An insulator infrared image fuzzy edge processing method with heuristic learning ability

Abstract

The application discloses an insulator infrared image fuzzy edge processing method with heuristic learning ability, and comprises the following steps: training a heuristic-constraint system and an edge processing network; acquiring an insulator infrared image to be measured; deriving an insulator infrared image temperature matrix to be measured; processing the insulator infrared image temperature matrix to be measured; calculating a primary edge feature matrix of the insulator infrared image temperature matrix to be measured; normalizing the double matrix to be measured; executing the edge processing network to obtain a processing result; the processing method generates a feature norm and a weighted sample-label group driven by a random vector through the heuristic-constraint system, continuously trains the edge processing network by using the weighted sample-label group, and uses the edge processing network to eliminate the fuzzy edge area of the insulator infrared image and realizes edge detection. The application has high technical advancement and wide applicability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence systems and visual inspection technology for power equipment in the power production field, specifically to a method for processing blurred edges of infrared images of insulators based on a heuristic-constraint system that enables heuristic learning. Background Technology

[0002] Insulators are crucial basic insulating components in the power transmission, distribution, and transformation processes of a power system. They provide electrical insulation and mechanical support between energized conductors and power poles, and their insulation effectiveness directly impacts the normal operation of the power grid. With increasing operating time, under the combined effects of electromechanical loads and environmental factors, their insulation and mechanical properties gradually decline, leading to insulator aging and deterioration. This can even result in serious accidents such as insulator bursting or string failure, causing widespread power outages and threatening the safe and stable operation of the power grid.

[0003] Traditional insulator inspection methods involve manual tower climbing. To improve inspection efficiency, reduce labor intensity, and ensure the safety of inspection personnel, infrared thermography is currently the most effective alternative. Automatically locating, identifying, and segmenting the insulator region within the infrared image is crucial for intelligently detecting insulator degradation. Edge detection is an important method for obtaining the position and geometry of insulator strings and is a vital prerequisite for image segmentation.

[0004] In practical applications, due to the poor contrast and single channel features of infrared images, the edge features of insulators in infrared images are often blurred. Currently, there is no effective method to eliminate the blurred areas at the edges of insulator infrared images. Summary of the Invention

[0005] The purpose of this invention is to provide a method for processing blurred edges in infrared images of insulators with heuristic learning capabilities, so as to solve the problems of edge loss, blurred areas and inaccurate edge positions that occur when detecting the edges of infrared images of insulator strings in the background art.

[0006] To achieve the above objectives, the present invention provides a method for processing blurred edges in infrared images of insulators with heuristic learning capabilities, characterized by comprising the following steps:

[0007] S1. Training Heuristic-Constraint System and Edge Processing Network;

[0008] S2. Obtain infrared images of the insulator under test using infrared imaging equipment;

[0009] S3. Export the infrared image temperature matrix of the insulator to be tested obtained in step S2;

[0010] S4. Process the temperature matrix to be measured exported in step S3;

[0011] S5. Calculate the primary edge feature matrix of the temperature matrix to be measured;

[0012] S6. Normalization of the two matrices under test;

[0013] S7. Execute the edge processing network to obtain the processing results.

[0014] Further, step S1 specifically involves: collecting a certain number of infrared images of insulators as a training sample set; exporting the temperature matrix of the training sample set using an infrared image acquisition device, and processing the temperature matrix of the training sample set; calculating and normalizing the primary edge feature matrix of the trained training sample set, and passing the normalized primary edge feature matrix of the training sample set to the paradigm generator, which provides input paradigm examples to the heuristic network and the constraint network respectively; after receiving the input paradigm examples, the heuristic network and the constraint network are trained with boundary labels and sample-label groups as convergence targets respectively; using the sample judgment network and the label judgment network to perform adversarial training with the heuristic network and the constraint network respectively, and using the qualified sample-label groups generated by the heuristic network for training the edge processing network.

[0015] Furthermore, processing the temperature matrix of the training sample set includes scaling and cropping the temperature matrix, which is calculated by the following formula (1):

[0016]

[0017] In equation (1), γ is the scaling factor, l sx To preset the image width, l sy To preset the image height, l x l is the width of the temperature matrix. y T represents the height of the temperature matrix. r T is the scaled temperature matrix, and T is the unscaled temperature matrix. f This is the temperature matrix after clipping.

[0018] Furthermore, the calculation method and normalization method of the primary edge feature matrix of the training sample set are as follows: Equations (2) and (3), respectively:

[0019] B f =ΔT f (2);

[0020]

[0021] In equations (2) and (3) above, B f B is the primary edge feature matrix, where Δ is the Laplacian operator; c This is the normalized primary edge feature matrix.

[0022] Furthermore, the paradigm generator generates paradigm examples using the following formula:

[0023] N s =N(μ,σ) 2 (4);

[0024]

[0025] S=(1-α)B c +αN f (6);

[0026] In equations (4) to (6) above, N s Let N be a normally distributed random number, μ be the mean of the normal distribution, σ be the standard deviation of the normal distribution, and N be the mean of the normal distribution. f The output is a truncated random number, α is the noise level, and S is the generation paradigm.

[0027] Furthermore, the heuristic network and the constraint network are trained with boundary labels and sample-label groups as convergence targets, respectively. The heuristic network is an artificial neural network whose input is the feature paradigm generated by the paradigm generator and whose output is the restored normalized temperature matrix and boundary labels. The constraint network is an artificial neural network whose input is the feature paradigm generated by the paradigm generator and whose output is the boundary labels. The heuristic network and the constraint network are implemented using convolutional neural networks or variational autoencoder methods.

[0028] Furthermore, the sample judgment network and label judgment network are respectively trained adversarially against the heuristic network and the constraint network. The sample judgment network is an artificial neural network whose input is the sample output by the heuristic network and the real sample, and whose output is the true probability of the sample. The label judgment network is an artificial neural network whose input is the boundary label output by the heuristic network, the boundary label output by the constraint network, and the real boundary label, and whose output is the true probability of the label. The sample judgment network and the label judgment network are implemented using convolutional neural networks or backpropagation neural networks.

[0029] Furthermore, the heuristic network generates qualified sample-label sets for training the edge processing network, where qualified sample-label sets are those that both the sample judgment network and the label judgment network fail to judge.

[0030] Furthermore, consistent with step S1, the processing of the temperature matrix to be measured in step S4 also includes scaling and clipping of the temperature matrix, and the scaling and clipping method is also calculated using equation (1); the calculation of the primary edge feature matrix of the temperature matrix to be measured in step S5 is also calculated using equation (2).

[0031] Furthermore, the normalization of the two matrices to be tested in step S6 includes the normalization of the clipped temperature matrix and the normalization of the primary edge feature matrix. The normalization method is as follows:

[0032]

[0033] In equation (7) above, T c The normalized temperature matrix, t a The preset starting temperature value, t b To preset the endpoint temperature, B c This is the normalized primary edge feature matrix.

[0034] Furthermore, in step S7, an edge processing network is executed to obtain the processing result, and the input of the edge processing network is T obtained in step S6. c With B c It is trained by generating qualified sample-label groups through a heuristic network, and the processing result is the probability distribution matrix of the insulator edge position; the edge processing network is implemented using convolutional neural network and variational autoencoder method.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] This invention provides a method for blurry edge processing of insulator infrared images with heuristic learning capabilities. It generates a feature paradigm and a weighted sample-label group driven by random vectors through a heuristic-constraint system. The weighted sample-label group is then used to continuously train an edge processing network, which is used to eliminate blurry edge regions in the insulator infrared image and achieve edge detection. This invention has high technical advancement and wide applicability.

[0037] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0038] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0039] Figure 1 This is a flowchart illustrating the insulator infrared image blur edge processing method with heuristic learning capability according to the present invention.

[0040] Figure 2 This is an infrared image of an insulator used in a preferred embodiment of the present invention;

[0041] Figure 3 The results are obtained from the preferred embodiment of the present invention. Detailed Implementation

[0042] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings. However, it should be noted that these embodiments are not intended to limit the present invention. Equivalent transformations or substitutions in function, method, or structure made by those skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0043] Please see Figure 1 This embodiment provides a method for processing blurred edges in infrared images of insulators with heuristic learning capabilities, characterized by the following steps:

[0044] S1. Training the heuristic-constraint system and edge processing network; specific steps include:

[0045] A certain number of infrared images of insulators are collected as a training sample set. The temperature matrix of the training sample set is exported using an infrared image acquisition device and processed. The primary edge feature matrix of the training sample set after training is calculated and normalized. The normalized primary edge feature matrix of the training sample set is then passed to the paradigm generator, which provides input paradigm examples to the heuristic network and the constraint network, respectively. After receiving the input paradigm examples, the heuristic network and the constraint network are trained with boundary labels and sample-label groups as convergence targets, respectively. The sample judgment network and the label judgment network are used to perform adversarial training with the heuristic network and the constraint network, respectively. The qualified sample-label groups generated by the heuristic network are then used for training the edge processing network.

[0046] In one specific implementation, step S1, processing the temperature matrix of the training sample set includes scaling and cropping the temperature matrix, which is calculated by the following formula (1):

[0047]

[0048] In equation (1), γ is the scaling factor, l sx To preset the image width, l sy To preset the image height, l x l is the width of the temperature matrix. y T represents the height of the temperature matrix. r T is the scaled temperature matrix, and T is the unscaled temperature matrix. f This is the trimmed temperature matrix. In this embodiment, l sx For 320, l sy For 240, l x For 640, l y It is 480.

[0049] In one specific implementation, the calculation method and normalization method of the primary edge feature matrix of the training sample set are as follows: (2) and (3), respectively:

[0050] B f =ΔT f (2);

[0051]

[0052] In equations (2) and (3) above, B f B is the primary edge feature matrix, where Δ is the Laplacian operator; c This is the normalized primary edge feature matrix.

[0053] In one specific implementation, the paradigm generator generates paradigm examples using the following formula:

[0054] N s =N(μ,σ) 2 (4);

[0055]

[0056] S=(1-α)B c +αN f (6);

[0057] In equations (4) to (6) above, N s and N s Let N be a normally distributed random number, μ be the mean of the normal distribution, σ be the standard deviation of the normal distribution, and N be the mean of the normal distribution. f The output is a truncated random number, α is the noise level, and S is the generation paradigm. In this embodiment, α is 0.3, μ is 0.5, and σ is 0.4.

[0058] In one specific implementation, the heuristic network and the constraint network are trained with boundary labels and sample-label sets as convergence targets, respectively. The heuristic network is an artificial neural network whose input is the feature paradigm generated by the paradigm generator, and whose output is the restored normalized temperature matrix and boundary labels. The constraint network is an artificial neural network whose input is the feature paradigm generated by the paradigm generator, and whose output is the boundary labels. The heuristic network and the constraint network are implemented using convolutional neural networks or variational autoencoders. The sample judgment network and the label judgment network are used for adversarial training against the heuristic network and the constraint network, respectively. The sample judgment network takes as input the samples output by the heuristic network and the real samples, and whose output is the true probability of the samples. The label judgment network takes as input the boundary labels output by the heuristic network, the boundary labels output by the constraint network, and the real boundary labels, and whose output is the true probability of the labels. The sample judgment network and the label judgment network are implemented using convolutional neural networks or backpropagation neural networks. The qualified sample-label sets generated by the heuristic network are used for training the edge processing network, where qualified sample-label sets are those that both the sample judgment network and the label judgment network fail to classify.

[0059] S2. Obtain infrared images of the insulator to be tested using infrared imaging equipment; the obtained infrared images of the insulator to be tested are the images to be inspected, and their quality should meet the requirements of relevant industry standards for infrared thermal imaging inspection of insulators.

[0060] S3. Export the infrared image temperature matrix of the insulator under test; specifically, the infrared image temperature matrix of the insulator under test can be exported through the relevant extended functions provided by the infrared image capturing equipment.

[0061] S4. The temperature matrix of the infrared image of the insulator to be tested exported in step S3 is processed, specifically including matrix scaling and cropping. The specific processing method is also as follows (1):

[0062]

[0063] In equation (1), γ is the scaling factor, l sx To preset the image width, l sy To preset the image height, l x l is the width of the temperature matrix. y T represents the height of the temperature matrix. r T is the scaled temperature matrix, and T is the unscaled temperature matrix. f This is the trimmed temperature matrix. In this embodiment, l sx For 320, l sy For 240, l x For 640, l y It is 480.

[0064] S5. Calculate the primary edge feature matrix of the temperature matrix to be measured; specifically, calculate the primary edge feature matrix of the clipped temperature matrix described in step S4. The calculation method for the primary edge feature matrix is ​​also as follows:

[0065] B f =ΔT f (2);

[0066] In the above formula, B f Let be the primary edge feature matrix, and Δ be the Laplacian operator.

[0067] S6. Normalization of the two matrices to be tested; specifically, this includes the normalization of the temperature matrix clipped in step S5 and the normalization of the primary edge feature matrix. The normalization method is as follows:

[0068]

[0069] In equation (7) above, T c The normalized temperature matrix, t a The preset starting temperature value, t b To preset the endpoint temperature, B c This is the normalized primary edge feature matrix. In this embodiment, t a At 5℃, t b The temperature is 45℃.

[0070] S7. Execute the edge processing network to obtain the processing result, where the input of the edge processing network is T obtained in step S6. c With B c It is trained by generating qualified sample-label groups through a heuristic network, and the processing result is the probability distribution matrix of the insulator edge position; the edge processing network is implemented using methods such as convolutional neural networks and variational autoencoders.

[0071] Using the above steps of the embodiments of the present invention to Figure 2 The infrared image of the insulator in the image is processed, and the resulting edge processing result is as follows: Figure 3 As shown.

[0072] In this invention, the artificial neural networks are not limited to the methods mentioned in this invention; other methods that meet the functional requirements of this invention are also applicable.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for processing blurred edges in infrared images of insulators with heuristic learning capability, characterized in that, The steps include the following: S1. Training the heuristic-constraint system and edge processing network; specifically including: collecting a certain number of infrared images of insulators as a training sample set; exporting the temperature matrix of the training sample set using infrared image acquisition equipment and processing the temperature matrix of the training sample set; calculating and normalizing the primary edge feature matrix of the training sample set after training, and passing the normalized primary edge feature matrix of the training sample set to the paradigm generator, which provides input paradigm examples to the heuristic network and the constraint network respectively; after receiving the input paradigm examples, the heuristic network and the constraint network are trained with boundary labels and sample-label groups as convergence targets respectively; using the sample judgment network and the label judgment network to perform adversarial training with the heuristic network and the constraint network respectively, and generating qualified samples by the heuristic network. - The label group is used for training the edge processing network. The constraint network is an artificial neural network that takes the feature paradigm generated by the paradigm generator as input and outputs the boundary labels. The heuristic and constraint networks are implemented using convolutional neural networks or variational autoencoders. The sample judgment network takes the sample output from the heuristic network and the real sample as input and outputs the true probability of the sample. The label judgment network takes the boundary labels output from the heuristic network, the boundary labels output from the constraint network, and the real boundary labels as input and outputs the true probability of the label. The sample judgment network and the label judgment network are implemented using convolutional neural networks or backpropagation neural networks. A qualified sample-label group is the sample-label group that both the sample judgment network and the label judgment network fail to classify. S2. Obtain infrared images of the insulator under test using infrared imaging equipment; S3. Export the temperature matrix of the infrared image of the insulator to be tested obtained in step S2; S4. Process the temperature matrix to be measured exported in step S3; S5. Calculate the primary edge feature matrix of the temperature matrix to be measured; S6. Normalize the dual matrix to be measured in step S4 and the primary edge feature matrix in step S5. S7. Input the temperature matrix and primary edge feature matrix normalized in step S6 into the edge processing network trained in step S1, and execute the edge processing network to obtain the processing result.

2. The processing method according to claim 1, characterized in that, The temperature matrix of the training sample set is processed by scaling and cropping the temperature matrix. The scaling and cropping are calculated by the following formula (1): (1)； In formula (1) For scaling factor, Preset image width, Preset image height, The width of the temperature matrix. For the height of the temperature matrix, This is the scaled temperature matrix. This is the temperature matrix before scaling. This is the temperature matrix after clipping.

3. The processing method according to claim 2, characterized in that, The calculation and normalization methods of the primary edge feature matrix of the training sample set are as follows: Equations (2) and (3) are respectively: (2)； (3)； In equations (2) and (3) above, This is the primary edge feature matrix. For the Laplace operator; This is the normalized primary edge feature matrix.

4. The processing method according to claim 1, characterized in that, The paradigm generator generates paradigm examples using the following formula: (4)； (5)； (6)； In equations (4) to (6) above Normally distributed random numbers, The mean of a normal distribution is . The standard deviation of the normal distribution. The output is a truncated random number. For noise level, This is the generated paradigm.

5. The processing method according to claim 1, characterized in that, The normalization of the two matrices to be tested in step S6 includes the normalization of the clipped temperature matrix and the normalization of the primary edge feature matrix. The normalization method is as follows: (7)； In equation (7) above, The normalized temperature matrix, To preset the starting temperature value, To preset the endpoint temperature, This is the normalized primary edge feature matrix.

6. The processing method according to claim 5, characterized in that, In step S7, the edge processing network is executed to obtain the processing result. The input of the edge processing network is the result obtained in step S6. and It is trained by generating qualified sample-label groups through a heuristic network, and the processing result is the probability distribution matrix of the insulator edge position; the edge processing network is implemented using convolutional neural network and variational autoencoder method.

Images