Bmax whose absolute value of the grayscale difference between j and j is in the interval [0,5] is taken as edgjud j The value of .

[0126] When calculating the edge judgment value, for B0, B45, B90, and B135, the greater the difference in grayscale value between the pixel points in the set and the pixel points in A, and the closer the grayscale value is to the grayscale value of j, the greater the probability that the set and the pixel point j are on the same image edge. Therefore, by obtaining the number of pixels in Bmax whose grayscale value is very different from that of j, the probability that j belongs to the edge of the image can be more accurately expressed. Without the need for complex calculations such as edge detection, the probability that the pixel point j belongs to the edge of the image can be quickly determined. The efficiency of noise reduction is effectively improved.

[0127] Preferably, the radius is set to 3.

[0128] Preferably, the edge judgment weighting coefficient is 0.3, the quantity weighting coefficient is 0.4, and the grayscale weighting coefficient is 0.3.

[0129] Preferably, performing grayscale optimization processing on the second processed image to obtain the image to be cut includes:

[0130] The second processed image is subjected to histogram equalization processing to obtain an image to be cut.

[0131] Specifically, performing histogram equalization processing can make the distribution of grayscale values ​​in the image more balanced and the image clearer, which is beneficial to improving the accuracy of appearance detection results.

[0132] Preferably, the image to be cut is cut to obtain a partial image of the appearance part of the transformer in the image to be cut, including:

[0133] Calculate the segmentation threshold of the image to be cut;

[0134] Obtain multiple connected domains consisting of pixel points greater than the segmentation threshold in the image to be segmented;

[0135] The local image is composed of the pixels in the connected domain with the largest area in the image to be cut.

[0136] Specifically, in the present invention, the grayscale value of the pixel point in the foreground part is greater than the grayscale value of the pixel point in the background part. Therefore, after calculating the segmentation threshold, the pixel points belonging to the background can be filtered out based on the segmentation threshold. Then, since there may be some areas in the background with grayscale values ​​close to the foreground, but the area of ​​the foreground part is generally larger, the present invention accurately obtains the local image of the appearance part of the transformer by filtering the composed connected domain based on the area.

[0137] Preferably, calculating the segmentation threshold of the image to be segmented includes:

[0138] The Otsu method is used to calculate the segmentation threshold of the image to be cut.

[0139] Specifically, other threshold calculation algorithms may also be used to obtain the segmentation threshold.

[0140] Preferably, the appearance image is identified to obtain the appearance detection result of the transformer, including:

[0141] The appearance image is input into a trained deep learning-based model (such as a convolutional neural network model) for image recognition to obtain a detection result of the appearance of the transformer.

[0142] Preferably, the appearance inspection result is qualified quality or unqualified quality.

[0143] Specifically, when the appearance image is identified, if any pre-set defect such as cracking, rusting, deformation, etc. is found in the appearance image, the appearance inspection result is that the quality does not meet the requirements.

[0144] Preferably, if Figure 2 As shown, it also includes an early warning terminal;

[0145] The image recognition terminal is also used to send the appearance detection results to the early warning terminal;

[0146] The early warning terminal is used to issue an early warning to the transformer operation and maintenance management personnel when the appearance inspection result shows that the quality is unqualified.

[0147] Specifically, the warning may be issued by popping up a prompt window on the computer used by the operation and maintenance manager.

[0148] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0149] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0150] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0151] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope parameters of the technical solutions of the embodiments of the present invention.

Claims

1. A transformer monitoring system based on image recognition, characterized in that: Including drones and image recognition terminals; The drone includes an environmental data acquisition module, a computing module, a shooting module, and a communication module; The environmental data acquisition module is used to obtain the current wind speed and light intensity after the UAV arrives at the location where the transformer appearance is photographed; The calculation module is used to calculate the number of shots based on wind speed and light intensity; The formula for calculating the number of shots is: numpho indicates the number of shots, winspd indicates the wind speed, winspdst indicates the set wind speed comparison value, ltig indicates the light intensity, ltigst indicates the set light intensity comparison value, α indicates the reference weight for wind speed, β indicates the reference weight for light intensity, and refqua indicates the maximum number of shots set; The shooting module is used to shoot the appearance of the transformer based on the number of shots to obtain an appearance image of the transformer; The communication module is used to send the appearance image of the transformer to the image recognition terminal; The image recognition terminal is used to recognize the appearance image and obtain the appearance detection result of the transformer; The image recognition terminal includes a screening module, a preprocessing module, a cutting module and a recognition module; The screening module is used to screen the appearance image transmitted by the communication module to obtain an image to be identified for subsequent identification; The preprocessing module is used to optimize the image to be recognized to obtain the image to be cut; The cutting module is used to cut the image to be cut, and obtain a partial image of the appearance part of the transformer in the image to be cut; The recognition module is used to recognize the local image and obtain the appearance detection result of the transformer; The appearance image transmitted by the communication module is screened to obtain an image to be identified for subsequent identification, including: If numpho is 1, the appearance image transmitted by the communication module is directly used as the image to be identified for subsequent identification; If numpho is greater than 1, numpho appearance images of the transformer are screened to obtain the image to be identified for subsequent identification; Screen numpho appearance images of the transformer to obtain images to be identified for subsequent identification, including: Calculate the screening value for each appearance image separately; The appearance image with the highest screening value is used as the image to be recognized; Calculate the filter value for each appearance image separately, including: Dispow is used to represent the distance between the camera module and the transformer, and the calculation formula of the screening radius R is: disref indicates the maximum shooting distance, nfrow and nfcol indicate the number of rows and columns of the appearance image, respectively, and min indicates obtaining the smaller value in the brackets; The screening value of each appearance image is calculated using the following formula: filval represents the screening value, cirpix represents the set of pixels in the circular area with a radius of R and the center of the appearance image, grayval i Represents the grayscale value of pixel i, and ncir represents the number of pixels in cirpix.

2. A transformer monitoring system based on image recognition according to claim 1, characterized in that: The appearance of the transformer is photographed based on the number of photographs to obtain the appearance image of the transformer, including: If numpho is 1, a single shot is taken of the appearance of the transformer to obtain the appearance image of the transformer; If numpho is greater than 1, the appearance of the transformer is photographed numpho times to obtain numpho appearance images of the transformer.

3. A transformer monitoring system based on image recognition according to claim 1, characterized in that: Send the transformer appearance image to the image recognition terminal, including The appearance image of the transformer is sent to the image recognition terminal via the 4G network or 5G network.

4. A transformer monitoring system based on image recognition according to claim 1, characterized in that: The image to be recognized is optimized to obtain the image to be cut, including: Performing grayscale transformation on the image to be identified to obtain a first processed image; Denoising the first processed image to obtain a second processed image; The second processed image is subjected to grayscale optimization processing to obtain an image to be cut.

5. The transformer monitoring system based on image recognition according to claim 1 is characterized in that: The image to be cut is cut to obtain a local image of the appearance of the transformer in the image to be cut, including: Calculate the segmentation threshold of the image to be cut; Obtain multiple connected domains consisting of pixel points greater than the segmentation threshold in the image to be segmented; The local image is composed of the pixels in the connected domain with the largest area in the image to be cut.

6. A transformer monitoring system based on image recognition according to claim 5, characterized in that: Calculate the segmentation threshold of the image to be cut, including: The Otsu method is used to calculate the segmentation threshold of the image to be cut.

7. The transformer monitoring system based on image recognition according to claim 1 is characterized in that: The appearance inspection result is qualified or unqualified.

8. The transformer monitoring system based on image recognition according to claim 7 is characterized in that: It also includes early warning terminals; The image recognition terminal is also used to send the appearance detection results to the early warning terminal; The early warning terminal is used to issue an early warning to the transformer operation and maintenance management personnel when the appearance inspection result shows that the quality is unqualified.

Images