Electric power inspection image super-resolution reconstruction method based on SRGAN

Abstract

The invention discloses an SRGAN-based electric power inspection image super-resolution reconstruction method, and relates to the technical field of image processing, and the method comprises the steps: collecting a high-resolution image of electric power equipment, constructing a data set suitable for electric power inspection, and carrying out the preprocessing; inputting the preprocessed low-resolution image into an SRGAN model, and training a generator and a discriminator in the SRGAN model; and inputting the test set and the verification set into the SRGAN model, and testing and verifying the trained SRGAN model. According to the method, the diversity and representativeness of the data set are ensured by constructing the high-resolution image data set specially used for electric power inspection, effective extraction and reconstruction of image features are realized by training the SRGAN model, the fidelity of the reconstructed image is improved, and finally, the visual quality and details of the image are improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a super-resolution reconstruction method for power inspection images based on SRGAN. Background Art

[0002] The key to ensuring the safe operation of power systems lies in effective power detection. However, traditional detection technologies face huge challenges in image resolution and are therefore unable to accurately detect finer texture details in electrical equipment. Super-resolution reconstruction technology aims to reconstruct high-resolution images from low-resolution images, inferring high-frequency details by leveraging the correlation and structural information between image pixels, thereby improving the visual quality and detail presentation of images.

[0003] Although SRCNN, ESPCN, VDSR and other methods have made some progress in the field of super-resolution reconstruction, these methods will be over-smoothed when the magnification exceeds 4 times, resulting in a lack of realism in the image. In addition, when processing power inspection images, the existing technology generally has problems such as insufficient reconstructed image details and loss of high-frequency details, which limits the accurate identification of minor defects and faults in power equipment. Therefore, there is an urgent need for a super-resolution reconstruction method that can maintain high-frequency details and texture details of the image at a large scale factor to improve the quality and detection accuracy of power inspection images. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a super-resolution reconstruction method for power inspection images based on SRGAN to solve the limitation problem of super-resolution reconstruction method in high-frequency detail restoration.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for super-resolution reconstruction of power inspection images based on SRGAN, which comprises:

[0008] Collect high-resolution images of power equipment, perform preprocessing, and construct a data set suitable for power inspection;

[0009] Input the prepared training set into the SRGAN model to train the generator and discriminator in the SRGAN model;

[0010] The test set is input into the SRGAN model to test and verify the trained SRGAN model.

[0011] As a preferred solution of the SRGAN-based super-resolution reconstruction method for power inspection images of the present invention, the specific steps of collecting high-resolution images of power equipment, performing data processing, and constructing a data set suitable for power inspection are as follows:

[0012] Photographing the power equipment, cables, distribution boxes, transformers, and insulators in a large substation;

[0013] The shooting process was carried out in different seasons, different weather conditions, and different time periods of the day;

[0014] The acquired high-resolution images are organized into a dataset dedicated to power inspection and divided into a training set, a test set, and a validation set;

[0015] The dataset is preprocessed by cropping and downsampling while preserving the corresponding original high-resolution images.

[0016] As a preferred solution of the SRGAN-based power inspection image super-resolution reconstruction method of the present invention, the pre-processed low-resolution image is input into the SRGAN model, and the generator and discriminator in the SRGAN model are trained. The specific steps are as follows:

[0017] Input the cropped and downsampled low-resolution images in the training set into the SRGAN model;

[0018] The generator in the SRGAN model reconstructs the preprocessed low-resolution images in the training set through convolution operations, and the discriminator in the SRGAN model distinguishes the reconstructed high-resolution images from the original high-resolution images;

[0019] At the beginning, the generator's generation effect is relatively poor, and the discriminator can easily judge it. The judgment result is fed back to the generator to optimize the generator, causing the high-resolution image reconstructed by the generator to be closer and closer to the original high-resolution image. The discriminator's judgment error will become larger. Then the discriminator is optimized based on the judgment error, making it difficult for the generator to deceive the discriminator, so the generator is optimized in turn;

[0020] During the entire training process, the generator and the discriminator continue to compete with each other in a cycle, and finally reach a Nash equilibrium.

[0021] As a preferred solution of the SRGAN-based power inspection image super-resolution reconstruction method of the present invention, the generator in the SRGAN model reconstructs the preprocessed low-resolution image through a convolution operation, and the specific steps are as follows:

[0022] Using skip connections in the high feature extraction layer composed of B residual blocks as the core of the generated network to enhance the flow of information across layers;

[0023] Two 3×3 convolution kernels and 64-channel convolution layers are used at the same time. A batch normalization layer is used after each convolution layer. Finally, ParametricReLU is used as the activation function and Elementwise Sum operation is used.

[0024] The deconvolution layer uses two trained sub-pixel convolutional layers to increase the input image resolution;

[0025] The reconstruction layer is responsible for reconstructing the final high-resolution image.

[0026] As a preferred solution of the SRGAN-based power inspection image super-resolution reconstruction method of the present invention, the discriminator in the SRGAN model distinguishes the reconstructed high-resolution image from the original high-resolution image, and the specific steps are as follows:

[0027] The activation function in the discriminator's network structure uses Leaky-ReLU;

[0028] Starting from the second convolutional layer of the 8 convolutional layers in the discriminator's network structure, a BN layer is added to each convolutional layer. After each strided convolution, the image resolution is reduced and the number of feature maps is doubled.

[0029] The final output layer uses two fully connected layers and activation functions to score the reconstructed high-resolution image and the original high-resolution image;

[0030] After each training session, the hyperparameters, learning rate, weight decay coefficient, and momentum are adjusted according to the performance on the validation set until the optimal hyperparameter combination is found.

[0031] As a preferred solution of the SRGAN-based power inspection image super-resolution reconstruction method of the present invention, wherein: the generator and the discriminator are optimized by calculating the loss function, and then trained again until they are indistinguishable. The specific steps are as follows:

[0032] The high-resolution image reconstructed by the generator and the original high-resolution image are sent to the VGG19Net network for feature extraction, and the root mean square error of the extracted high semantic features is calculated. The expression is as follows:

[0033]

[0034] in, refers to the content loss function, is the generator defined, θ G is the generator parameter, I HR is the reconstructed high-resolution image, Φ i，jrepresents the jth convolutional layer before the i-th maximum pooling of VGG19Net, W i，j and H i，j is the dimension of VGG19Net, i is the index variable of the convolutional layer in the VGG19Net network, and j is the feature map index variable in the convolutional layer;

[0035] By calculating the adversarial loss, we can obtain the high-resolution image reconstructed by the discriminator generator. The probability of being true is expressed as follows:

[0036]

[0037] in To combat losses, is the discriminator defined, θ D is the discriminator parameter, I LR is the input low-resolution image, N is the number of images;

[0038] Combining content loss and adversarial loss into a perceptual loss function;

[0039] The parameters of the discriminator are fixed and the parameters of the generator are adjusted to train the generator.

[0040] The parameters of the generator are fixed and the values ​​of the parameters of the discriminator are adjusted. The discriminator is trained by maximizing the score of the original high-resolution image in the discriminator and minimizing the score of the reconstructed high-resolution image.

[0041] Through adversarial training of the generator and the discriminator and alternating optimization, the generator finally reconstructs a high-resolution image that the discriminator cannot distinguish.

[0042] As a preferred solution of the SRGAN-based power inspection image super-resolution reconstruction method of the present invention, the content loss and the adversarial loss are combined into a perceptual loss function, and the specific steps are as follows:

[0043] The SRGAN model combines the weighted sum of content loss and adversarial loss into perceptual loss, which is expressed as follows:

[0044]

[0045] in, is the content loss defined, l SR is the perceptual loss function.

[0046] As a preferred solution of the SRGAN-based power inspection image super-resolution reconstruction method of the present invention, wherein: the test set is input into the SRGAN model, and the trained SRGAN model is tested, and the specific steps are as follows:

[0047] The test set is input into the trained SRGAN model to perform peak signal-to-noise ratio and structural similarity index tests;

[0048] The high-resolution images reconstructed by the generator have good visual effects in terms of peak signal-to-noise ratio and structural similarity, and the generator and discriminator reach Nash equilibrium, indicating that the SRGAN model is successful.

[0049] In a second aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the SRGAN-based power inspection image super-resolution reconstruction method as described in the first aspect of the present invention is implemented.

[0050] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the SRGAN-based power inspection image super-resolution reconstruction method as described in the first aspect of the present invention is implemented.

[0051] The beneficial effects of the present invention are as follows: the present invention ensures the diversity and representativeness of the data set by capturing images of power equipment in different seasons, weather conditions and time periods, and then crops and downsamples the images to construct a high-resolution image data set specifically for power inspection. Through the training of the SRGAN model, effective extraction and reconstruction of image features are achieved, the fidelity of the reconstructed image is improved, and finally the image visual quality and details are improved. By calculating the perceptual loss and guiding the training and optimization of the SRGAN model, the dual optimization of the generated image in terms of content and vision is achieved, and the realism and visual quality of the image reconstruction are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0053] Figure 1 This is a flow chart of the super-resolution reconstruction method of power inspection images based on SRGAN in Example 1.

[0054] Figure 2 Flowchart for training and optimizing the generator and discriminator guided by the loss function in Example 1. DETAILED DESCRIPTION

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0058] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for super-resolution reconstruction of power inspection images based on SRGAN, comprising the following steps:

[0059] S1. Collect high-resolution images of power equipment, build a dataset suitable for power inspection, and perform preprocessing.

[0060] Furthermore, we photographed the power equipment, cables, distribution boxes, transformers, and insulators in a large substation to ensure that the data set reflects the actual situation of the power equipment;

[0061] The shooting process was carried out in different seasons, weather conditions, and time periods of the day to ensure the diversity and representativeness of the dataset;

[0062] Organize the acquired high-resolution images into a dataset dedicated to power inspection;

[0063] Furthermore, the dataset is preprocessed by cropping and downsampling, while the corresponding original high-resolution images are saved and divided into training set, test set, and validation set;

[0064] 80% of the data set is divided into training set, 10% of the data set is divided into test set, and 10% of the data set is divided into validation set;

[0065] The selected images in the dataset are cropped to 2040 pixels on one side and a multiple of 4 pixels on the other side, and then the cropped images are downsampled by 4 times using bicubic interpolation;

[0066] The test set is collected. The images obtained must be of different sizes, but the length and width of the images must be multiples of 4. 20% of the images with rich texture information are selected from the cropped images to form a new test set. Finally, the selected images are downsampled by 4 times using the bicubic interpolation method.

[0067] The preprocessing method of the validation set is the same as that of the training set.

[0068] S2. The preprocessed low-resolution image and the high-resolution image are input into the SRGAN model together to train the generator and discriminator in the SRGAN model.

[0069] Furthermore, SRGAN (Super-Resolution Generative Adversarial Network) is a deep learning model based on adversarial training, specifically for image super-resolution, which includes two key networks: generator and discriminator;

[0070] The generator and discriminator in the SRGAN model are trained by inputting the cropped and downsampled low-resolution images in the training set into the SRGAN model.

[0071] Furthermore, the generator in the SRGAN model reconstructs the preprocessed low-resolution images in the training set through convolution operations;

[0072] The network structure of the generator is divided into low feature extraction layer, high feature extraction layer, deconvolution layer and reconstruction layer;

[0073] The core of the entire network is a high-feature extraction layer composed of B residual blocks, which uses skip connections to enhance the flow of information across layers and prevent the gradient vanishing problem caused by the deepening of the network;

[0074] Two 3×3 convolution kernels and 64-channel convolution layers are used at the same time. A batch normalization layer is used after each convolution layer. Finally, ParametricReLU is used as the activation function and Elementwise Sum operation is used.

[0075] The deconvolution layer uses two trained sub-pixel convolutional layers to increase the input image resolution;

[0076] Furthermore, the discriminator in the SRGAN model distinguishes between the reconstructed high-resolution image and the original high-resolution image;

[0077] The activation function in the discriminator network structure uses Leaky-ReLU (α=0.2) to prevent some negative output necrosis;

[0078] Starting from the second convolutional layer of the 8 convolutional layers in the discriminator's network structure, a BN layer is added to each convolutional layer to improve network stability and running speed. After each strided convolution, the image resolution is reduced and the number of feature maps is doubled.

[0079] Finally, the output layer uses two fully connected layers and an activation function, which is equivalent to a binary classification network. It scores the reconstructed high-resolution image and the original high-resolution image. Its output is a scalar value. The output scalar value of the reconstructed high-resolution image is 1, and the output scalar value of the original high-resolution image is 0.

[0080] After each training session, use the validation set to evaluate the performance of the model. Adjust the hyperparameters, learning rate, weight decay coefficient, and momentum according to the performance on the validation set until the optimal hyperparameter combination is found. This process can help prevent the model from overfitting the training data and improve the performance of the model on unseen data.

[0081] Furthermore, the high-resolution image reconstructed by the generator and the original high-resolution image are fed into the VGG19Net network for feature extraction, and the root mean square error of the extracted high semantic features is calculated. The expression is as follows:

[0082]

[0083] in, For the defined content loss, is the generator defined, θ G is the generator parameter, which represents the weight and bias of the L layer of the generated network, I HR is the reconstructed high-resolution image, Φ i，j represents the jth convolutional layer before the i-th maximum pooling of VGG19Net, W i，j and H i，j is the dimension of VGG19Net, i is the index variable of the convolutional layer in the VGG19Net network, and j is the feature map index variable in the convolutional layer;

[0084] Furthermore, adversarial loss is used to enhance the realism of reconstructed high-resolution images and improve the visual quality of reconstructed high-resolution images. By calculating adversarial loss, the generated high-resolution images reconstructed by the discriminator generator are obtained. The probability and expression of being true are as follows:

[0085]

[0086] in, is the adversarial loss defined, is the discriminator defined, θ D is the discriminator parameter, I LRis the input low-resolution image, N is the number of images;

[0087] Furthermore, the SRGAN model combines the weighted sum of content loss and adversarial loss into perceptual loss, which improves the realism and visual quality of the image while preserving the image content.

[0088]

[0089] in, is the content loss defined, l SR is the perceptual loss function;

[0090] Furthermore, the parameters of the discriminator are fixed and the values ​​of the parameters of the generator are adjusted to train the generator;

[0091] The parameters of the generator are fixed and the values ​​of the parameters of the discriminator are adjusted. The discriminator is trained by maximizing the score of the original high-resolution image in the discriminator and minimizing the score of the reconstructed high-resolution image.

[0092] At the beginning, the generator's generation effect is relatively poor, and the discriminator can easily judge it. The judgment result is fed back to the generator to optimize the generator, causing the "fake data" generated by the generator to be closer and closer to the real data, and the discriminator's judgment error will become larger. Then the discriminator is optimized based on the judgment error, making it difficult for the generator to deceive the discriminator, so the generator is optimized in turn. During the entire training process, the generator and the discriminator constantly compete with each other, and finally reach a Nash equilibrium.

[0093] S3. Input the test set into the SRGAN model to test the trained SRGAN model.

[0094] Furthermore, the test set is input into the trained SRGAN model to test the peak signal-to-noise ratio and structural similarity index;

[0095] The high-resolution images reconstructed by the generator have good visual effects in terms of peak signal-to-noise ratio and structural similarity, and the generator and discriminator reach Nash equilibrium, indicating that the SRGAN model is successful.

[0096] This embodiment also provides a computer device, which is suitable for the case of the SRGAN-based power inspection image super-resolution reconstruction method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the SRGAN-based power inspection image super-resolution reconstruction method proposed in the above embodiment.

[0097] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0098] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for super-resolution reconstruction of power inspection images based on SRGAN proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0099] In summary, the present invention takes images of power equipment in different seasons, weather conditions and time periods to ensure the diversity and representativeness of the data set, then crops and downsamples the images to construct a high-resolution image data set specifically for power inspection. Through the training of the SRGAN model, effective extraction and reconstruction of image features are achieved, the fidelity of the reconstructed image is improved, and finally the visual quality and details of the image are improved. By calculating the perceptual loss and guiding the training and optimization of the SRGAN model, the dual optimization of the generated image in terms of content and vision is achieved, and the realism and visual quality of the image reconstruction are improved.

Claims

1. A super-resolution reconstruction method for power inspection images based on SRGAN, characterized by: include, Collect high-resolution images of power equipment, construct a data set suitable for power inspection, and perform preprocessing; The preprocessed low-resolution image and high-resolution image are input into the SRGAN model together to train the generator and discriminator in the SRGAN model; The test set is input into the SRGAN model to test the trained SRGAN model.

2. The method for super-resolution reconstruction of power inspection images based on SRGAN as claimed in claim 1, characterized in that: The specific steps of collecting high-resolution images of power equipment, constructing a data set suitable for power inspection, and preprocessing are as follows: Photographing the power equipment, cables, distribution boxes, transformers, and insulators in a large substation; The shooting process was carried out in different seasons, different weather conditions, and different time periods of the day; The acquired high-resolution images are organized into a dataset dedicated to power inspection and divided into a training set, a test set, and a validation set; The dataset is preprocessed by cropping and downsampling while preserving the corresponding original high-resolution images.

3. The method for super-resolution reconstruction of power inspection images based on SRGAN as claimed in claim 2, characterized in that: The preprocessed low-resolution image and high-resolution image are input into the SRGAN model together, and the generator and discriminator in the SRGAN model are trained. The specific steps are as follows: Input the cropped and downsampled low-resolution images in the training set into the SRGAN model; The generator in the SRGAN model reconstructs the preprocessed low-resolution images in the training set through convolution operations, and the discriminator in the SRGAN model distinguishes the reconstructed high-resolution images from the original high-resolution images; At the beginning, the generator's generation effect is relatively poor, and the discriminator can easily judge it. The judgment result is fed back to the generator to optimize the generator, causing the high-resolution image reconstructed by the generator to be closer and closer to the original high-resolution image. The discriminator's judgment error will become larger. Then the discriminator is optimized based on the judgment error, making it difficult for the generator to deceive the discriminator, so the generator is optimized in turn; During the entire training process, the generator and the discriminator continue to compete with each other in a cycle, and finally reach a Nash equilibrium.

4. The method for super-resolution reconstruction of power inspection images based on SRGAN as claimed in claim 3, characterized in that: The generator in the SRGAN model reconstructs the preprocessed low-resolution image through convolution operation. The specific steps are as follows: Residual links are used in the high feature extraction layer composed of B residual blocks as the core of the generated network to enhance the flow of information across layers; Two 3×3 convolution kernels and 64-channel convolution layers are used at the same time. A batch normalization layer is used after each convolution layer. Finally, ParametricReLU is used as the activation function and Elementwise Sum operation is used. The deconvolution layer uses two trained sub-pixel convolutional layers to increase the input image resolution.

5. The method for super-resolution reconstruction of power inspection images based on SRGAN as claimed in claim 4, characterized in that: The discriminator in the SRGAN model distinguishes the reconstructed high-resolution image from the original high-resolution image. The specific steps are as follows: The activation function in the discriminator's network structure uses Leaky-ReLU; Starting from the second convolutional layer of the 8 convolutional layers in the discriminator's network structure, a BN layer is added to each convolutional layer. After each strided convolution, the image resolution is reduced and the number of feature maps is doubled. The final output layer uses two fully connected layers and activation functions to score the reconstructed high-resolution image and the original high-resolution image; After each training session, the hyperparameters, learning rate, weight decay coefficient, and momentum are adjusted according to the performance on the validation set until the optimal hyperparameter combination is found.

6. The method for super-resolution reconstruction of power inspection images based on SRGAN as claimed in claim 5, characterized in that: The generator and discriminator are optimized by calculating the loss function, and then trained again until they are indistinguishable. The specific steps are as follows: The high-resolution image reconstructed by the generator and the original high-resolution image are sent to the VGG19Net network for feature extraction, and the root mean square error of the extracted high semantic features is calculated. The expression is as follows: in, refers to the content loss function, is the generator defined, θ G is the generator parameter, I HR is the reconstructed high-resolution image, Φ i,j represents the jth convolutional layer before the i-th maximum pooling of VGG19Net, W i,j and H i,j is the dimension of VGG19Net, i is the index variable of the convolutional layer in the VGG19Net network, and j is the feature map index variable in the convolutional layer; By calculating the adversarial loss, we can obtain the high-resolution image reconstructed by the discriminator generator. The probability of being true is expressed as follows: in, To combat losses, is the discriminator defined, θ D is the discriminator parameter, I LR is the input low-resolution image, N is the number of images; Combining content loss and adversarial loss into a perceptual loss function; The parameters of the discriminator are fixed and the parameters of the generator are adjusted to train the generator. The parameters of the generator are fixed and the values ​​of the parameters of the discriminator are adjusted. The discriminator is trained by maximizing the score of the original high-resolution image in the discriminator and minimizing the score of the reconstructed high-resolution image. Through adversarial training of the generator and the discriminator and alternating optimization, the generator finally reconstructs a high-resolution image that the discriminator cannot distinguish.

7. The method for super-resolution reconstruction of power inspection images based on SRGAN as claimed in claim 6, characterized in that: The content loss and adversarial loss are combined into a perceptual loss function. The specific steps are as follows: The SRGAN model combines the weighted sum of content loss and adversarial loss into perceptual loss, which is expressed as follows: in, is the content loss defined, l SR is the perceptual loss function.

8. The method for super-resolution reconstruction of power inspection images based on SRGAN as claimed in claim 7, characterized in that: The test set is input into the SRGAN model to test the trained SRGAN model. The specific steps are as follows: The test set is input into the trained SRGAN model to perform peak signal-to-noise ratio and structural similarity index tests; The high-resolution images reconstructed by the generator have good visual effects in terms of peak signal-to-noise ratio and structural similarity, and the generator and discriminator reach Nash equilibrium, indicating that the SRGAN model is successful.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the SRGAN-based power inspection image super-resolution reconstruction method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the SRGAN-based power inspection image super-resolution reconstruction method according to any one of claims 1 to 8 are implemented.

Images