Rust identification method and device for substation equipment, terminal equipment and storage medium

Abstract

The invention discloses a substation equipment corrosion identification method and device, terminal equipment and a storage medium, and belongs to the technical field of equipment corrosion identification, and the method comprises the steps: obtaining a target inspection image of a substation and text data used for describing each piece of to-be-identified equipment of the substation, according to the method, component features used for representing visual features of corrosion and text features used for representing semantic description of corrosion are extracted through a corrosion recognition model, finally, all the component features and all the text features are fused, and a corrosion result of a target inspection image can be output according to the fused features. Through the automatic image processing and text feature extraction technology, the labor intensity of manual inspection can be reduced, the misjudgment risk caused by human factors is reduced, and rapid and accurate identification of equipment corrosion is realized. According to the invention, the problems of low corrosion identification efficiency and inaccurate identification in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment corrosion identification, and in particular to a method, device, terminal equipment and storage medium for identifying corrosion of substation equipment. Background Art

[0002] Corrosion of substation equipment is a common problem in power systems. It can lead to decreased equipment performance, increased safety hazards, and even equipment failure, impacting the stable operation of the power system. Identifying corrosion in substation equipment is crucial because timely detection and treatment of corrosion issues can prevent equipment failures, reduce maintenance costs, extend equipment life, and ensure stable power system operation.

[0003] However, traditional manual inspection methods have many shortcomings. Traditional technologies mainly rely on the experience and intuitive judgment of inspectors. For example, it is necessary to manually identify inspection images and check the descriptive language of the equipment to determine whether there is rust. This is not only time-consuming and labor-intensive, but also easily affected by subjective factors. Due to the need for manual inspection and identification of rust, the existing technology has the problem of low rust recognition efficiency and inaccurate recognition. Summary of the Invention

[0004] The embodiments of the present invention provide a method, device, terminal device and storage medium for identifying corrosion of substation equipment. Through automated image processing and text feature extraction technology, the labor intensity of manual inspections can be reduced, the risk of misjudgment and missed judgment caused by human factors can be reduced, and rapid and accurate identification of corrosion of substation equipment can be achieved. This can effectively solve the problems of low efficiency and inaccurate identification of corrosion in the existing technology.

[0005] An embodiment of the present invention provides a method for identifying corrosion of substation equipment, comprising:

[0006] Obtaining a target inspection image corresponding to the substation and text data describing each device to be identified in the substation; wherein the target inspection image includes: a plurality of devices to be identified;

[0007] Inputting the target inspection image and the text data into a preset corrosion recognition model, so that the corrosion recognition model segments each device to be identified in the target inspection image into a plurality of component region maps, and extracts component features corresponding to each component region map for characterizing the visual characteristics of corrosion; extracting text features corresponding to each component region map for characterizing the semantic description of corrosion based on the text data; fusing each component feature with each text feature to generate a fused feature, and outputting the corrosion result corresponding to the target inspection image based on the fused feature;

[0008] The corrosion result includes: a result of a device used to characterize the presence of corrosion in the target inspection image, or a result of a device used to characterize the absence of corrosion in the target inspection image.

[0009] Preferably, the component features include: texture features, color features, shape features, material features and edge features;

[0010] The step of extracting component features corresponding to each component region map and used to characterize visual features of corrosion includes:

[0011] For each component region map, texture features used to describe texture complexity are extracted based on the surface texture changes of the component;

[0012] The color gamut in the component area map is statistically analyzed to extract color features for quantifying the distribution of the rust area.

[0013] Extracting shape features for characterizing the geometric shape of the rusted area from the component area map according to the contour shape of the component in the component area map;

[0014] Capture the long-range dependencies between pixels in the component area map and extract the material features used to characterize the surface material of the component based on the long-range dependencies;

[0015] The edge information of the component area map is captured according to the gradient calculation method of the convolution kernel, and edge features for characterizing the edge contour of the component are generated.

[0016] Preferably, extracting texture features for describing texture complexity based on surface texture changes of the component includes:

[0017] The shallow convolution kernel in the rust recognition model is used to capture the fine-grained texture of rust particles in the component area map to characterize the component surface.

[0018] The deep convolution kernel in the rust recognition model is used to capture the coarse-grained texture in the component area map that is used to characterize the patchy distribution of the component surface.

[0019] The fine-grained texture and the coarse-grained texture are multi-scale fused to generate texture features for describing texture complexity.

[0020] Preferably, the text data includes: the installation environment of each component, the installation position of each component, the rust degree of each component, the component color of each component, the component material of each component, and the component structure of each component;

[0021] The step of extracting text features corresponding to each component region diagram and used for representing the semantic description of corrosion according to the text data includes:

[0022] For each component area map, a corrosion risk feature is generated to characterize and describe the corrosion risk level of each component based on the installation environment, installation location, and corrosion level of each component.

[0023] Generate attribute features for characterizing and describing visual attributes of the components according to the component color of each component, the component material of each component, and the component structure of each component;

[0024] A text feature for representing a semantic description of rust is generated based on the corrosion risk feature and the attribute feature.

[0025] Preferably, the fusing of each component feature with each text feature to generate a fused feature includes:

[0026] Through the projection layer in the rust recognition model, each component feature and each text feature are mapped into the same feature space to generate several target component features and several target text features;

[0027] Based on the cross-modal attention mechanism, the features of each target component are fused with the features of each target text to generate fused features.

[0028] Preferably, the training process of the preset corrosion recognition model includes:

[0029] The inspection image samples and the text data samples corresponding to the inspection images are used as training samples; wherein the inspection image samples include: a plurality of equipment samples to be identified;

[0030] Taking each training sample and the actual corrosion result corresponding to each training sample as input and the predicted corrosion result of each training sample as output, the corrosion recognition model to be trained is iteratively trained until the corrosion recognition model converges, thereby generating a completed corrosion recognition model.

[0031] Preferably, it also includes:

[0032] During each iteration of training the corrosion recognition model, a training sample is input into the corrosion recognition model, so that the corrosion recognition model generates a corrosion prediction result corresponding to the training sample based on each component feature sample and each text feature sample in the training sample;

[0033] The corrosion prediction results are compared with the actual corrosion results, and the network parameters of the corrosion recognition model are adjusted according to the comparison results.

[0034] Based on the above method embodiments, the present invention provides corresponding device embodiments.

[0035] An embodiment of the present invention provides a corrosion identification device for substation equipment, comprising: a data acquisition module and a corrosion result generation module;

[0036] The data acquisition module is used to acquire a target inspection image corresponding to the substation and text data used to describe each device to be identified in the substation; wherein the target inspection image includes: a plurality of devices to be identified;

[0037] The corrosion result generation module is used to input the target inspection image and the text data into a preset corrosion recognition model, so that the corrosion recognition model divides each device to be identified in the target inspection image into a number of component area maps, and extracts the component features corresponding to each component area map for characterizing the visual features of corrosion; extracts the text features corresponding to each component area map for characterizing the semantic description of corrosion based on the text data; fuses each component feature with each text feature to generate a fused feature, and outputs the corrosion result corresponding to the target inspection image based on the fused feature; wherein the corrosion result includes: a result for characterizing the device with corrosion in the target inspection image, or a result for characterizing the device without corrosion in the target inspection image.

[0038] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0039] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the corrosion identification method for substation equipment described in the above-mentioned embodiment of the invention.

[0040] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0041] Another embodiment of the present invention provides a storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the corrosion identification method for substation equipment described in the above-mentioned embodiment of the invention.

[0042] The following beneficial effects are achieved by implementing the present invention:

[0043] Embodiments of the present invention provide a method, apparatus, terminal device, and storage medium for identifying corrosion of substation equipment. The present invention first obtains a target inspection image corresponding to the substation and text data describing each device to be identified in the substation. The target inspection image and text data are then input into a preset corrosion recognition model. The corrosion recognition model segments each device to be identified in the target inspection image into several component region maps, extracts component features corresponding to each component region map for visually characterizing corrosion, and extracts text features corresponding to each component region map for semantically characterizing corrosion based on the text data. Finally, the component features are fused with the text features to generate fused features. Based on the fused features, the corrosion result corresponding to the target inspection image is output. Compared with the prior art, the present invention can reduce the labor intensity of manual inspections and the risk of misjudgment and missed detection caused by human factors through automated image processing and text feature extraction technology. This allows for rapid and accurate identification of corrosion in substation equipment, providing reliable data support for equipment maintenance and improving the operational safety and stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The present invention is a flowchart of a method for identifying corrosion of substation equipment provided by an embodiment of the present invention.

[0045] Figure 2 This is a schematic diagram of a cross-modal information fusion process provided by an embodiment of the present invention.

[0046] Figure 3 The present invention is a schematic structural diagram of a corrosion identification device for substation equipment provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0048] like Figure 1 As shown, in order to solve the problem of low efficiency and inaccurate corrosion recognition in the prior art, an embodiment of the present invention provides a corrosion recognition method for substation equipment, comprising:

[0049] Step S1: Acquire a target inspection image corresponding to a substation and text data describing each device to be identified in the substation; wherein the target inspection image includes: a plurality of devices to be identified;

[0050] Step S2: Inputting the target inspection image and the text data into a preset corrosion recognition model, so that the corrosion recognition model segments each device to be identified in the target inspection image into a plurality of component region maps, and extracts component features corresponding to each component region map for characterizing the visual characteristics of corrosion; extracting text features corresponding to each component region map for characterizing the semantic description of corrosion based on the text data; fusing each component feature with each text feature to generate a fused feature, and outputting the corrosion result corresponding to the target inspection image based on the fused feature;

[0051] The corrosion result includes: a result of a device used to characterize the presence of corrosion in the target inspection image, or a result of a device used to characterize the absence of corrosion in the target inspection image.

[0052] Regarding step S1, in a preferred embodiment, the target inspection image corresponding to the substation can be an image captured on-site at the substation by a camera or other image acquisition device, which can capture and record the appearance status of each device in the substation. Schematically, the inspection image can include the following contents:

[0053] Overall device image: displays the overall appearance of the device, including basic information such as shape, size, and color.

[0054] Equipment detail images: Focus on specific parts or details of the equipment, such as connectors, fasteners, insulators, etc. These parts are often prone to corrosion problems.

[0055] Multi-angle images: Images of equipment taken from different angles provide a more comprehensive perspective, helping to identify corrosion issues that are difficult to see from a single angle.

[0056] Furthermore, the text data input by the present invention is a supplementary description of the inspection image, which can provide additional information about the substation equipment. Schematically, the text data may include: equipment name and number, which are used to uniquely identify each device. Equipment type, such as transformer, circuit breaker, disconnector, etc. Equipment location, the specific location information in the substation, helps to locate the location where the corrosion problem occurs. Rust description, that is, a text description of the equipment corrosion condition, such as the degree of rust, rust location, rust type, etc., the above text information can be combined with image features to improve the accuracy of subsequent corrosion identification.

[0057] For example, traditional corrosion detection methods often rely on visual features, which can make it difficult to distinguish actual corrosion from other similar visual features (such as dust accumulation and stains) on the target component. However, the present invention can more accurately distinguish features similar to corrosion by analyzing and fusing the feature dimensions of image and text data, thereby improving the accuracy of corrosion detection.

[0058] Regarding step S2, in a preferred embodiment, the training process of the preset corrosion recognition model includes:

[0059] The inspection image samples and the text data samples corresponding to the inspection images are used as training samples; wherein the inspection image samples include: a plurality of equipment samples to be identified;

[0060] Taking each training sample and the actual corrosion result corresponding to each training sample as input and the predicted corrosion result of each training sample as output, the corrosion recognition model to be trained is iteratively trained until the corrosion recognition model converges, thereby generating a completed corrosion recognition model.

[0061] During each iteration of training the corrosion recognition model, a training sample is input into the corrosion recognition model, so that the corrosion recognition model generates a corrosion prediction result corresponding to the training sample based on each component feature sample and each text feature sample in the training sample;

[0062] The corrosion prediction results are compared with the actual corrosion results, and the network parameters of the corrosion recognition model are adjusted according to the comparison results.

[0063] Illustratively, the present invention uses a variety of training samples during the training process, including images of different devices, different angles, and different lighting conditions, as well as rich text data, so that the model has strong generalization capabilities, so that it can handle rust identification tasks in various complex scenarios and adapt to the inspection needs of different substations.

[0064] In a preferred embodiment, the present invention can also introduce data enhancement technology (such as random rotation, flipping, scaling, etc.) and regularization methods to enable the model to learn more robust feature representations during the training process. When applied, it can still maintain a high recognition accuracy even in the face of challenges such as noise, occlusion, and deformation.

[0065] In a preferred embodiment, the corrosion recognition model of the present invention includes: a component detection model, a corrosion detection model, and a multimodal large model; schematically, first, the present invention can use the YOLOv11 target detection algorithm to train a component detection model and an equipment corrosion detection model, which are respectively used to identify various components in the substation equipment and preliminarily detect whether there is corrosion on these components; the multimodal large model is used to perform secondary verification and identification on the suspected rusted components identified by the equipment corrosion detection model, that is, the multimodal large model combines the image feature extraction model and the text encoding model, and can extract visual features in the image and fuse them with the corrosion features described in the text, so as to more accurately judge the corrosion situation in the image and output the final corrosion result. It can be understood that the YOLOv11 target detection algorithm is a further developed and optimized version based on the YOLO (You Only Look Once) series of algorithms. The target detection task can be regarded as a single regression problem. The bounding box, category, and other information of the target in the image can be predicted through a single forward propagation.

[0066] Specifically, in a preferred embodiment, the present invention can annotate the substation drone visible light inspection images in VOC format, mark the location and category of each component, and then use the annotated images and corresponding annotation files to train component detection models and equipment corrosion detection models based on the YOLOv11 target detection algorithm.

[0067] Furthermore, after training is complete, the component detection model can be used to locate corroded components in the image. After cropping the image of the corroded component, the cropped image is passed to the device's corrosion detection model for initial recognition. The multimodal large model then performs a secondary verification and recognition to determine whether the corroded area in the image is true rust or another phenomenon with similar characteristics, thereby more accurately determining the rust condition in the image. After the multimodal large model extracts image features, it can fuse these features with semantic features related to "rust" to determine whether there is actual rust in the image.

[0068] Therefore, the embodiment of the present invention can filter the recognition results of the initial recognition based on the secondary review results of the multimodal large model, and finally output the true corrosion results, thereby reducing the false detection of the algorithm and improving the recognition accuracy.

[0069] When using the YOLO11 target detection model, the present invention adopts an improved CSP (Cross Stage Partial Networks) architecture, which enhances the feature extraction capability by optimizing the feature flow path in the network. At the same time, it can reduce computational overhead, improve the model's inference speed, reduce redundant calculations, and promote the effective transmission of information.

[0070] This paper also utilizes the PANet (Path Aggregation Network) architecture, a network structure that enhances the feature fusion capabilities of the object detection model. By combining features from different layers, the model's ability to detect objects at different scales is improved. Compared to traditional object detection models, this model architecture offers stronger feature extraction capabilities and improves detection speed while maintaining high accuracy.

[0071] Therefore, the present invention adopts a cascade method of component detection, corrosion detection model to multimodal large model, which reduces the interference of redundant information in the corrosion detection process while ensuring the detection capability of a single model, and can further improve the detection accuracy of rust targets.

[0072] In a preferred embodiment, the component features of the present invention include: texture features, color features, shape features, material features, and edge features;

[0073] When extracting component features corresponding to each component region map for representing the visual features of rust through the rust recognition model, the specific features may be:

[0074] For each component region map, texture features used to describe texture complexity are extracted based on the surface texture changes of the component;

[0075] The color gamut in the component area map is statistically analyzed to extract color features for quantifying the distribution of the rust area.

[0076] Extracting shape features for characterizing the geometric shape of the rusted area from the component area map according to the contour shape of the component in the component area map;

[0077] Capture the long-range dependencies between pixels in the component area map and extract the material features used to characterize the surface material of the component based on the long-range dependencies;

[0078] The edge information of the component area map is captured according to the gradient calculation method of the convolution kernel, and edge features for characterizing the edge contour of the component are generated.

[0079] Schematically, by extracting texture features, we can describe the structural organization and arrangement of a component's surface, reflecting both the microscopic and macroscopic structures of the surface. In rust identification, texture features can be used to distinguish rusted areas from normal areas. Rust often causes changes in surface texture, such as rust spots and rust marks. Extracting texture features can more effectively identify these subtle texture changes, thereby improving the accuracy of rust identification.

[0080] In a preferred embodiment, extracting texture features for describing texture complexity based on surface texture changes of a component includes:

[0081] The shallow convolution kernel in the rust recognition model is used to capture the fine-grained texture of rust particles in the component area map to characterize the component surface.

[0082] The deep convolution kernel in the rust recognition model is used to capture the coarse-grained texture in the component area map that is used to characterize the patchy distribution of the component surface.

[0083] The fine-grained texture and the coarse-grained texture are multi-scale fused to generate texture features for describing texture complexity.

[0084] In the above process, the embodiment of the present invention uses convolutional kernels at different levels within the rust recognition model to capture both fine-grained and coarse-grained textures within the component region image. These textures are then fused at multiple scales to generate texture features that describe texture complexity. The fusion of multi-scale texture features enables the model to better adapt to component images of varying scales and resolutions, maintaining stable recognition performance across diverse scenarios.

[0085] It's understandable that the shallow convolution kernels in the rust recognition model have a smaller receptive field and can capture small-scale texture features on the component surface, such as rust particles. These shallow convolution kernels can identify microstructural changes on the component surface, thus assisting in rust recognition. Deep convolution kernels, on the other hand, have a larger receptive field and can capture large-scale texture features on the component surface, such as patchy rust areas. Coarse-grained texture features reflect the macroscopic structure of the component surface, and deep convolution kernels can identify the overall morphology and distribution of rust areas.

[0086] By fusing fine-grained and coarse-grained textures at multiple scales, we can generate a rich texture feature map. This map encompasses both the microstructure and macrostructure of the component surface, providing a more comprehensive description of the component's texture characteristics. Therefore, by capturing both fine-grained and coarse-grained textures, the present invention can more comprehensively describe the texture characteristics of a component's surface.

[0087] In a preferred embodiment, rust typically causes changes in the surface color of a component, such as reddish brown, yellow, dark brown, etc. By extracting color features, the color distribution of the rusted area can be quantified, further assisting in the judgment of rust identification.

[0088] In rust identification, shape features can be used to identify geometric features such as the boundary, shape, and size of the rust area. By extracting shape features, the rust area can be located more accurately and its severity can be assessed.

[0089] Material features reflect the material information of the component surface, such as metal, plastic, etc. In rust identification, material features can distinguish components of different materials and their rust characteristics, thereby more accurately judging the rust situation.

[0090] In rust identification, edge features can identify the boundaries of components and their rusted areas. By extracting edge features, the contours of the rusted areas can be more clearly outlined, thereby more accurately identifying the rust situation.

[0091] In summary, the embodiments of the present invention can more comprehensively describe the characteristics of a component and its corroded area by extracting the component's texture features, color features, shape features, material features, and edge features, thereby improving the accuracy and efficiency of rust identification.

[0092] In a preferred embodiment, the text data includes: the installation environment of each component, the installation position of each component, the degree of rust of each component, the component color of each component, the component material of each component, and the component structure of each component;

[0093] The step of extracting text features corresponding to each component region diagram and used for representing the semantic description of corrosion according to the text data includes:

[0094] For each component area map, a corrosion risk feature is generated to characterize and describe the corrosion risk level of each component based on the installation environment, installation location, and corrosion level of each component.

[0095] Generate attribute features for characterizing and describing visual attributes of the components according to the component color of each component, the component material of each component, and the component structure of each component;

[0096] A text feature for representing a semantic description of rust is generated based on the corrosion risk feature and the attribute feature.

[0097] Illustratively, the aforementioned text data includes the component's installation environment, location, corrosion level, color, material, and structure, describing various component attributes and conditions. Based on this information, a corrosion risk signature is generated. This signature reflects the potential corrosion risk of a component in different environments and locations, serving as an important basis for assessing component corrosion risk.

[0098] Based on the component color, component material and component structure information in the text data, the attribute features are generated, which describe the visual attributes of the component, such as color, material and structural characteristics. The visual attributes can also be related to the occurrence and development of rust, and can be integrated with the shape features and material features in the component features to further accurately identify whether the component has visual rust.

[0099] Therefore, by combining corrosion risk features and attribute features, the present invention can generate more comprehensive and accurate text features, thereby improving the accuracy and efficiency of rust identification.

[0100] In a preferred embodiment, the fusing of each component feature with each text feature to generate a fused feature includes:

[0101] Through the projection layer in the rust recognition model, each component feature and each text feature are mapped into the same feature space to generate several target component features and several target text features;

[0102] Based on the cross-modal attention mechanism, the features of each target component are fused with the features of each target text to generate fused features.

[0103] Schematically, in order to ensure that component features and text features can be compared and fused on the same dimension, the projection layer of the present invention can use methods such as matrix projection and fully connected layer mapping to convert features of different modalities into target component features and target text features with the same dimension.

[0104] The cross-modal attention mechanism refers to the fusion and alignment of information by calculating the similarity or attention weight between different modal features. In the fusion process of the present invention, each target component feature can be regarded as a "query", and each target text feature can be regarded as a "key-value pair". By calculating the similarity (or attention weight) between the query and the key-value pair, the most relevant text feature can be selected for each component feature for fusion. In principle, the fusion method can be a weighted summation or splicing method.

[0105] Illustratively, within the cross-modal attention mechanism, the present invention can calculate the similarity or attention weight between each target component feature and the target text feature. Based on the calculated similarity or attention weight, the target component features and the target text features are fused to generate fused features. The fused features are then input into subsequent layers of the rust recognition model for further classification and recognition.

[0106] Through the above process, the present invention can achieve the effective fusion of component features and text features, thereby making full use of information from two different modalities to improve the accuracy and efficiency of corrosion recognition.

[0107] In a preferred embodiment, Figure 2As shown in the cross-modal information fusion process diagram, the present invention can use the image feature extraction model in the multimodal large model to extract the image feature map of the image (i.e., the above-mentioned component features). These image feature maps contain visual features such as color, texture, and shape. The visual features are then mapped to a specific embedding space through a multi-layer perceptron projector. At the same time, the input text information (i.e., the above-mentioned text features for semantic description of rust) is also mapped to a specific embedding space using the text encoding model in the multimodal large model. The image information and text information in the same embedding space are fused to achieve the purpose of cross-modal information fusion. Finally, the fused information is passed to the base large model so that the base large model generates results through an autoregressive mechanism, such as "there are traces of rust on the surface of the main transformer" or "there are no traces of rust on the surface of the main transformer."

[0108] Schematically, the process of text feature extraction is the process of text encoding, which converts the content of the text into a text feature matrix with specified dimensions through a text encoding model. The feature matrix is ​​a collection of a fixed number of numbers, where these numbers are arranged in a specific order. These numbers can be understood as the digital form of the meaning of the text, and the text matrices of texts with similar meanings have a higher similarity.

[0109] The fused features contain both visual information such as image texture and color, and the semantic description of the text, allowing the model to integrate the features of both. The fused features are a deep integration of visual information and semantic information of the text. The image features provide information about the visual characteristics of rust such as color, texture, and edges, while the text features introduce semantic priors about the concept of "rust".

[0110] Furthermore, through a fusion mechanism, the model of the present invention can accurately capture key areas related to "rust" in the image and understand whether these areas match the description in the text. Fusion features are essentially a joint encoding of image and text information, containing both visual patterns and semantic guidance. Therefore, the model can use this feature to more comprehensively judge the rust content in the image. The present invention, through a large multimodal model, can effectively improve the model's understanding of the recognition target, thereby improving the algorithm's detection accuracy.

[0111] like Figure 3 As shown, based on the above embodiments of the corrosion identification method for various substation equipment, the present invention provides corresponding device embodiments;

[0112] An embodiment of the present invention provides a corrosion identification device for substation equipment, comprising: a data acquisition module and a corrosion result generation module;

[0113] The data acquisition module is used to acquire a target inspection image corresponding to the substation and text data used to describe each device to be identified in the substation; wherein the target inspection image includes: a plurality of devices to be identified;

[0114] The corrosion result generation module is used to input the target inspection image and the text data into a preset corrosion recognition model, so that the corrosion recognition model divides each device to be identified in the target inspection image into a number of component area maps, and extracts the component features corresponding to each component area map for characterizing the visual features of corrosion; extracts the text features corresponding to each component area map for characterizing the semantic description of corrosion based on the text data; fuses each component feature with each text feature to generate a fused feature, and outputs the corrosion result corresponding to the target inspection image based on the fused feature; wherein the corrosion result includes: a result for characterizing the device with corrosion in the target inspection image, or a result for characterizing the device without corrosion in the target inspection image.

[0115] It should be noted that the device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without paying any creative effort.

[0116] Those skilled in the art can clearly understand that, for the sake of convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0117] Based on the above-mentioned embodiments of the corrosion identification method for various substation equipment, the present invention provides corresponding embodiments of terminal equipment.

[0118] An embodiment of the present invention provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, a corrosion identification method for substation equipment described in any method embodiment of the present invention is implemented.

[0119] The terminal device may be a computing terminal device such as a desktop computer, a notebook computer, a palmtop computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0120] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0121] The memory can be used to store the computer program, and the processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created based on the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device or other volatile solid-state storage device.

[0122] Based on the above-mentioned embodiments of the corrosion identification method for various substation equipment, the present invention provides corresponding embodiments of storage media items.

[0123] An embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a corrosion identification method for substation equipment described in any method embodiment of the present invention.

[0124] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. The computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0125] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying corrosion of substation equipment, characterized in that: include: Obtaining a target inspection image corresponding to the substation and text data describing each device to be identified in the substation; wherein the target inspection image includes: a plurality of devices to be identified; Inputting the target inspection image and the text data into a preset corrosion recognition model, so that the corrosion recognition model segments each device to be identified in the target inspection image into a plurality of component region maps, and extracts component features corresponding to each component region map for characterizing the visual characteristics of corrosion; extracting text features corresponding to each component region map for characterizing the semantic description of corrosion based on the text data; fusing each component feature with each text feature to generate a fused feature, and outputting the corrosion result corresponding to the target inspection image based on the fused feature; The corrosion result includes: a result of a device used to characterize the presence of corrosion in the target inspection image, or a result of a device used to characterize the absence of corrosion in the target inspection image.

2. A method for identifying corrosion of substation equipment according to claim 1, characterized in that: The component features include: texture features, color features, shape features, material features and edge features; The step of extracting component features corresponding to each component region map and used to characterize visual features of corrosion includes: For each component region map, texture features used to describe texture complexity are extracted based on the surface texture changes of the component; The color gamut in the component area map is statistically analyzed to extract color features for quantifying the distribution of the rust area. Extracting shape features for characterizing the geometric shape of the rusted area from the component area map according to the contour shape of the component in the component area map; Capture the long-range dependencies between pixels in the component area map and extract the material features used to characterize the surface material of the component based on the long-range dependencies; The edge information of the component area map is captured according to the gradient calculation method of the convolution kernel, and edge features for characterizing the edge contour of the component are generated.

3. A method for identifying corrosion of substation equipment according to claim 2, characterized in that: The extraction of texture features for describing texture complexity based on the surface texture changes of the component includes: The shallow convolution kernel in the rust recognition model is used to capture the fine-grained texture of rust particles in the component area map to characterize the component surface. The deep convolution kernel in the rust recognition model is used to capture the coarse-grained texture in the component area map that is used to characterize the patchy distribution of the component surface. The fine-grained texture and the coarse-grained texture are multi-scale fused to generate texture features for describing texture complexity.

4. A method for identifying corrosion of substation equipment according to claim 3, characterized in that: The text data includes: the installation environment of each component, the installation position of each component, the rust degree of each component, the component color of each component, the component material of each component, and the component structure of each component; The step of extracting text features corresponding to each component region diagram and used for representing the semantic description of corrosion according to the text data includes: For each component area map, a corrosion risk feature is generated to characterize and describe the corrosion risk level of each component based on the installation environment, installation location, and corrosion level of each component. Generate attribute features for characterizing and describing visual attributes of the components according to the component color of each component, the component material of each component, and the component structure of each component; A text feature for representing a semantic description of rust is generated based on the corrosion risk feature and the attribute feature.

5. A method for identifying corrosion of substation equipment according to claim 4, characterized in that: The fusion of each component feature with each text feature to generate fused features includes: Through the projection layer in the rust recognition model, each component feature and each text feature are mapped into the same feature space to generate several target component features and several target text features; Based on the cross-modal attention mechanism, the features of each target component are fused with the features of each target text to generate fused features.

6. A method for identifying corrosion of substation equipment according to claim 5, characterized in that: The training process of the preset corrosion recognition model includes: The inspection image samples and the text data samples corresponding to the inspection images are used as training samples; wherein the inspection image samples include: a plurality of equipment samples to be identified; Taking each training sample and the actual corrosion result corresponding to each training sample as input and the predicted corrosion result of each training sample as output, the corrosion recognition model to be trained is iteratively trained until the corrosion recognition model converges, thereby generating a completed corrosion recognition model.

7. A method for identifying corrosion of substation equipment according to claim 6, characterized in that: Also includes: During each iteration of training the corrosion recognition model, a training sample is input into the corrosion recognition model, so that the corrosion recognition model generates a corrosion prediction result corresponding to the training sample based on each component feature sample and each text feature sample in the training sample; The corrosion prediction results are compared with the actual corrosion results, and the network parameters of the corrosion recognition model are adjusted according to the comparison results.

8. A corrosion identification device for substation equipment, characterized in that: include: Data acquisition module and corrosion result generation module; The data acquisition module is used to acquire a target inspection image corresponding to the substation and text data used to describe each device to be identified in the substation; wherein the target inspection image includes: a plurality of devices to be identified; The corrosion result generation module is used to input the target inspection image and the text data into a preset corrosion recognition model, so that the corrosion recognition model divides each device to be identified in the target inspection image into a number of component area maps, and extracts the component features corresponding to each component area map for characterizing the visual features of corrosion; extracts the text features corresponding to each component area map for characterizing the semantic description of corrosion based on the text data; fuses each component feature with each text feature to generate a fused feature, and outputs the corrosion result corresponding to the target inspection image based on the fused feature; wherein the corrosion result includes: a result for characterizing the device with corrosion in the target inspection image, or a result for characterizing the device without corrosion in the target inspection image.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for identifying corrosion of substation equipment according to any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the corrosion identification method for substation equipment according to any one of claims 1 to 7.

Images