A transmission reconstruction method and system for substation main equipment inspection images

Abstract

The application discloses a transmission reconstruction method and system for substation main equipment inspection images, and the method comprises the following steps: collecting substation main equipment inspection images, performing main equipment category identification and classification, and obtaining corresponding main equipment category labels; performing adaptive coding according to the substation main equipment inspection images and the equipment category labels, and obtaining compressed inspection images; inputting the main equipment category labels into a corresponding super-resolution generative adversarial network submodel, and performing category-guided reconstruction processing on the compressed inspection images, and outputting high-resolution reconstructed images. The method can be deployed in a substation local area network or a safety zone IV, can work cooperatively with an inspection robot, a camera and an edge computing node, and can realize real-time collection, compression, transmission and reconstruction of inspection images.

Description

Technical Field

[0001] This invention relates to a method and system for transmitting and reconstructing inspection images of main equipment in substations, belonging to the fields of image processing and intelligent operation and maintenance technology of power systems. Background Technology

[0002] The safe operation of substations relies on real-time monitoring and inspection of main equipment (including circuit breakers, transformers, busbars, disconnectors, etc.). Substation inspections primarily involve manual inspections or the use of inspection equipment (drones, robots, cameras) to collect on-site images and videos. Due to the increasing number of devices and the growing demand for inspections, the amount of image data collected on-site has increased dramatically. In multi-threaded inspection scenarios, 60-80 images can be collected per second, with each image typically ranging from 1 to 4 MB in size. Existing substation bandwidth (usually 100 Mbps) cannot guarantee transmission speeds, leading to delays in fault detection. Existing image compression methods (such as JPEG, PNG, HEVC, etc.) can cause severe image distortion at high compression ratios, affecting subsequent analysis and judgment. This fails to meet the dual requirements of image quality and transmission efficiency for substation inspections, impacting the safe and stable operation of the substation.

[0003] Existing learning-based super-resolution reconstruction methods (such as SRCNN and DRCN) improve image quality by increasing the number of network layers, but in practical applications, they still suffer from problems such as training instability, vanishing gradients, and slow reconstruction speed, making it difficult to meet the requirements of high frame rates and real-time performance. Although Generative Adversarial Networks (GANs) have achieved some success in image detail restoration, existing methods still have the following shortcomings in substation inspection scenarios: on the one hand, they do not perform adaptive compression based on equipment type, resulting in the underutilization of image features of different equipment types; on the other hand, they lack multi-loss function optimization strategies during training, which cannot guarantee the stability and high fidelity of generated images under high compression rates.

[0004] Therefore, there is an urgent need for an image processing method that classifies and identifies various main equipment in substations, combines adaptive compression strategies, and uses generative adversarial networks for super-resolution reconstruction. This method aims to improve image transmission efficiency and reconstruction quality, and meet the comprehensive requirements of intelligent inspection systems for real-time performance, reliability, and image quality. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for transmitting and reconstructing inspection images of substation main equipment. In view of the problems of large data volume, severe compression distortion, low transmission efficiency and insufficient reconstruction accuracy of traditional inspection images in multi-type equipment scenarios, this invention proposes a full-link intelligent visual processing system that integrates "intelligent recognition-adaptive compression-type-guided reconstruction-high-fidelity restoration", realizing end-to-end intelligent optimization from image acquisition, compression encoding to reconstruction and restoration.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution.

[0007] In a first aspect, the present invention provides an optimized transmission method for inspection images of main equipment in substations, comprising:

[0008] Collect images of main equipment in the substation during inspection, identify and classify the main equipment categories, and obtain the corresponding main equipment category labels;

[0009] Based on the inspection images of the main equipment in the substation and the equipment category labels, adaptive encoding is performed to obtain compressed inspection images;

[0010] The main device category label is input into the corresponding super-resolution generative adversarial network sub-model, and the compressed inspection image is subjected to category-guided reconstruction processing to output a high-resolution reconstructed image.

[0011] Furthermore, the main device category identification adopts a lightweight convolutional neural network model. On the basic architecture including convolutional layers, batch normalization layers and activation layers, the Hard-Swish function replaces the original ReLU activation function, and an attention mechanism is embedded after each channel.

[0012] The main equipment category labels include, but are not limited to, circuit breakers, transformers, capacitors, insulators, and busbars.

[0013] Furthermore, based on the substation main equipment inspection images and the main equipment category labels, a depthwise separable convolutional coding network is used as a compression sub-model. Combined with quantization rate, convolutional kernel size and feature enhancement strategies, the inspection images are adaptively encoded to maximize the preservation of key feature information and reduce compression distortion.

[0014] Adaptive encoding is based on the main device category label and the image's texture complexity index. The optimal model selection is determined using a rule function, expressed as:

[0015]

[0016] in, For regular functions, , as well as This is an RGRU compressed network model;

[0017] The texture complexity index of the image is calculated from the edge density, color variance and local gradient change of the inspection image, and is used to characterize the richness of detail and compression sensitivity of the inspection image.

[0018] Based on the type of main equipment in the substation, the corresponding RGRU compression network model is selected to compress the inspection images to ensure the compression rate.

[0019] In the method of this invention, for main devices with complex structures and fine textures, a high-frequency enhancement model based on perceptual loss optimization is used to ensure texture details and edge continuity; for main devices at a macro scale, a reconstruction model based on structural consistency optimization is used to maintain the overall outline and shape stability.

[0020] Furthermore, the adaptively compressed inspection image is transmitted to the server. Based on the identified main device category, the server calls the corresponding Super-Resolution Generative Adversarial Network (SRGAN) sub-model to perform type-guided reconstruction processing on the compressed image, generating a high-resolution reconstructed image for high-fidelity restoration of different device features.

[0021] Furthermore, the SAN sub-model includes a generator structure and a discriminator structure;

[0022] The generator structure includes a feature extraction layer, a residual module, a feature fusion module, and an information reconstruction module connected in sequence.

[0023] The discriminator structure includes a convolutional layer, a Leaky-ReLU activation unit, a max pooling layer, and a batch normalization layer, used to determine whether the generated image closely approximates the distribution of the real image.

[0024] Furthermore, the generator structure introduces a residual network and a dense feature fusion module to enhance the transmission capability of multi-scale features through cross-layer feature connections and feature reuse mechanisms.

[0025] For different main device categories, the generator structure dynamically adjusts the feature fusion weights to enhance texture details and optimize structural consistency, thereby improving the detail performance and edge restoration accuracy of the generated image.

[0026] Furthermore, during the training phase of the super-resolution generative adversarial network sub-model, the loss function of the generator structure adopts a multi-objective optimization function, including perceptual loss, Wasserstein distance loss, and pixel mean square error loss.

[0027] The perceived loss High-level semantic features of generated and real images are extracted using a pre-trained network, and Euclidean distance is calculated to maintain consistency of high-level semantic features. The expression is as follows:

[0028] ;

[0029] in, This represents the total number of feature elements; Image data is input to the generator network; For the input image The generated image output after reconstruction by the generator network G; The real image corresponding to the input image; (·) represents the feature mapping operation for extracting high-level semantic features of images using a pre-trained network; To generate an image The i-th feature element in the feature map extracted by the pre-trained network; For real images The i-th feature element in the feature map extracted by the pre-trained network;

[0030] Wasserstein distance loss The expression used to measure the difference between the distribution of generated images and real images is:

[0031] ;

[0032] in, The probability distribution of the input data, Used to distinguish this specific distribution model; For the mapping function of the generator network, For the input image The generated image output after passing through the generator network; The mapping function for the discriminator network. For the discriminator network to generate images The output scalar value for evaluating authenticity;

[0033] The pixel mean square error loss The expression used to constrain overall brightness and color deviation is:

[0034] ;

[0035] in, To generate an image The feature value of the i-th pixel in the image; Let be the feature value of the i-th pixel in the real reference image y.

[0036] Furthermore, gamma correction and noise suppression processing are performed on the high-resolution reconstructed image to improve the reliability of image recognition and automatic analysis.

[0037] Secondly, the present invention provides a transmission and reconstruction system for inspection images of main equipment in substations, comprising:

[0038] The image acquisition module is used to acquire images of the main equipment in the substation during inspection, perform main equipment category identification and classification, and obtain the corresponding main equipment category label.

[0039] The image compression module is used to perform adaptive encoding based on the substation main equipment inspection image and the equipment category label to obtain the compressed inspection image;

[0040] The image reconstruction module is used to input the main device category label into the corresponding super-resolution generative adversarial network sub-model, and to perform category-guided reconstruction processing on the compressed inspection image to output a high-resolution reconstructed image.

[0041] Thirdly, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the method for transmitting and reconstructing inspection images of substation main equipment as described in any of the first aspects.

[0042] Fourthly, the present invention provides a computer device, comprising:

[0043] Memory, used to store computer programs / instructions;

[0044] A processor is configured to execute the computer program / instructions to implement the steps of the method for transmitting and reconstructing images of substation main equipment as described in any of the first aspects.

[0045] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0046] 1. The transmission and reconstruction method for inspection images of substation main equipment provided by the present invention firstly acquires inspection images of substation main equipment, performs main equipment category identification and classification, and obtains corresponding main equipment category labels; by accurately classifying and extracting features from the main equipment categories, the method of the present invention realizes real-time judgment of the category and complexity of the inspection images;

[0047] Secondly, based on the substation main equipment inspection image and the equipment category label, adaptive encoding is performed to obtain the compressed inspection image. The method of this invention uses a lightweight convolutional neural network model as the front end, and adaptively selects the optimal compression and reconstruction sub-network according to the main equipment category to form a differentiated visual optimization mechanism for multiple scenarios, which significantly improves the compression rate and reconstruction quality of the image.

[0048] Finally, the main device category label is input into the corresponding super-resolution generative adversarial network sub-model, and the compressed inspection image is subjected to category-guided reconstruction processing to output a high-resolution reconstructed image. The method of the present invention can improve the image transmission efficiency by about 185% under the same bandwidth conditions, and maintain the image structure similarity index of no less than 0.94 in high frame rate and multi-threaded inspection scenarios, thus ensuring image quality and usability.

[0049] 2. The method of the present invention is deployed in the local area network of the substation or in the safety zone IV, and works in collaboration with inspection robots, cameras and edge computing nodes to realize the acquisition, compression, transmission and reconstruction of real-time inspection images.

[0050] 3. The computer-readable storage medium and computer device provided by the present invention can execute the steps of the transmission and reconstruction method for substation equipment inspection images provided by the present invention. Attached Figure Description

[0051] Figure 1 This is an overall flowchart of an optimized transmission method for inspection images of substation main equipment provided according to an embodiment of the present invention.

[0052] Figure 2 A schematic diagram of an improved ShuffleNet network structure for an optimized transmission method of inspection images of substation main equipment provided by an embodiment of the present invention.

[0053] Figure 3 This is a schematic diagram of the encoder structure of a compression model for an optimized transmission method for inspection images of substation main equipment provided in an embodiment of the present invention.

[0054] Figure 4 This is a schematic diagram of the decoder structure of a compression model for an optimized transmission method for inspection images of substation main equipment provided in an embodiment of the present invention.

[0055] Figure 5 This is a flowchart illustrating the generator structure of an optimized transmission method for inspection images of substation main equipment provided in an embodiment of the present invention.

[0056] Figure 6 This is a flowchart illustrating the discriminator structure of an optimized transmission method for inspection images of substation main equipment provided in an embodiment of the present invention. Detailed Implementation

[0057] It should be noted that:

[0058] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0059] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0060] Example 1

[0061] like Figure 1 As shown in the figure, this embodiment introduces an optimized transmission method for inspection images of main equipment in substations, including:

[0062] Collect images of main equipment in the substation during inspection, identify and classify the main equipment categories, and obtain the corresponding main equipment category labels;

[0063] Based on the inspection images of the main equipment in the substation and the equipment category labels, adaptive encoding is performed to obtain compressed inspection images;

[0064] The main device category label is input into the corresponding super-resolution generative adversarial network sub-model, and the compressed inspection image is subjected to category-guided reconstruction processing to output a high-resolution reconstructed image.

[0065] Furthermore, in this embodiment, on-site inspection images are acquired using a substation inspection robot or a fixed camera. The image resolution is 1920×1080, and the size of a single image is between 1 and 4 MB. A multi-threaded acquisition method is adopted, which can capture 60 to 80 inspection images within 1 second, effectively ensuring the inspection coverage and real-time data acquisition. Among them, the sample inspection images used for training and testing cover various main devices, totaling 10,000 images, and are divided into training set and test set in an 8:2 ratio, that is, the number of training set samples is 8,000 images and the number of validation set samples is 2,000 images.

[0066] Furthermore, a large number of main equipment inspection images collected on-site are used as the training set. In this embodiment, the number of training set samples is set to 8,000, and the main equipment categories are labeled, including circuit breakers, transformers, capacitors, insulators, and busbars.

[0067] Furthermore, considering that the present invention can be used not only for station servers but also for mobile inspection equipment such as drones and robots, the method of the present invention is based on the lightweight convolutional neural network ShuffleNet. ShuffleNet mainly solves the problems of large computational cost per layer and information isolation caused by group convolution by using group convolution and channel shuffle.

[0068] like Figure 2 As shown, the improved ShuffleNet classification model, based on convolutional layers, batch normalization layers, and activation layers, addresses the limited number of substation main equipment inspection image samples by replacing the original ReLU activation function with the Hard-Swish function. This avoids the excessive feature sparsity and information loss caused by ReLU in low-precision and small networks. The Hard-Swish function expression is as follows:

[0069] ;

[0070] in, The activation function is applied to the current input feature elements. The processed output feature activation values; Let i be the i-th feature element input to the activation function.

[0071] Furthermore, to address the issue of inconsistent equipment sizes in substations—with the volume of main transformers and insulators differing by hundreds of times—leading to a low recognition rate for small targets in the original Shuffle Unit structure, this invention embeds a Squeeze-and-Excitation (SE) attention mechanism module after each Shuffle Unit structure. This module adaptively enhances the response of important features through a feature recalibration mechanism at the channel dimension, thereby improving the stability of small target recognition and ultimately achieving rapid and accurate classification of equipment such as circuit breakers, transformers, busbars, and disconnectors.

[0072] The proposed method of this invention proposes an improved model based on Shuffle Unit, which has been validated on a test set. The ShuffleNet classification model achieves a recognition rate of 72.8% for active equipment inspection images.

[0073] Furthermore, based on the classified inspection images, a depthwise separable convolutional coding network is used as a compression sub-model. Combined with quantization rate, convolutional kernel size and feature enhancement strategies, the inspection images are adaptively encoded to maximize the preservation of key feature information and reduce compression distortion.

[0074] In this embodiment, the system outputs main device category labels, including transformers, circuit breakers, insulators, and capacitors. Since different devices exhibit significant differences in image texture details, edge complexity, and space occupancy, using a uniform compression model would result in compression distortion or resource waste. Therefore, this invention designs a lightweight main device category adaptive selection module (ACS) that integrates the main device category with the image's texture complexity index. The ACS module selects the most suitable compression model structure from the preset RGRU model library. It comprises three core parts:

[0075] 1. Complexity Evaluation Unit:

[0076] Calculate the texture complexity index of the input inspection image. In this embodiment, the index is obtained by combining edge density, color variance and local gradient change, and is used to characterize the image detail richness and compression sensitivity.

[0077] 2. Model Library Management Unit:

[0078] In this embodiment, several RGRU model variants were designed to adapt to different compression ratios and feature fidelity requirements, including RGRU-L (Large), RGRU-M (Medium), and RGRU-S (Small):

[0079] The RGRU-L (Large) is suitable for master devices with complex features and rich structural details (such as main transformers), with an emphasis on preserving texture and detail.

[0080] The RGRU-M (Medium) is suitable for medium-complex master equipment (such as circuit breakers and capacitors), balancing compression ratio and clarity.

[0081] The RGRU-S (Small) is suitable for main equipment with a small main body or a high background ratio (such as insulators and busbars), and pursues a high compression ratio.

[0082] 3. Adaptive Decision Unit:

[0083] Based on the obtained category labels and texture complexity index This embodiment uses a rule function. The optimal model selection result is calculated using the following expression:

[0084] ;

[0085] Based on the identified main device category, the corresponding RGRU compression network is selected to compress the inspection images, ensuring the compression rate.

[0086] Furthermore, the compression model structure consists of two parts: an encoder structure and a decoder structure, such as... Figure 3 As shown, the encoder structure includes a five-layer convolutional network and GRU units, and extracts multi-scale features by progressively reducing the resolution of the feature maps, ultimately outputting a highly efficient compressed representation (Binarizer) after binarization; as shown Figure 4 As shown, the decoder structure includes a four-layer convolutional network and a GRU unit. It reconstructs image features by progressively upsampling the compressed representation and generates a compressed image suitable for super-resolution reconstruction. The size of the compressed inspection image is controlled between 50 and 100KB.

[0087] This method reduces the size of the inspection image to less than 20% of the original by fusing depthwise separable convolution with a GRU-designed RGRU compression network, while maintaining the integrity of the core structural information. Compared with the original image, it significantly reduces the transmission load and effectively preserves key textures and device details.

[0088] Furthermore, the adaptively compressed inspection images are transmitted to the server. Multi-threaded transmission ensures real-time image upload and avoids data backlog. Based on the identified main device category, the server calls the corresponding Super-Resolution Generative Adversarial Network (SRGAN) sub-model to perform category-guided reconstruction processing on the compressed image, generating a high-resolution reconstructed image. In this embodiment, preliminary anomaly screening can also be achieved by combining edge computing main devices, further improving the processing efficiency of the main station.

[0089] To address the issues of detail loss and feature blurring caused by compressed images, this embodiment employs an SRGAN sub-model at the decoding end, combining the recognition results to perform "category-driven reconstruction" to achieve high-fidelity restoration of compressed images. The training set consists of compressed images and their corresponding high-resolution real images, ensuring that the model learns a non-linear mapping from low-quality to high-quality images.

[0090] Furthermore, the SRGAN sub-model includes a generator structure and a discriminator structure;

[0091] like Figure 5As shown, the generator structure includes a feature extraction layer, a residual module, a feature fusion module, and an information reconstruction module connected in sequence. The feature extraction layer extracts low-resolution image features; the residual module enhances gradient propagation and feature reuse, improving detail recovery capabilities; the feature fusion module connects the outputs of each residual block to form dense features; and the information reconstruction module outputs a super-resolution reconstructed image. In this embodiment, the generator structure dynamically adjusts the residual block depth and the number of feature channels for different main device categories. For devices with complex textures and rich details, such as transformers, the generator uses a deeper residual network structure with a residual block count ranging from 16 to 32. In this embodiment, 24 residual blocks are used, and the number of feature channels in each residual block is set to 64 to 128 to enhance the expression of high-frequency texture information and detail reconstruction capabilities.

[0092] For devices with relatively simple structures and few texture variations, such as insulators, the generator adopts a lightweight shallow network structure with a residual block count ranging from 4 to 8. In this embodiment, it is set to 6 residual blocks and the number of feature channels is set to 32 to 64. This effectively reduces computational complexity and improves reconstruction speed while ensuring basic reconstruction quality.

[0093] The generator incorporates a residual network and a dense feature fusion mechanism, which enhances the transfer capability of multi-scale features through cross-layer connections and feature reuse. At the same time, the feature fusion weights are adaptively adjusted according to the host device category, which effectively improves the edge restoration accuracy and texture performance of the generated image while optimizing structural consistency.

[0094] like Figure 6 As shown, the discriminator structure includes a convolutional layer, a Leaky-ReLU activation unit, a max pooling layer, and a batch normalization layer, which are used to probabilistically distinguish between the generated image and the real image, prompting the generator to continuously optimize to approximate the distribution of the real high-resolution image.

[0095] Furthermore, during the training phase of the super-resolution generative adversarial network sub-model, the loss function of the generator structure adopts a multi-objective optimization function, including perceptual loss, Wasserstein distance, and pixel mean square error loss, as shown in Table 1:

[0096] Table 1 Loss Function Types

[0097]

[0098] In this embodiment, the perceptual loss is based on the VGG19 pre-trained network, which extracts high-level semantic features between the generated and real images and calculates the Euclidean distance to maintain the consistency of high-level semantic features. The definition is as follows:

[0099] ;

[0100] in, The total number of feature elements in the feature map extracted from the VGG19 network; Image data is input to the generator network; For the input image The generated image output after reconstruction by the generator network G; The real image corresponding to the input image; (·) represents the feature mapping operation for extracting high-level semantic features of images using the VGG19 pre-trained network; To generate an image The i-th feature element in the feature map extracted by the VGG19 network; For real images The i-th feature element in the feature map extracted by the VGG19 network.

[0101] Wasserstein distance loss is used to measure the difference between the generated image and the real image distribution, ensuring the stability of generation training. The definition is as follows:

[0102] ;

[0103] in, The probability distribution of the input data, Used to distinguish this specific distribution model; For the mapping function of the generator network, For the input image The generated image output after passing through the generator network; The mapping function for the discriminator network. For the discriminator network to generate images The output is a scalar value evaluating the authenticity of the data.

[0104] Pixel mean square error loss is used to constrain overall brightness and color deviation, maintaining the overall brightness and color consistency of the generated image. The definition is as follows:

[0105] ;

[0106] in, To generate an image The feature value of the i-th pixel in the image; Let be the feature value of the i-th pixel in the real reference image y;

[0107] The overall loss function is defined as:

[0108] .

[0109] Furthermore, gamma correction and noise suppression processing are performed on the high-resolution reconstructed image to improve the reliability of image recognition and automatic analysis.

[0110] Furthermore, in this embodiment, the optimization algorithm used for model training is Adam, and the initial learning rate is set to... The training iterations were set to 2000, the learning rate was gradually reduced with the number of iterations, and the magnification was set to 4x to ensure maximum restoration of compressed image details. After the model was trained, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) metrics were used to evaluate the quality of the reconstructed images on the validation set.

[0111] To verify the effectiveness of the method of this invention, the PSNR and SSIM of the generated reconstructed images were evaluated and compared with traditional methods such as Bicubic, SRCNN, and DRCN. Experimental results show that the method of this invention improves the PSNR by an average of about 2.5dB and the SSIM by about 0.05~0.08. Moreover, the reconstructed high-resolution inspection images have clear details, natural texture restoration, and better visual effects than existing technologies. At the same time, while maintaining an SSIM of over 0.94, the method of this invention improves the compression transmission efficiency by about 185%, achieving low-latency, high-fidelity intelligent image transmission and reconstruction.

[0112] The high-resolution reconstructed images generated by the method of this invention can be directly applied to power grid inspection systems, supporting rapid analysis and remote monitoring. They provide highly robust and real-time visual data support for multi-source inspection systems such as UAVs and robots in substations, and have high engineering practical value.

[0113] Example 2

[0114] Based on the optimized transmission method for inspection images of substation main equipment described in Embodiment 1, this embodiment introduces an optimized transmission system for inspection images of substation main equipment, including:

[0115] The image acquisition module is used to acquire images of the main equipment in the substation during inspection, perform main equipment category identification and classification, and obtain the corresponding main equipment category label.

[0116] The image compression module is used to perform adaptive encoding based on the substation main equipment inspection image and the equipment category label to obtain the compressed inspection image;

[0117] The image reconstruction module is used to input the main device category label into the corresponding super-resolution generative adversarial network sub-model, and to perform category-guided reconstruction processing on the compressed inspection image to output a high-resolution reconstructed image.

[0118] Example 3

[0119] Based on the optimized transmission method for inspection images of substation main equipment described in Embodiment 1, this embodiment introduces a computer-readable storage medium storing a computer program / instruction. When the computer program / instruction is executed by a processor, it implements the steps of the optimized transmission method for inspection images of substation main equipment as described in any of Embodiment 1.

[0120] Example 4

[0121] Based on the optimized transmission method for inspection images of substation main equipment described in Embodiment 1, this embodiment provides a computer device, including:

[0122] Memory, used to store computer programs / instructions;

[0123] A processor is configured to execute the computer program / instructions to implement the steps of the optimized transmission method for inspection images of substation main equipment as described in any one of Embodiments 1.

[0124] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0126] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0128] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for transmitting and reconstructing inspection images of main equipment in substations, characterized in that, include: Collect images of main equipment in the substation during inspection, identify and classify the main equipment categories, and obtain the corresponding main equipment category labels; Based on the inspection images of the main equipment in the substation and the equipment category labels, adaptive encoding is performed to obtain compressed inspection images; The main device category label is input into the corresponding super-resolution generative adversarial network sub-model, and the compressed inspection image is subjected to category-guided reconstruction processing to output a high-resolution reconstructed image.

2. The method for transmitting and reconstructing inspection images of substation main equipment according to claim 1, characterized in that, The main device category identification adopts a lightweight convolutional neural network model. On the basic architecture including convolutional layers, batch normalization layers and activation layers, the Hard-Swish function replaces the original ReLU activation function, and an attention mechanism is embedded after each channel. The main equipment category labels include, but are not limited to, circuit breakers, transformers, capacitors, insulators, and busbars.

3. The method for transmitting and reconstructing inspection images of substation main equipment according to claim 1, characterized in that, The adaptive encoding is based on the host device category label and the image's texture complexity index. The optimal model selection is determined using a rule function, expressed as: ； in, For regular functions, , as well as This is an RGRU compressed network model; The texture complexity index of the image is calculated from the edge density, color variance and local gradient change of the inspection image, and is used to characterize the richness of detail and compression sensitivity of the inspection image. Based on the identified main device category, the corresponding RGRU compression network model is selected to compress the inspection image to ensure the compression rate.

4. The method for transmitting and reconstructing inspection images of substation main equipment according to claim 1, characterized in that, The super-resolution generative adversarial network sub-model includes a generator structure and a discriminator structure; The generator structure includes a feature extraction layer, a residual module, a feature fusion module, and an information reconstruction module connected in sequence. The discriminator structure includes a convolutional layer, a Leaky-ReLU activation unit, a max pooling layer, and a batch normalization layer, used to determine whether the generated image closely approximates the distribution of the real image.

5. The method for transmitting and reconstructing inspection images of substation main equipment according to claim 4, characterized in that, The generator structure introduces a residual network and a dense feature fusion module to enhance the transmission capability of multi-scale features through cross-layer feature connections and feature reuse mechanisms. For different main device categories, the generator structure dynamically adjusts the feature fusion weights to achieve texture detail enhancement and structural consistency optimization.

6. The method for transmitting and reconstructing inspection images of substation main equipment according to claim 4, characterized in that, During the training phase of the super-resolution generative adversarial network sub-model, the loss function of the generator structure adopts a multi-objective optimization function, including perceptual loss, Wasserstein distance loss, and pixel mean square error loss. The perceived loss High-level semantic features of generated and real images are extracted using a pre-trained network, and Euclidean distance is calculated to maintain consistency of high-level semantic features. The expression is as follows: ； in, This represents the total number of feature elements; Image data is input to the generator network; For the input image The generated image output after reconstruction by the generator network G; The real image corresponding to the input image; (·) represents the feature mapping operation for extracting high-level semantic features of images using a pre-trained network; To generate an image The i-th feature element in the feature map extracted by the pre-trained network; For real images The i-th feature element in the feature map extracted by the pre-trained network; Wasserstein distance loss The expression used to measure the difference between the distribution of generated images and real images is: ； in, The probability distribution of the input data, Used to distinguish this specific distribution model; For the mapping function of the generator network, For the input image The generated image output after passing through the generator network; The mapping function for the discriminator network. For the discriminator network to generate images The output scalar value for evaluating authenticity; The pixel mean square error loss The expression used to constrain overall brightness and color deviation is: ； in, To generate an image The feature value of the i-th pixel in the image; Let be the feature value of the i-th pixel in the real reference image y.

7. The method for transmitting and reconstructing inspection images of substation main equipment according to claim 1, characterized in that, The high-resolution reconstructed image is subjected to gamma correction and noise suppression processing to improve the reliability of image recognition and automatic analysis.

8. A transmission and reconstruction system for inspection images of main equipment in substations, characterized in that, include: The image acquisition module is used to acquire images of the main equipment in the substation during inspection, perform main equipment category identification and classification, and obtain the corresponding main equipment category label. The image compression module is used to perform adaptive encoding based on the substation main equipment inspection image and the equipment category label to obtain the compressed inspection image; The image reconstruction module is used to input the main device category label into the corresponding super-resolution generative adversarial network sub-model, and to perform category-guided reconstruction processing on the compressed inspection image to output a high-resolution reconstructed image.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the method for transmitting and reconstructing images of substation main equipment as described in any one of claims 1 to 7.

10. A computer device, characterized in that, include: Memory, used to store computer programs / instructions; A processor is configured to execute the computer program / instructions to implement the steps of the method for transmitting and reconstructing images of substation main equipment as described in any one of claims 1 to 7.

Images