Power image sample comprehensive enhancement method and system based on super-resolution and abnormal scene simulation

Abstract

The invention discloses a power image sample comprehensive enhancement method and system based on super-resolution and abnormal scene simulation, and the method comprises the steps: collecting power image samples, including normal scene and abnormal scene images, carrying out the quality evaluation of the power images through a deep neural network, and screening out a high-quality power image sample as a basic data set; performing high-quality reconstruction on the screened low-quality power image by using a depth super-resolution technology; an abnormal scene simulator is designed in combination with an actual abnormal scene in power production, and an abnormal scene image is expanded through a data enhancement technology and a deep learning generation model; a deep super-resolution technology is combined with abnormal scene expansion, a comprehensive enhancement strategy is generated, and finally a comprehensive power image data set is constructed. The method not only improves the definition and detail expressive force of the power image sample, but also greatly enriches the diversity of the sample library.

Description

Technical Field

[0001] The present invention relates to a method and system for comprehensive enhancement of electric power image samples, and in particular to a method and system for comprehensive enhancement of electric power image samples based on super-resolution and abnormal scene simulation. Background Art

[0002] In the daily operations and monitoring of the power industry, power image samples serve as the foundation for training large-scale visual models in power production scenarios. Their quality and diversity are crucial to the performance and accuracy of these models. However, existing power image sample libraries face two core challenges: first, image quality varies widely, with low resolution, insufficient clarity, and poor contrast being common issues; second, samples of abnormal scenes are scarce, making it difficult to meet the model's learning requirements for complex situations.

[0003] Traditional image enhancement and sample expansion methods, such as simple rotation, scaling, and flipping, can increase the number of samples to a certain extent, but the resulting images often lack authenticity and diversity, making it difficult to simulate the complex conditions of actual power production. With the rapid development of deep learning technology, advanced algorithms such as super-resolution techniques and generative adversarial networks (GANs) have provided new ideas for processing power image samples. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method and system for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation, which realizes comprehensive enhancement of power image samples through super-resolution technology based on deep learning algorithm and abnormal scene simulation technology based on generative adversarial network.

[0005] Technical solution: The present invention provides a comprehensive enhancement method for power image samples based on super-resolution and abnormal scene simulation, comprising:

[0006] (1) Collect power image samples, including normal scene and abnormal scene images, use deep neural network to evaluate the quality of power images, and select high-quality power image samples with a quality score of 0.8 to 1 as the basic data set;

[0007] (2) Using deep super-resolution technology, we reconstruct the low-quality power images with a quality score of 0 to 0.8 into high quality.

[0008] (3) Based on the actual abnormal scenarios in power production, an abnormal scenario simulator is designed, and abnormal scene image expansion is performed through data enhancement technology and deep learning generation model;

[0009] (4) Combine deep super-resolution technology with abnormal scene expansion to generate a comprehensive enhancement strategy and finally construct a comprehensive power image dataset.

[0010] Preferably, the deep neural network in step (1) adopts a convolutional neural network (CNN) as a model for image quality assessment, and the convolutional neural network (CNN) includes 1 input layer, 7 convolutional layers, 1 fully connected layer and 1 output layer.

[0011] Preferably, the convolutional neural network (CNN) performs quality assessment, including:

[0012] After preprocessing, the collected power image samples are divided into training set, validation set, test set and data set to be processed;

[0013] The image quality evaluation indicator Peak Signal-to-Noise Ratio (PSNR) was used to automatically evaluate the quality of the training set images. Experts were invited to review and adjust some of the training set images to obtain a quality score for each image.

[0014] Select Mean Squared Error (MSE) as the loss function to measure the difference between the predicted rating and the actual rating, and select Adam optimizer to update the parameters;

[0015] During the training process, the validation set is used to monitor the model performance. After the training is completed, the generalization ability of the model is evaluated on the test set.

[0016] The dataset to be processed is input into the trained convolutional neural network (CNN) to perform quantitative scoring on the image quality. Based on the scoring results, image samples with scores between 0.8 and 1 are selected as high-quality images as the basic dataset, and image samples with scores between 0 and 0.8 are selected as low-quality images.

[0017] Preferably, the deep super-resolution technology adopts a super-resolution reconstruction model SRCNN of a convolutional neural network, which includes three convolutional layers, respectively used for low-quality image feature extraction, nonlinear mapping, and reconstruction of high-resolution images.

[0018] Preferably, the super-resolution reconstruction includes:

[0019] The collected power image samples are divided into high-quality dataset, training set, validation set and low-quality dataset; the low-quality dataset is used as input and the high-quality image is used as the target output to train the super-resolution reconstruction SRCNN model, calculate the prediction results through forward propagation, and calculate the gradient and update the model parameters through back propagation; the trained super-resolution reconstruction SRCNN model is used to perform super-resolution reconstruction on the low-quality dataset.

[0020] Preferably, the data enhancement technology in step (3) includes but is not limited to rotating, scaling, flipping, and adding noise to the power image.

[0021] Preferably, the deep learning generation model is a generative adversarial network (GANs), which performs quality assessment on the generated image, including calculating the structural similarity index (SSIM) and the peak signal-to-noise ratio (PSNR), and comparing and verifying the generated layer image with the actual power image.

[0022] The present invention provides a comprehensive enhancement system for power image samples based on super-resolution and abnormal scene simulation, comprising:

[0023] Image quality assessment module: This module collects power image samples, including normal and abnormal scene images, uses a deep neural network to assess the quality of power images, and selects high-quality power image samples with a quality score of 0.8 to 1 as the basic dataset.

[0024] Super-resolution enhancement module: used to reconstruct high-quality low-quality power images with a quality score of 0 to 0.8 using deep super-resolution technology;

[0025] Abnormal scenario simulation module: used to design an abnormal scenario simulator based on actual abnormal scenarios in power production, and to expand abnormal scenario images through data enhancement technology and deep learning generation models;

[0026] Comprehensive enhancement strategy module: used to combine deep super-resolution technology with abnormal scene expansion to generate a comprehensive enhancement strategy and ultimately construct a comprehensive power image dataset.

[0027] A computer device includes one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors. When the programs are executed by the processors, the steps of the method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation are implemented.

[0028] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation.

[0029] Beneficial effects: Compared with the existing technology, the present invention has the following significant advantages: (1) It can automatically screen out high-quality image samples and use deep super-resolution technology to reconstruct low-resolution images with high resolution, significantly improving the image's detail expression and visual effects; (2) Combining data enhancement technology and deep learning generation model, a large number of diverse and high-quality abnormal scene images are generated. This not only solves the problem of insufficient abnormal scene samples, but also provides rich training data for the visual big model; (3) Through comprehensive enhancement strategy, personalized enhancement processing is provided for each original power image, and a comprehensive power image dataset containing both high-quality basic samples and rich abnormal scene samples is constructed, which significantly improves the training efficiency and accuracy of the model, making its application in power production scenarios more reliable and effective; (4) Deep learning and computer vision technology are applied to the enhancement and generation of power image samples, and a new comprehensive enhancement method is proposed, which fully considers the actual needs and characteristics of power production and has extremely high practicality; (5) It not only improves the quality and diversity of power image samples, but also provides a more solid data foundation for the application of visual big models in power production scenarios, and promotes the power industry to develop in a more intelligent and efficient direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The figure is a flow chart of the method of the present invention.

[0031] Figure 2 Flowchart for generating a synthetic data set for the method of the present invention.

[0032] Figure 3 This is a flowchart of the deep super-resolution reconstruction steps described in the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0034] like Figure 1 As shown, the present invention provides a comprehensive enhancement method for power image samples based on super-resolution and abnormal scene simulation, comprising:

[0035] (1) Collect power image samples, including normal scene and abnormal scene images, use deep neural network to evaluate the quality of power images, and select high-quality power image samples with a quality score of 0.8 to 1 as the basic data set;

[0036] The normal scenarios include image data of power equipment, power grid lines, and substation environments. Abnormal scenarios include power equipment failures (line short circuit, equipment overheating, equipment damage, etc.), personnel illegal operations (not wearing work clothes, not wearing safety helmets, smoking, staying in dangerous areas, etc.), and environmental abnormalities (oil leakage, damaged dials, damaged power equipment, equipment obstacles, equipment fireworks, etc.). During the specific implementation process, 120,000 image samples were collected.

[0037] Convolutional neural network (CNN) is selected as the core model for image quality assessment to perform feature extraction and quality assessment on power images. The network structure of convolutional neural network (CNN) includes 1 input layer, 7 convolutional layers, 1 fully connected layer, and 1 output layer.

[0038] Input layer: accepts preprocessed power images as input. The image size is unified to 224x224x3, with RGB three channels.

[0039] The pre-processing shown includes but is not limited to the following operations:

[0040] Resizing: The original power image is resized to a uniform size of 224x224 pixels to ensure that all input images have the same size for easier network processing;

[0041] Channel processing: If the original image is not RGB three-channel, convert it to RGB format. If it is a grayscale image, copy the grayscale value to the three channels to simulate an RGB image;

[0042] Normalization: Normalize the pixel values ​​of the image, for example, scaling the pixel values ​​from the range of 0-255 to the range of 0-1 or -1 to 1, which helps the network converge faster and improves training stability

[0043] Data augmentation: Increase data diversity by rotating, flipping, scaling, cropping, and other operations on the original image, thereby improving the generalization ability of the model;

[0044] In the specific implementation process, denoising, contrast enhancement, and histogram equalization operations are also included.

[0045] Detailed configuration of convolutional layer:

[0046] Convolutional layer 1: The number of convolution kernels is 32, the convolution kernel size is 3x3, the stride is 1, the padding is 'same', and the activation function is ReLU.

[0047] Convolutional layer 2: The number of convolution kernels is 64, the convolution kernel size is 3x3, the stride is 1, the padding is 'same', the activation function is ReLU, followed by a maximum pooling layer (Max Pooling Layer), the pooling window size is 2x2, and the stride is 2.

[0048] The third convolutional layer has 128 kernels, a kernel size of 3x3, a stride of 1, padding of 'same', and a ReLU activation function.

[0049] The fourth convolutional layer has 128 kernels, a kernel size of 3x3, a stride of 1, padding of 'same', and a ReLU activation function. It is followed by a maximum pooling layer with a pooling window size of 2x2 and a stride of 2.

[0050] The 5th convolutional layer has 256 kernels, a kernel size of 3x3, a stride of 1, padding of 'same', and a ReLU activation function.

[0051] The sixth convolutional layer has 256 kernels, a kernel size of 3x3, a stride of 1, padding of 'same', and a ReLU activation function.

[0052] The 7th convolutional layer has 256 kernels, a kernel size of 3x3, a stride of 1, padding of 'same', a ReLU activation function, and is followed by a maximum pooling layer with a pooling window size of 2x2 and a stride of 2.

[0053] Fully connected layer: After flattening the feature map output by the 7th convolutional layer into a one-dimensional vector, it is connected to a fully connected layer containing 512 neurons, and the activation function is ReLU.

[0054] Output layer: The fully connected layer is followed by a neuron, and the output is the image quality score, which ranges from 0 to 1, indicating the relative quality of the image.

[0055] The process of image quality assessment using convolutional neural network (CNN) includes:

[0056] (11) 40,000 power image samples were collected and preprocessed as training set A, 10,000 as validation set B, 10,000 as test set C, and 60,000 as the dataset to be processed D.

[0057] (12) In order to improve the accuracy and reliability of labels, the results of expert evaluation and automatic evaluation are combined, and the image quality evaluation indicator peak signal-to-noise ratio (PSNR) is used to automatically evaluate the quality of the training set images. Experts are then invited to review and adjust the images in the training set that have uncertainty or boundary conditions, and the quality score of each image is obtained as the target output during training.

[0058] (13) The mean square error (MSE) is used as the loss function to measure the difference between the predicted score and the actual score.

[0059]

[0060] Among them, MSE represents mean square error, n represents the number of samples, and y i represents the true value of the i-th sample, that is, the actual quality score of the image, Represents the predicted value of the i-th sample, that is, the image quality score output by the network.

[0061] (14) Select the Adam optimizer and set the appropriate learning rate to 0.001. The two momentum parameters of the Adam optimizer: the decay rate β1 of the first-order moment (gradient) is set to 0.9, and the decay rate β2 of the second-order moment (gradient square) is set to 0.999.

[0062] Based on the dataset size and model complexity, the number of training rounds is set to 80 to ensure that the model fully learns the image features.

[0063] During the training process, the validation set B is used to monitor the model performance to prevent overfitting; after training, the model generalization ability is evaluated on the test set C.

[0064] The 60,000 images in dataset D are input into the trained CNN network, which extracts key features from power images and quantitatively scores the image quality based on indicators such as clarity, contrast, and target object integrity.

[0065] Based on the quality score, high-quality image samples were selected as the basic dataset. The scoring threshold was set to 0.8, and scores between 0.8 and 1 indicate high-quality images. This resulted in a basic dataset A1 containing high-quality power images, providing reliable data support for subsequent super-resolution processing and abnormal scene simulation.

[0066] (2) Using deep super-resolution technology, the filtered low-resolution power images are reconstructed into high resolution;

[0067] The super-resolution reconstruction SRCNN model based on convolutional neural network is used to reconstruct low-resolution power images into high-resolution ones. Figure 3 The specific steps are as follows:

[0068] From the 120,000 collected images, 40,000 high-quality, high-resolution images were selected, 35,000 of which were used as the training set E, 5,000 as the validation set F, and 20,000 low-resolution images were selected as the dataset G. These images should cover a variety of power equipment, personnel, and environmental features to ensure the generalization ability of the model.

[0069] The images in the high-resolution datasets E and F are reduced to low-resolution images using bilinear interpolation and bicubic interpolation methods, and the images are reduced to 1 / 4 of the original size to obtain datasets E1 and F1 respectively.

[0070] The super-resolution reconstruction SRCNN model typically consists of three convolutional layers. The first layer has a convolution kernel size of 9x9 and is used for feature extraction to extract low-frequency features of the image; the second layer has a convolution kernel size of 1x1 and is used for nonlinear mapping; and the third layer has a convolution kernel size of 5x5 and is used to reconstruct high-resolution images.

[0071] The ReLU (Rectified Linear Unit) activation function is used after each convolutional layer to increase nonlinearity; the mean square error (MSE) is used to calculate the pixel difference between the reconstructed image and the real image to evaluate the model performance.

[0072] Training the super-resolution reconstruction SRCNN model includes:

[0073] The model weights and biases were initialized using random initialization, and the Adam optimization algorithm was used to update the model parameters. The SRCNN model was trained using the low-resolution dataset E1 as input and high-resolution images as target outputs. During training, predictions were calculated using forward propagation, and gradients were calculated and model parameters were updated using backpropagation.

[0074] The initial learning rate is set to 0.001 and can be gradually reduced as training progresses. The batch size is set to 32 based on the GPU memory capacity. The number of training epochs is set to 70 based on the dataset size and model complexity. During training, the validation set F1 score is regularly used to evaluate model performance to avoid overfitting. During training, model parameters are saved after each epoch for use in the testing phase. The best model is selected based on the validation set performance and saved.

[0075] The trained super-resolution reconstruction SRCNN model is used to perform super-resolution reconstruction on the low-resolution dataset G, restoring the high-frequency detail information in the image and making the originally blurred features of power equipment, personnel and environment clearly identifiable.

[0076] (3) Based on the actual abnormal scenarios in power production, an abnormal scenario simulator is designed, and abnormal scene image expansion is performed through data enhancement technology and deep learning generation model;

[0077] Combining data augmentation technology and deep learning generative models (GANs), we can generate diverse images of abnormal scenes. Specifically, we include the following:

[0078] Based on actual abnormal scenarios in power production, we define abnormal types and characteristics and design an abnormal scenario simulator. The simulator can simulate abnormal situations such as power equipment failures, illegal human operation, and environmental abnormalities.

[0079] Among them, power equipment failures include line short circuits, equipment overheating, equipment damage, etc., which may manifest as abnormal current and voltage, increased surface temperature of the equipment, or signs of damage on the appearance of the equipment.

[0080] Personnel violations include not wearing work clothes, not wearing a safety helmet, smoking, staying in dangerous areas, etc. Abnormal behavior of personnel may be captured through video surveillance.

[0081] Environmental anomalies include oil leakage, damaged dials, damaged electrical equipment, equipment obstructions, equipment fireworks, etc.

[0082] According to the actual abnormal scenarios in power production, corresponding mathematical models or physical models are established.

[0083] An abnormal scenario simulator was developed using the Python programming language and the MATLAB simulation tool to describe the occurrence and evolution of abnormal events and to simulate and visualize abnormal scenarios. Data augmentation techniques (such as rotation, scaling, flipping, and adding noise) and deep learning generative models (GANs) were used to generate diverse abnormal scene images.

[0084] First, we selected 5,000 abnormal images from the 120,000 collected images to form dataset H and annotated them, marking abnormal areas or objects. Dataset H was processed using data augmentation techniques, including rotation: rotating the image by a certain angle to simulate abnormal scenes from different perspectives. Scaling: enlarging or reducing the image by a certain ratio to simulate abnormal scenes observed at different distances. Flipping: flipping the image horizontally or vertically to increase image diversity. Adding noise: adding random noise (Gaussian noise, salt and pepper noise) to the image to simulate noise interference during image transmission or acquisition. The result is dataset H1 containing 50,000 images.

[0085] The generator is responsible for generating abnormal scene images. Its input is a random noise vector and its output is the generated image.

[0086] The discriminator is responsible for determining whether the input image is a real image or a generated image. Its input is a real image or a generated image, and the output is the judgment result, whether it is real or fake.

[0087] The cross entropy loss function is used to train the model; the Adam optimizer is selected to update the model parameters; the batch size is set to 64 images; the learning rate is 0.0002; and the number of iterations is determined based on the convergence of the model.

[0088] Use the trained GANs model to generate a variety of abnormal scene images. These images should include various types of abnormalities, such as power equipment failures, personnel violations, and environmental anomalies. By adjusting the generator's input noise vector and model parameters, abnormal scene images with different characteristics and details can be generated.

[0089] On the basis of maintaining the authenticity of the original image, abnormal scene images similar to real power images are generated by adjusting parameters and model structure.

[0090] Generator parameters: Adjust the number of convolutional layers, convolution kernel size, step size and other parameters of the generator to change the characteristics and details of the generated image.

[0091] Discriminator parameters: Adjust the discriminator's network structure, loss function weights and other parameters to improve its ability to distinguish between real images and generated images.

[0092] Training parameters: Adjust training parameters such as batch size, learning rate, and number of iterations to optimize the model training process.

[0093] Generate abnormal scene images using the adjusted GANs model.

[0094] The quality of the generated images is evaluated, such as calculating indicators such as the structural similarity index SSIM and the peak signal-to-noise ratio PSNR.

[0095] The generated images are compared with real power images to ensure their visual similarity.

[0096] (4) Combine deep super-resolution technology with abnormal scene expansion to generate a comprehensive enhancement strategy and finally construct a comprehensive power image dataset.

[0097] Combining super-resolution enhancement and abnormal scene expansion methods creates a comprehensive enhancement strategy. Each original power image undergoes personalized enhancement processing, including super-resolution reconstruction and abnormal scene simulation. Appropriate enhancement methods and parameters are selected based on the image content and actual needs. For example, super-resolution enhancement is prioritized for images containing critical power equipment, while abnormal scene simulation is performed for images requiring simulation of abnormal conditions.

[0098] Construct a comprehensive power image dataset that contains both high-quality basic samples and rich abnormal scene samples. The dataset should contain image samples from various power production scenarios to meet the training requirements of large visual models.

[0099] The present invention also provides a comprehensive enhancement system for power image samples based on super-resolution and abnormal scene simulation, comprising:

[0100] Image quality assessment module: This module collects power image samples, including normal and abnormal scene images, uses a deep neural network to assess the quality of power images, and selects high-quality power image samples with a quality score of 0.8 to 1 as the basic dataset.

[0101] Super-resolution enhancement module: used to reconstruct high-quality low-quality power images with a quality score of 0 to 0.8 using deep super-resolution technology;

[0102] Abnormal scenario simulation module: used to design an abnormal scenario simulator based on actual abnormal scenarios in power production, and to expand abnormal scenario images through data enhancement technology and deep learning generation models;

[0103] Comprehensive enhancement strategy module: used to combine deep super-resolution technology with abnormal scene expansion to generate a comprehensive enhancement strategy and ultimately construct a comprehensive power image dataset.

Claims

1. A comprehensive enhancement method for power image samples based on super-resolution and abnormal scene simulation, characterized in that: include: (1) Collect power image samples, including normal scene and abnormal scene images, use deep neural network to evaluate the quality of power images, and select high-quality power image samples with a quality score of 0.8 to 1 as the basic data set; (2) Using deep super-resolution technology, we reconstruct the low-quality power images with a quality score of 0 to 0.8 into high quality. (3) Based on the actual abnormal scenarios in power production, an abnormal scenario simulator is designed, and abnormal scene image expansion is performed through data enhancement technology and deep learning generation model; (4) Combine deep super-resolution technology with abnormal scene expansion to generate a comprehensive enhancement strategy and finally construct a comprehensive power image dataset.

2. The method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation according to claim 1, characterized in that: The deep neural network in step (1) adopts a convolutional neural network (CNN) as a model for image quality assessment, and the convolutional neural network (CNN) includes 1 input layer, 7 convolutional layers, 1 fully connected layer and 1 output layer.

3. The method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation according to claim 2, characterized in that: The convolutional neural network (CNN) performs quality assessment, including: After preprocessing, the collected power image samples are divided into training set, validation set, test set and data set to be processed; The image quality evaluation indicator Peak Signal-to-Noise Ratio (PSNR) was used to automatically evaluate the quality of the training set images. Experts were invited to review and adjust some of the training set images to obtain a quality score for each image. Select Mean Squared Error (MSE) as the loss function to measure the difference between the predicted rating and the actual rating, and select Adam optimizer to update the parameters; During the training process, the validation set is used to monitor the model performance. After the training is completed, the generalization ability of the model is evaluated on the test set. The dataset to be processed is input into the trained convolutional neural network (CNN) to perform quantitative scoring on the image quality. Based on the scoring results, image samples with scores between 0.8 and 1 are selected as high-quality images as the basic dataset, and image samples with scores between 0 and 0.8 are selected as low-quality images.

4. The method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation according to claim 1, characterized in that: The deep super-resolution technology adopts the super-resolution reconstruction model SRCNN of convolutional neural network, which includes three convolution layers for low-quality image feature extraction, nonlinear mapping and reconstruction of high-resolution images.

5. The method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation according to claim 4, characterized in that: The super-resolution reconstruction comprises: The collected power image samples are divided into high-quality dataset, training set, validation set and low-quality dataset; the low-quality dataset is used as input and the high-quality image is used as the target output to train the super-resolution reconstruction SRCNN model, calculate the prediction results through forward propagation, and calculate the gradient and update the model parameters through back propagation; the trained super-resolution reconstruction SRCNN model is used to perform super-resolution reconstruction on the low-quality dataset.

6. The method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation according to claim 1, characterized in that: The data enhancement technology in step (3) includes but is not limited to rotating, scaling, flipping, and adding noise to the power image.

7. The method for comprehensive enhancement of power image samples based on super-resolution and abnormal scene simulation according to claim 1, characterized in that: The deep learning generation model is a generative adversarial network (GANs), which performs quality assessment on the generated images, including calculating the structural similarity index (SSIM) and the peak signal-to-noise ratio (PSNR), and comparing and verifying the generated layer images with actual power images.

8. A comprehensive enhancement system for power image samples based on super-resolution and abnormal scene simulation, characterized by: include: Image quality assessment module: This module collects power image samples, including normal and abnormal scene images, uses a deep neural network to assess the quality of power images, and selects high-quality power image samples with a quality score of 0.8 to 1 as the basic dataset. Super-resolution enhancement module: used to reconstruct high-quality low-quality power images with a quality score of 0 to 0.8 using deep super-resolution technology; Abnormal scenario simulation module: used to design an abnormal scenario simulator based on actual abnormal scenarios in power production, and to expand abnormal scenario images through data enhancement technology and deep learning generation models; Comprehensive enhancement strategy module: used to combine deep super-resolution technology with abnormal scene expansion to generate a comprehensive enhancement strategy and ultimately construct a comprehensive power image dataset.

9. A computer device, characterized in that: The method comprises one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of a comprehensive enhancement method for power image samples based on super-resolution and abnormal scene simulation are implemented as described in any one of claims 1 to 7.

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 a comprehensive enhancement method for power image samples based on super-resolution and abnormal scene simulation are implemented as described in any one of claims 1 to 7.

Images