Deteriorated insulator automatic detection method based on infrared temperature image

Abstract

The invention belongs to the technical field of insulator fault detection, and particularly discloses a degraded insulator automatic detection method based on an infrared temperature image, which comprises the following steps of: carrying a refrigeration thermal infrared imager on an unmanned aerial vehicle, and obtaining a degraded insulator by improving an insulator steel cap segmentation and temperature extraction algorithm; the influence on the temperature reading of the insulator caused by factors such as large dust coverage and light reflection on the insulator is eliminated, so that the temperature reading of the insulator is more accurate; various complete insulator temperature curves of insulators at different positions and with different degradation degrees under the conditions of different wind speeds, humidity, environment temperatures and the like can be obtained through simulation, the temperature curves are converted into AD value data according to the temperature and AD value conversion relation of the refrigeration thermal infrared imager, a data set is made, and then through a recurrent neural network, the temperature curve of the insulators at different positions and different degradation degrees is calculated. According to the invention, tiny changes of the temperature curve of the insulator under various conditions can be learned, so that the temperature change range of different positions of each insulator can be accurately judged, and the degraded insulator can be automatically and accurately detected.

Description

Technical Field

[0001] The invention relates to the technical field of insulator fault detection, in particular to an automatic detection method for deteriorated insulators based on infrared temperature images. Background Art

[0002] Zero value detection of porcelain insulators in power transmission lines is one of the important tasks to ensure safe and stable operation of transmission lines. As the voltage level increases, the number of insulators used also increases, forming an insulator string. However, as the operating time increases, the insulators will gradually deteriorate. At this time, in order to ensure normal conduction, it is necessary to detect the deterioration of the insulators. The most common detection methods are observation, spark fork method, infrared temperature measurement method and ultraviolet imaging method. Among them, infrared temperature measurement method is the most common detection method, which is generally performed by drone operation. However, the existing infrared temperature measurement method has the following problems:

[0003] (1) There is a problem of temperature inaccuracy in Celsius temperature processing:

[0004] When judging deteriorated insulators based on temperature, whether the temperature is accurate is a key factor, because in order to accurately judge the deteriorated insulator based on the slight change in temperature, the Celsius temperature can only retain two decimal places. At this time, judging the deteriorated insulator based on the Celsius temperature is not accurate enough, because when the resistance of the deteriorated insulator is above 500MΩ, the heat generation of the deteriorated insulator is close to that of a normal insulator. At this time, more accurate temperature data is needed to distinguish between deteriorated insulators and normal insulators.

[0005] (2) Lack of dust coverage and reflection judgment affects the actual insulator temperature reading:

[0006] Ceramic insulators are easily covered with dust, and composite insulators are prone to reflection when shooting infrared videos. Both of these situations will affect the actual temperature of the insulator. For example, the ceramic insulator itself is a normal insulator, but there is a certain area of ​​dust coverage at the steel cap. The temperature in the dust-covered area is close to 0 degrees Celsius. At this time, the average temperature of the steel cap will be affected by the dust and be lower, so this place will be misjudged as a zero-value degraded insulator; the temperature value of the composite insulator's reflective area will be higher than that of a normal insulator, resulting in the misjudgment of the reflective area as a low-value degraded insulator. These phenomena will lead to misjudgment and affect the accuracy of judging degraded insulators.

[0007] (3) Lack of algorithms to automatically identify deteriorated insulators based on small fluctuations in the insulator temperature curve:

[0008] Some methods for judging the temperature of insulators are to set thresholds, and some methods judge deteriorated insulators according to the change of the standard temperature curve of insulators. However, these methods are not flexible enough. First, due to the different distribution voltages at different positions of insulators, the heating temperatures caused by deterioration at different positions of the insulator string are also different. Detecting only based on one threshold will lead to a large number of false detections. Second, some judge deteriorated insulators according to the standard temperature curve of insulators. By comparing the obtained temperature curve of the insulator with the standard curve, and then judging whether it is a deteriorated insulator. This method is feasible but inefficient, and the made standard curve is too single. Although the temperature curve of the insulator is an irregular U-shaped structure under ideal conditions, under different natural conditions such as wind speed, ambient temperature, and humidity, the actually photographed temperature of the insulator will not exactly conform to a smooth irregular U-shaped curve. Judging only according to the standard temperature curve will greatly increase the false detection rate in actual judgment. Summary of the Invention

[0009] In order to solve the above technical problems, the present invention provides an automatic detection method for deteriorated insulators based on infrared temperature images.

[0010] To achieve the above object, the present invention is implemented according to the following technical solution:

[0011] An automatic detection method for deteriorated insulators based on infrared temperature images includes the following steps:

[0012] S1. On the premise of knowing the resistance values, ambient temperature, humidity, and wind speed of normal insulators and deteriorated insulators for experiments, obtain the infrared images of the marked normal insulators and deteriorated insulators through a refrigerated infrared thermal imager; and extract the temperature data in the infrared images to obtain the actual temperature of the insulators; wherein, the insulators include ceramic insulators and composite insulators;

[0013] S2. Simulate the simulated temperature of the insulator that is the same as the actual temperature of the insulator in comsol simulation, convert the simulated insulator temperature obtained by the simulation into the AD value of the actual temperature of the insulator, and make an insulator AD value data set;

[0014] S3. Use the AD value data set to train the designed recurrent neural network to automatically judge normal insulators and deteriorated insulators, and obtain a trained recurrent neural network model;

[0015] S4. On the premise of knowing the resistance value, ambient temperature, humidity, and wind speed of the insulator, use the refrigerated infrared thermal imager carried by the UAV to collect the infrared image of the insulator; perform image segmentation on the infrared image of the insulator and exclude the influence of a large amount of dust coverage or reflection factors to obtain the real-time temperature of the insulator; input the obtained real-time temperature of the insulator into the trained recurrent neural network model, and the recurrent neural network model automatically judges whether the insulator is a normal insulator or a deteriorated insulator.

[0016] Further, in the step S2, the simulated insulator temperature is temperature data with eight decimal places.

[0017] Further, in the step S2, the simulated insulator temperature is converted into the AD value of the actual temperature of the insulator through a fitting function, and the fitting function is:

[0018] y = A1x 2 + A2x + A3;

[0019] A1, A2, and A3 are constant coefficients, which are obtained through the corresponding table of Celsius temperature and AD value of the refrigerated thermal imager.

[0020] Further, in the step S3, the recurrent neural network is an RNN recurrent neural network, and dropout is added to reduce overfitting; tanh is selected as the activation function, and the sigmiod is used for classification by the classifier.

[0021] Further, in the step S4, for ceramic insulators, the method for obtaining the real-time temperature of ceramic insulators is as follows:

[0022] (1) Perform guided filtering on the infrared image of the ceramic insulator. Remove image noise and smooth the image through guided filtering to enhance image details; after guided filtering, perform detail enhancement on the image. Let the temperature data without guided filtering be represented by b, the temperature data after guided filtering be represented by d, and the temperature data after detail processing be represented by f. Then the detail processing method is:

[0023] f = b + 0.8(b - d) - min(b + 0.8(b - d));

[0024] (2) Perform image segmentation on the infrared image of the ceramic insulator through MaskRcnn, segment out the steel cap part of the ceramic insulator, and record the coordinates of the steel cap bounding box;

[0025] (3) Binarize the segmented image to obtain a binary image, use the binary image for contour detection to obtain the contour of the steel cap bounding box, calculate the contour area, and set the contour area threshold to 30 pixels: If the contour area in pixels is less than 30, it is regarded as a false detection; exclude the parts determined as false detections and retain the remaining bounding box coordinates; find the minimum bounding rectangle for the contours that meet the threshold requirements, further optimize the irregularly segmented bounding boxes to obtain a more accurate rectangular insulator segmentation box, and record the four corner coordinates of each steel cap rectangular segmentation box of the ceramic insulator;

[0026] (4) According to the four corner coordinates of the rectangular bounding box obtained by the minimum bounding rectangle method, record the number of pixels within the bounding box as p i , and record the temperature of each pixel within the bounding box as t i , at this time, by comparing the absolute value of the difference between the lowest temperature and the highest temperature represented by the pixels within the steel cap bounding box of the same ceramic insulator, if the absolute value of the temperature difference is greater than 1.5 degrees Celsius, then it is determined that there is a large area covered with dust at this steel cap, and discard it by calculating the proportion of the temperature difference from the maximum value above 1.5 degrees Celsius. Let the maximum temperature within the rectangular bounding box be T max , the total number of pixels within the entire rectangular bounding box is P s , those lower than T max - 1.5 degrees Celsius are recorded as the number of temperature pixels P l , then the dust coverage ratio P is:

[0027]

[0028] If P ≥ 50%, discard the temperature data that is more than 0.8 degrees Celsius lower than the maximum value; if 10% < P < 50%, discard the data that is more than 1 degree Celsius lower than the maximum value; if P ≤ 10%, discard the data that is more than 1.5 degrees Celsius lower than the maximum value. Process the remaining data according to the following formula to obtain the average temperature avg(t) of the steel cap of the ceramic insulator, that is, the real-time temperature of the ceramic insulator:

[0029]

[0030] pi represents the number of pixels that meet the conditions, and t i represents the temperature data corresponding to each pixel.

[0031] Further, in step S4, for the composite insulator, the method for obtaining the real-time temperature of the composite insulator is:

[0032] 1) Perform guided filtering on the infrared image of the composite insulator. Remove image noise and smooth the image through guided filtering to enhance image details. After guided filtering, perform detail enhancement on the image. Denote the temperature data without guided filtering as b, the temperature data after guided filtering as d, and the temperature data after detail processing as f. Then the detail processing method is as follows:

[0033] f = b + 0.8(b - d) - min(b + 0.8(b - d));

[0034] 2) Perform image segmentation on the infrared image of the composite insulator through MaskRcnn;

[0035] 3) Binarize the segmented image to obtain a binary image. Use the binary image for contour detection to obtain the contour of the bounding box of the composite insulator, and calculate the contour area. The contour area threshold is set to 3000 pixels. If the contour area in pixels is less than 3000, it is regarded as a false detection. Exclude the falsely detected part and retain the remaining bounding box coordinates. For the contours meeting the threshold requirements, find the minimum bounding rectangle to further optimize the irregularly segmented bounding box and obtain a more accurate rectangular insulator segmentation box, and record the coordinates of the central axis of the composite insulator;

[0036] 4) Extract the temperature data of all points on the central axis of the composite insulator and plot them as the temperature curve of the central axis of the composite insulator;

[0037] 5) Since the reflective points occur at the peaks of the temperature curve of the central axis of the composite insulator, mark all the peaks of the temperature curve of the central axis of the composite insulator. Calculate the distance between each peak. According to the multi-digit method, find the regular distances, record this distance as D, and compare the other peak distances D i with D. If |D i - D| ≤ 2, the corresponding peak is also recorded as caused by reflection, and mark the serial numbers of all qualified peaks;

[0038] 6) Calculate the half-wave width of all peaks and exclude the obviously irregular peaks according to the quartile method. Record and mark the serial numbers of the peaks meeting the half-wave peak width requirements;

[0039] 7) Perform an intersection operation on the peak serial numbers in steps 5) and 6). The peaks that meet both conditions in steps 5) and 6) are the peaks with high temperature caused by reflection. Extract the peaks caused by reflection, eliminate the peaks through moving average filtering, eliminate the influence of reflection on the misjudgment of the composite insulator, and obtain the real-time temperature of the composite insulator.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) The temperature data is refined, improving the accuracy of judging deteriorated insulators based on temperature. Generally, ordinary infrared cameras can only read the temperature to two decimal places in degrees Celsius. However, when the resistance of the insulator is above 500 MΩ, the heat generated by the deteriorated insulator is close to that of a normal insulator. Therefore, in this invention, a cooled infrared thermal imager is used instead, and the AD value read is more accurate. Moreover, the temperature obtained by simulation can be accurate to eight decimal places. Therefore, the temperature obtained by simulation is already very accurate and can be converted into an AD value through a fitting function, and the dataset remains accurate and reliable.

[0042] (2) The steel cap segmentation technology of ceramic insulators is optimized. Before segmentation, guided filtering and detail enhancement are performed first to smooth the image and ensure that the original insulator temperature data will not be lost. Then, the steel cap bounding box of the ceramic insulator is segmented by Mask RCNN. Whether it is a normal steel cap bounding box is judged according to the area of the bounding box. Then, the minimum bounding rectangle processing is performed on the bounding box. Through these two steps, problems such as segmentation errors and inaccurate and irregular bounding box ranges can be excluded.

[0043] (3) A dust judgment algorithm is proposed to eliminate misjudgments caused by factors such as dust. Since the deteriorated insulator is judged based on temperature, when there is a large area of connected dust and other pollutants on the steel cap of the ceramic insulator, the temperature of the covered area is close to 0 degrees Celsius, which will affect the calculation of the average temperature of the insulator and cause the temperature to be on the low side. Therefore, a dust judgment algorithm is proposed to exclude the influence of a large area of connected dust and other pollutants on the insulator temperature.

[0044] (4) A reflection judgment algorithm for composite insulators is proposed to eliminate misjudgments of deteriorated insulators caused by reflection. The temperature at the reflection area will be higher than the heat generation temperature of a normal insulator, and it is easy to judge the reflection area as a low-value deteriorated insulator. This invention can exclude the phenomenon of temperature increase of composite insulators caused by reflection, prevent misjudging the temperature anomaly caused by reflection as a deteriorated insulator, and improve the accuracy of judging whether a composite insulator is a deteriorated insulator based on temperature.

[0045] (5) Automatically and accurately judge deteriorated insulators based on temperature. Different positions and environments of insulators will affect the temperature change of insulators, and the temperature curve of insulators is variable. Judging only based on the standard temperature curve will result in a large error. Therefore, training and learning are carried out through an RNN recurrent neural network, and the captured temperature data is directly input into the RNN model to automatically judge deteriorated insulators. Moreover, since a large amount of insulator temperature data under various environments and resistances can be obtained through simulation, the model can fully learn the relationships of various insulator temperature data, and the accuracy of judging deteriorated insulators is higher. Description of the Drawings

[0046] Figure 1Flow chart of the automatic detection method for deteriorated insulators based on the infrared temperature image of ceramic insulators.

[0047] Figure 2 Flow chart of the automatic detection method for deteriorated insulators based on the infrared temperature image of composite insulators. Detailed implementation manners

[0048] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the invention.

[0049] Embodiment 1

[0050] As Figure 1 shown, this embodiment exemplarily shows an automatic detection method for deteriorated insulators based on infrared temperature images, which is for ceramic insulators. The specific steps are as follows:

[0051] S1. On the premise of knowing the resistance values, ambient temperature, humidity, and wind speed of normal ceramic insulators and deteriorated ceramic insulators for experiments, obtain the infrared images of the marked normal ceramic insulators and deteriorated ceramic insulators through a refrigerated infrared thermal imager; and extract the temperature data in the infrared images to obtain the actual temperature of the ceramic insulators.

[0052] S2. Simulate the simulated temperature of the ceramic insulator that is the same as the actual temperature of the ceramic insulator in comsol simulation, convert the simulated temperature of the simulated ceramic insulator into the AD value of the actual temperature of the ceramic insulator, and make a ceramic insulator AD value data set; specifically:

[0053] Compare with the insulator temperature data actually captured. According to the resistance value of the insulator in step S1, in comsol simulation, by changing the conductivity and relative permittivity, change the resistance value of the simulated insulator, simulate the ambient temperature, humidity and wind speed of the actual insulator, and simulate the temperature data of the insulator with the same resistance value and ambient temperature as in the actual environment in comsol simulation, and compare it with the temperature data of the experimental insulator actually captured. If the simulation data is consistent with the actual data, it means that the temperature value obtained by simulating the insulator is credible.

[0054] Change the position and resistance value of the deteriorated ceramic insulator in the insulator string, simulate the temperature change of various ceramic insulators under different environmental factors, and obtain the second-order fitting function by fitting the Celsius temperature x and AD value y correspondence table of the refrigerated infrared thermal imager:

[0055] y = A1x 2 + A2x + A3;

[0056] Table 1 Correspondence between AD value and Celsius degree

[0057]

[0058]

[0059] A1, A2, and A3 are constant coefficients, which are obtained through the correspondence table between the Celsius temperature of the refrigerated thermal imager and the AD value (see Table 1). The simulated temperature data with eight decimal places is converted into the AD value captured by the infrared thermal imager through a fitting function to create a data set. The data set is divided into a training set and a test set. To ensure the persuasiveness of the effect, the training set uses a mixture of simulated data and some actual captured data, and the test set uses only the actual captured data.

[0060] S3. Use the AD value data set to train the designed recurrent neural network to automatically identify normal ceramic insulators and deteriorated ceramic insulators, and obtain a trained recurrent neural network model. Specifically:

[0061] Step 1: Designate the first insulator away from the conductor side of the ceramic insulator as No. 1, and number the remaining insulators in sequence; read the temperature data of the ceramic insulators, and convert the temperature data into a vector according to the insulator numbers, and input the AD value data of the ceramic insulators into the recurrent neural network in sequence each time.

[0062] Step 2: Select the RNN recurrent neural network, add dropout to reduce overfitting, select tanh as the activation function, and the final result is divided into normal ceramic insulators and deteriorated ceramic insulators, which is a binary classification problem, and sigmiod is used for classification.

[0063] Step 3: Train the training set and test the trained model with the test set to detect the training effect of the model.

[0064] S4. On the premise of knowing the resistance value, ambient temperature, humidity, and wind speed of the ceramic insulator, collect the infrared image of the ceramic insulator through a refrigerated infrared thermal imager carried by a drone; perform image segmentation on the infrared image of the ceramic insulator and eliminate the influence of a large amount of dust coverage or reflection factors to obtain the real-time temperature of the ceramic insulator. The method for obtaining the real-time temperature of the ceramic insulator is as follows:

[0065] (1) Perform guided filtering on the infrared image of the ceramic insulator. Remove image noise, smooth the image, and enhance image details through guided filtering; after guided filtering, perform detail enhancement on the image. Let the temperature data without guided filtering be represented by b, the temperature data after guided filtering be represented by d, and the temperature data after detail processing be represented by f. Then the detail processing method is:

[0066] f = b + 0.8(b - d) - min(b + 0.8(b - d));

[0067] (2) Use MaskRcnn to perform image segmentation on the infrared image of the ceramic insulator, segment the steel cap part of the ceramic insulator, and record the coordinates of the bounding box of the steel cap.

[0068] (3) Binarize the segmented image to obtain a binary image, use the binary image for contour detection to obtain the contour of the bounding box of the steel cap, calculate the contour area, and set the contour area threshold to 30 pixels: if the contour area in pixels is less than 30, it is regarded as a false detection; exclude the part determined to be a false detection and retain the remaining bounding box coordinates; find the minimum bounding rectangle for the contours that meet the threshold requirements, further optimize the irregularly segmented bounding box to obtain a more accurate rectangular insulator segmentation box, and record the four corner coordinates of each rectangular segmentation box of the steel cap of the ceramic insulator.

[0069] (4) According to the four corner coordinates of the rectangular bounding box obtained by the minimum bounding rectangle method, record the number of pixels within the bounding box as p i , and record the temperature of each pixel within the bounding box as t i , at this time, by comparing the absolute value of the difference between the lowest temperature and the highest temperature represented by the pixels within the bounding box of the steel cap of the same ceramic insulator, if the absolute value of the temperature difference is greater than 1.5 degrees Celsius, then it is determined that there is a large area covered with dust at this steel cap, and discard it by calculating the proportion of the temperature difference from the maximum value above 1.5 degrees Celsius. Let the maximum temperature within the rectangular bounding box be T max , the total number of pixels within the entire rectangular bounding box is P s , lower than T max - the number of temperature pixels at 1.5 degrees Celsius is recorded as P l , then the dust coverage ratio P is:

[0070]

[0071] If P ≥ 50%, discard the temperature data that is more than 0.8 degrees Celsius lower than the maximum value. If 10% < P < 50%, discard the data that is more than 1 degree Celsius lower than the maximum value. If P ≤ 10%, discard the data that is more than 1.5 degrees Celsius lower than the maximum value. Process the remaining data according to the following formula to obtain the average temperature avg(t) of the steel cap of the ceramic insulator, that is, the real-time temperature of the ceramic insulator:

[0072]

[0073] pi represents the number of pixels that meet the conditions, and t i represents the temperature data corresponding to each pixel.

[0074] Finally, input the obtained real-time temperature of the ceramic insulator into the trained recurrent neural network model, and the recurrent neural network model automatically determines whether the insulator is a normal ceramic insulator or a deteriorated ceramic insulator.

[0075] Embodiment 2

[0076] As Figure 2 shown, this embodiment exemplarily demonstrates an automatic detection method for deteriorated insulators based on infrared temperature images, which is applicable to composite insulators. The specific steps are as follows:

[0077] S1. On the premise of knowing the resistance values, ambient temperature, humidity, and wind speed of the known normal composite insulators and deteriorated composite insulators for experiments, obtain the infrared images of the marked normal composite insulators and deteriorated composite insulators through a refrigerated infrared thermal imager; and extract the temperature data in the infrared images to obtain the actual temperature of the composite insulators.

[0078] S2. Simulate the simulated temperature of the composite insulator that is the same as the actual temperature of the composite insulator in comsol simulation, convert the simulated temperature of the composite simulated insulator into the AD value of the actual temperature of the composite insulator, and make a dataset of the AD values of the composite insulators; specifically:

[0079] Compare with the temperature data of the actually photographed composite insulators. According to the resistance value of the composite insulator in step S1, in comsol simulation, by changing the conductivity and relative permittivity, change the resistance value of the simulated insulator, simulate the ambient temperature, humidity, and wind speed of the actual insulator, and simulate the temperature data of the insulator with the same resistance value and ambient temperature as in the actual environment in comsol simulation, and compare it with the temperature data of the actually photographed experimental composite insulators. If the simulation data is consistent with the actual data, it indicates that the temperature value obtained by simulating the composite insulator is credible.

[0080] Change the position and resistance value of the deteriorated composite insulator in the insulator string, simulate the temperature change situations of various composite insulators under different environmental factors, and obtain the second-order fitting function by fitting the correspondence table of the Celsius temperature x and the AD value y of the refrigerated infrared thermal imager:

[0081] y = A1x 2 + A2x + A3;

[0082] A1, A2, and A3 are constant coefficients, which are obtained from the correspondence table of the Celsius temperature and the AD value of the refrigerated thermal imager (see Table 1). Convert the simulated temperature data with eight decimal places into the AD value photographed by the infrared thermal imager through the fitting function, and make a dataset. The dataset is divided into a training set and a test set. To ensure the persuasiveness of the effect, the training set adopts a mixture of simulation data and some actually photographed data, and the test set completely adopts the actually photographed data.

[0083] S3. Use the AD value dataset to train the designed recurrent neural network, automatically identify normal composite insulators and deteriorated composite insulators, and obtain a trained recurrent neural network model. Specifically:

[0084] Step 1: Designate the insulator temperature data on the side far from the conductor of the composite insulator as No. 1, and number the remaining temperature data in sequence; read the temperature data of the composite insulator, and transform the temperature data into a vector according to the insulator number. Each time, input the AD value data of the composite insulator into the recurrent neural network in sequence.

[0085] Step 2: Select the RNN recurrent neural network, add dropout to reduce overfitting, select tanh as the activation function. The final result is divided into normal ceramic insulators and deteriorated ceramic insulators, which is a binary classification problem, and sigmiod is used for classification.

[0086] Step 3: Train the training set and test the trained model with the test set to detect the training effect of the model.

[0087] S4. On the premise of knowing the resistance value, ambient temperature, humidity, and wind speed of the composite insulator, use a refrigerated infrared thermal imager carried by a drone to collect the infrared image of the composite insulator; perform image segmentation on the infrared image of the composite insulator and exclude the influence of a large amount of dust coverage or reflection factors to obtain the real-time temperature of the composite insulator.

[0088] The method for obtaining the real-time temperature of the composite insulator is as follows:

[0089] 1) Perform guided filtering on the infrared image of the composite insulator. Remove image noise and smooth the image through guided filtering to enhance image details; after guided filtering, perform detail enhancement on the image. Denote the temperature data without guided filtering as b, the temperature data after guided filtering as d, and the temperature data after detail processing as f. Then the detail processing method is:

[0090] f = b + 0.8(b - d) - min(b + 0.8(b - d));

[0091] 2) Perform image segmentation on the infrared image of the composite insulator through MaskRcnn.

[0092] 3) Binarize the segmented image to obtain a binary image. Use the binary image for contour detection to obtain the contour of the composite insulator bounding box, calculate the contour area, and set the contour area threshold to 3000 pixels. If the contour area in pixels is less than 3000, it is regarded as a false detection; exclude the part determined to be a false detection and retain the remaining bounding box coordinates; find the minimum bounding rectangle for the contours that meet the threshold requirements, further optimize the irregularly segmented bounding box to obtain a more accurate rectangular insulator segmentation box, and record the coordinates of the central axis of the composite insulator.

[0093] 4) Extract the temperature data of all points on the central axis of the composite insulator and plot it as a temperature curve of the central axis of the composite insulator.

[0094] 5) Since the reflective points occur at the peaks of the temperature curve of the central axis of the composite insulator, mark all the peaks of the temperature curve of the central axis of the composite insulator; calculate the spacing between each peak, find the regular spacing according to the multi-digit method, record this spacing as D, and compare the other peak spacing D i with D. If |D i - D| ≤ 2, the corresponding peak is also recorded as caused by reflection, and mark the serial numbers of all qualified peaks.

[0095] 6) Calculate the half-wave width of all peaks, and exclude the obviously irregular peaks according to the quartile method. Record and mark the serial numbers of the peaks that meet the half-wave peak width requirements.

[0096] 7) Perform an intersection operation on the peak serial numbers marked in steps 5) and 6). The peaks that meet both conditions in steps 5) and 6) are the peaks with elevated temperature caused by reflection; extract the peaks caused by reflection, eliminate the peaks through moving average filtering, eliminate the influence of reflection on the misjudgment of the composite insulator, and obtain the real-time temperature of the composite insulator.

[0097] Finally, input the obtained real-time temperature of the composite insulator into the trained recurrent neural network model, and the recurrent neural network model automatically determines whether the insulator is a normal composite insulator or a deteriorated composite insulator.

[0098] The technical solution of the present invention is not limited to the limitations of the above specific embodiments. Any technical deformation made according to the technical solution of the present invention falls within the protection scope of the present invention.

Claims

1. A method for automatically detecting deteriorated insulators based on infrared temperature images, characterized in that: The following steps are involved: S1. Under the premise of knowing the resistance values, ambient temperature, humidity and wind speed of normal insulators and deteriorated insulators used in the experiment, obtain infrared images of marked normal insulators and deteriorated insulators through a cooled infrared thermal imager; and extract temperature data in the infrared image to obtain the actual temperature of the insulator; wherein the insulator includes ceramic insulators and composite insulators; S2. In the comsol simulation, a simulated insulator temperature that is the same as the actual insulator temperature is simulated, and the simulated insulator temperature obtained by simulation is converted into an AD value of the actual insulator temperature, and an insulator AD value data set is prepared; S3, using the AD value data set to train the designed recurrent neural network, automatically distinguishing normal insulators and deteriorated insulators, and obtaining a trained recurrent neural network model; S4. Under the premise of known insulator resistance, ambient temperature, humidity and wind speed, the infrared image of the insulator is collected by the cooled infrared thermal imager carried by the UAV; the infrared image of the insulator is segmented and the influence of a large amount of dust cover or reflective factors is eliminated to obtain the real-time temperature of the insulator; the obtained real-time temperature of the insulator is input into the trained recurrent neural network model, and the recurrent neural network model automatically determines whether the insulator is a normal insulator or a deteriorated insulator.

2. The automatic detection method for deteriorated insulators based on infrared temperature images according to claim 1 is characterized in that: In the step S2, the simulated insulator temperature is temperature data with eight decimal places.

3. The automatic detection method for deteriorated insulators based on infrared temperature images according to claim 1 is characterized in that: In step S2, the simulated insulator temperature is converted into the AD value of the actual insulator temperature by a fitting function, and the fitting function is: <h2 style=";text-align:left;direction:ltr">y = A1x<h2 style=";text-align:left;direction:ltr"> 2 <h2 style=";text-align:left;direction:ltr"> +A2x+A3; A1, A2, and A3 are constant coefficients, which are obtained through the corresponding table between the Celsius temperature of the cooling thermal imager and the AD value.

4. The automatic detection method for deteriorated insulators based on infrared temperature images according to claim 1 is characterized in that: In step S3, the recurrent neural network is an RNN recurrent neural network, and dropout is added to reduce overfitting; tanh is selected as the activation function, and the classifier uses sigmiod for classification.

5. The automatic detection method for deteriorated insulators based on infrared temperature images according to claim 1 is characterized in that: In step S4, for the ceramic insulator, the method for obtaining the real-time temperature of the ceramic insulator is: (1) Guided filtering is performed on the infrared image of the ceramic insulator to remove image noise, smooth the image, and enhance image details. After guided filtering, the image is enhanced in detail. The temperature data before guided filtering is represented by b, the temperature data after guided filtering is represented by d, and the temperature data after detailed processing is represented by f. The detailed processing method is: f=b+0.8(bd)-min(b+0.8(bd)); (2) Use MaskRcnn to segment the infrared image of the ceramic insulator, segment the steel cap part of the ceramic insulator and record the coordinates of the steel cap selection box; (3) Binarize the segmented image to obtain a binary image, use the binary image to perform contour detection, obtain the contour of the steel cap frame, calculate the contour area, and set the contour area threshold to 30 pixels: if the contour area pixel is less than 30, it is considered a false detection; exclude the part determined as a false detection, and retain the coordinates of the remaining frame selection box; find the minimum circumscribed rectangle for the contour that meets the threshold requirement, further optimize the irregular segmentation frame, obtain a more accurate rectangular insulator segmentation frame, and record the four corner coordinates of each steel cap rectangular segmentation frame of the ceramic insulator; (4) The coordinates of the four corners of the rectangular selection box obtained by the minimum bounding rectangle method are recorded as the number of pixels within the selection box range as p i , the temperature of each pixel within the selection box is recorded as t i At this time, by comparing the absolute value of the difference between the lowest temperature and the highest temperature represented by the pixels within the same ceramic insulator steel cap selection box, if the absolute value of the temperature difference is greater than 1.5 degrees Celsius, it is judged that there is a large amount of dust covered area at this steel cap. By calculating the ratio of the temperature difference with the maximum value of more than 1.5 degrees Celsius, it is discarded. The maximum temperature in the rectangular selection box is set as T max , the number of pixels within the entire rectangular selection box is P s , lower than T max The number of pixels with a temperature of -1.5 degrees Celsius is denoted as P l , then the dust coverage ratio P is: If P ≥ 50%, discard the temperature data that is 0.8 degrees Celsius less than the maximum value. If 10% < P < 50%, discard the data that is 1 degree Celsius less than the maximum value. If P ≤ 10%, discard the data that is 1.5 degrees Celsius less than the maximum value. The remaining data is processed according to the following formula to obtain the average temperature avg (t) of the ceramic insulator steel cap, that is, the real-time temperature of the ceramic insulator: pi represents the number of pixels that meet the conditions, t i Represents the temperature data corresponding to each pixel.

6. The automatic detection method for deteriorated insulators based on infrared temperature images according to claim 1 is characterized in that: In step S4, for the composite insulator, the method for obtaining the real-time temperature of the composite insulator is: 1) Guided filtering is performed on the infrared image of the composite insulator to remove image noise, smooth the image, and enhance image details. After guided filtering, the image is enhanced in detail. The temperature data that has not been processed by guided filtering is represented by b, the temperature data that has been processed by guided filtering is represented by d, and the temperature data after detailed processing is represented by f. The detailed processing method is: f=b+0.8(bd)-min(b+0.8(bd)); 2) Perform image segmentation on the infrared image of the composite insulator using MaskRcnn; 3) Binarize the segmented image to obtain a binary image, use the binary image to perform contour detection, obtain the contour of the composite insulator selection box, calculate the contour area, and set the contour area threshold to 3000 pixels. If the contour area pixel is less than 3000, it is considered a false detection; exclude the part determined as a false detection, and retain the coordinates of the remaining selection box; find the minimum circumscribed rectangle for the contour that meets the threshold requirement, further optimize the irregular selection box, obtain a more accurate rectangular insulator segmentation box, and record the coordinates of the composite insulator centerline; 4) Extracting the temperature data of all points on the center axis of the composite insulator and drawing a temperature curve of the center axis of the composite insulator; 5) Because the reflection point occurs at the peak of the temperature curve of the center axis of the composite insulator, mark all the peaks of the temperature curve of the center axis of the composite insulator; calculate the spacing between each peak, find the regular spacing according to the multi-digit method, record this spacing as D, and record the spacing of other peaks as D i Compared with D, if |D i -D|≤2, the corresponding peak is also recorded as caused by reflection, and all peaks that meet the conditions are marked with serial numbers; 6) Calculate the half-wave width of all peaks, exclude obviously irregular peaks according to the quartile method, and record and mark the peaks that meet the half-wave peak width requirements; 7) The peak marking numbers of step 5) and step 6) are subjected to the same operation, and the peak that satisfies both conditions of step 5) and step 6) is the peak with a higher temperature caused by reflection; The peak caused by reflection is extracted and eliminated through sliding average filtering, thereby eliminating the influence of reflection on the misjudgment of composite insulators and obtaining the real-time temperature of the composite insulator.

Images