Training method of power transmission line icing identification model, power transmission line icing identification method and device, computer equipment, storage medium and computer program product

Abstract

The invention relates to a training method and device of a power transmission line icing recognition model, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring an icing image sample set for a power transmission line; inputting the icing image sample into a down-sampling network in a to-be-trained power transmission line icing identification model, and extracting image features of the icing image sample through the down-sampling network; inputting the image features of the icing image sample into an up-sampling network in a to-be-trained power transmission line icing identification model, and generating an icing prediction image based on the image features through the up-sampling network; and training the power transmission line icing identification model according to loss information determined by the icing actual information of the icing image sample and the icing prediction information of the icing prediction image. By adopting the method, the icing condition of the power transmission line can be accurately identified.

Description

Technical Field

[0001] The present application relates to the technical field of power systems, and in particular to a training method, apparatus, computer equipment, storage medium, and computer program product for a transmission line icing recognition model. Background Art

[0002] In winter, when temperatures are low and moisture is abundant, ice easily forms when supercooled water droplets attached to power lines solidify due to the release of latent heat. Large amounts of ice accumulation can trigger a series of serious power accidents, such as line strand breakage, which directly weakens the mechanical strength of the line and affects the stability of power transmission. It can also cause insulator reversal, disrupting their normal insulation performance and potentially leading to leakage, short circuits, and even tower tilting or collapse, paralyzing the entire transmission line and causing widespread power outages. This can cause significant inconvenience to social production and people's lives, and may also lead to major safety accidents.

[0003] Currently, the identification of ice coverage on transmission lines typically involves using digital image processing techniques such as adaptive threshold segmentation and morphological filtering to detect iced power lines from simple background images. Alternatively, an edge detection algorithm using the LOG operator and multi-scale wavelet transforms combined with Hough line detection can be used to identify the edges of iced transmission lines. Alternatively, tension sensors, meteorological sensors, and microwave detection devices are installed on transmission lines to measure parameters such as conductor load, ambient temperature and humidity, air pressure, tower tilt angle, and microwave signal changes. Mathematical models are then used to calculate ice thickness and set alarm thresholds to implement ice monitoring and alarms. As can be seen, existing technologies are unable to accurately identify ice coverage.

[0004] Therefore, traditional technologies have the problem of being unable to accurately identify the icing conditions of transmission lines. Summary of the Invention

[0005] Based on this, it is necessary to provide a training method, device, computer equipment, computer-readable storage medium and computer program product for a transmission line icing identification model that can accurately identify the icing conditions of transmission lines in response to the above technical problems.

[0006] A method for training a transmission line ice recognition model, the method comprising:

[0007] Acquire an ice-covered image sample set for a transmission line; the ice-covered image sample set includes a plurality of ice-covered image samples;

[0008] The ice-covered image sample is input into the downsampling network in the transmission line ice-covered recognition model to be trained. The multi-scale feature extraction unit of the downsampling network extracts the multi-scale feature map of the ice-covered image sample. The semantic feature extraction unit of the downsampling network extracts the semantic feature map of the ice-covered image sample. The feature fusion unit of the downsampling network fuses the multi-scale feature map and the semantic feature map based on the gated attention mechanism to obtain the image features of the ice-covered image sample.

[0009] The image features of the ice image samples are input into the upsampling network in the transmission line ice recognition model to be trained, and the ice prediction image is generated based on the image features through the upsampling network;

[0010] The transmission line icing recognition model is trained based on the actual icing information of the icing image samples and the loss information determined by the icing prediction information of the icing prediction images.

[0011] In one embodiment, obtaining a sample set of ice-covered images of a transmission line includes:

[0012] By simulating an icing environment of a transmission line, an ice thickness image set obtained by a multispectral imager for ice thickness is obtained, and an ice surface image set obtained by a high-speed polarization camera for ice surface is obtained; the ice thickness image set includes ice thickness image subsets corresponding to multiple acquisition moments; the ice thickness image subset corresponding to any acquisition moment includes ice thickness images of different bands acquired at the acquisition moment; the ice surface image set includes multiple ice surface images;

[0013] Performing visualization processing on images in the ice thickness image set to obtain processed images, and segmenting images in the ice surface image set to obtain segmented images;

[0014] An ice-covered image sample set is generated based on each processed image and each segmented image.

[0015] In one embodiment, the images in the ice thickness image set are visualized to obtain processed images, including:

[0016] The principal component analysis model is used to perform principal component analysis on ice thickness images of different bands collected at any acquisition time, and the principal component analysis results corresponding to any acquisition time are obtained;

[0017] Based on the principal component analysis results corresponding to any acquisition time, the ice thickness images of different bands in the ice thickness image subset corresponding to any acquisition time are fused to obtain the fused ice thickness image corresponding to any acquisition time;

[0018] The contrast between the ice layer and the conductor in the fused ice thickness image corresponding to any acquisition time is enhanced by the adaptive histogram equalization method, and the contrast-enhanced ice thickness image corresponding to any acquisition time is obtained.

[0019] The contrast-enhanced ice thickness image corresponding to each acquisition moment is used as each processed image.

[0020] In one embodiment, segmenting images in the ice-covered surface image set to obtain segmented images includes:

[0021] Divide any ice-covered surface image according to a preset window size and a preset window step size to obtain a set of sub-blocks corresponding to any ice-covered surface image;

[0022] Using a quadtree decomposition method, a quadtree decomposition is performed on a set of sub-blocks corresponding to any ice-covered surface image, so as to determine a sub-block containing ice edge features in the set of sub-blocks of any ice-covered surface image as a target sub-block;

[0023] Determining a segmentation method for the ice-covered surface image based on a target sub-block corresponding to any ice-covered surface image;

[0024] According to the segmentation method of each ice-covered surface image, each ice-covered surface image is segmented to obtain each segmented image.

[0025] In one embodiment, the ice cover prediction image is an ice cover area prediction image or an ice cover thickness prediction image; the ice cover area prediction image is determined based on a pixel-level segmentation mask; the pixel-level segmentation mask is used to identify whether each pixel position in the ice cover area prediction image is an ice cover area; the ice cover thickness prediction image is determined based on an ice cover thickness distribution map; the ice cover thickness distribution map is used to identify the ice cover thickness at each pixel position in the ice cover thickness prediction image.

[0026] A method for identifying ice coating on a transmission line, the method comprising:

[0027] Acquire an image of ice cover to be identified on a transmission line;

[0028] The ice image to be identified is input into the downsampling network in the pre-trained transmission line ice identification model. The multi-scale feature extraction unit of the downsampling network extracts the multi-scale feature map of the ice image to be identified. The semantic feature extraction unit of the downsampling network extracts the semantic feature map of the ice image to be identified. The feature fusion unit of the downsampling network fuses the multi-scale feature map and the semantic feature map based on the gated attention mechanism to obtain the image features of the ice image to be identified.

[0029] Input the image features of the ice-covered image to be identified into the upsampling network in the pre-trained transmission line ice-covered recognition model, and generate an ice-covered recognition image based on the image features through the upsampling network;

[0030] Based on the ice cover recognition image, an ice cover condition of the ice cover image to be recognized is determined.

[0031] A training device for a transmission line ice coating recognition model, the device comprising:

[0032] An acquisition module is used to acquire an ice-covered image sample set for a transmission line; the ice-covered image sample set includes a plurality of ice-covered image samples;

[0033] A feature extraction module is used to input ice-covered image samples into a downsampling network in a transmission line ice-covered recognition model to be trained, extract a multi-scale feature map of the ice-covered image samples through a multi-scale feature extraction unit of the downsampling network, extract a semantic feature map of the ice-covered image samples through a semantic feature extraction unit of the downsampling network, and fuse the multi-scale feature map and the semantic feature map through a feature fusion unit of the downsampling network based on a gated attention mechanism to obtain image features of the ice-covered image samples;

[0034] An image generation module is used to input the image features of the ice image sample into the upsampling network in the transmission line ice recognition model to be trained, and generate an ice prediction image based on the image features through the upsampling network;

[0035] The training module is used to train the transmission line icing recognition model according to the actual icing information of the icing image samples and the loss information determined by the icing prediction information of the icing prediction image.

[0036] A computer device includes a memory and a processor, wherein the memory stores a computer program and the processor implements the steps of the above method when executing the computer program.

[0037] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0038] A computer program product comprises a computer program, which implements the steps of the above method when executed by a processor.

[0039] The training method, device, computer equipment, storage medium and computer program product of the transmission line icing recognition model are as follows: obtaining an icing image sample set for the transmission line; the icing image sample set includes multiple icing image samples; inputting the icing image samples into a downsampling network in the transmission line icing recognition model to be trained; extracting a multi-scale feature map of the icing image samples through a multi-scale feature extraction unit of the downsampling network; extracting a semantic feature map of the icing image samples through a semantic feature extraction unit of the downsampling network; and fusion of the multi-scale feature maps based on a gated attention mechanism by a feature fusion unit of the downsampling network. The feature map and the semantic feature map are fused to obtain the image features of the ice-covered image samples; the image features of the ice-covered image samples are input into the upsampling network in the transmission line icing recognition model to be trained, and the icing prediction image is generated based on the image features through the upsampling network; the transmission line icing recognition model is trained according to the loss information determined by the actual icing information of the ice-covered image samples and the icing prediction information of the icing prediction image; in this way, the transmission line icing recognition model for accurately identifying the icing condition of the transmission line can be trained quickly and efficiently, which is conducive to accurately and efficiently identifying the icing condition of the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 A diagram illustrating an application environment for a training method for a transmission line ice coating recognition model in one embodiment;

[0042] Figure 2 A schematic flow chart of a method for training a transmission line ice recognition model in one embodiment;

[0043] Figure 3 1 is a flow chart of a method for identifying ice coating on a power transmission line according to an embodiment;

[0044] Figure 4 Schematic diagram of a flow chart of a method for training a transmission line icing recognition model in another embodiment;

[0045] Figure 5 A structural block diagram of a training device for a transmission line icing recognition model in one embodiment;

[0046] Figure 6 This is a structural block diagram of a device for identifying ice coating on a power transmission line according to an embodiment;

[0047] Figure 7 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0049] The training method of the transmission line ice recognition model provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, the terminal 102 communicates with the server 104 via a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104 or placed on the cloud or other network servers. The server 104 obtains an ice-covered image sample set for the transmission line; the ice-covered image sample set includes multiple ice-covered image samples; the server 104 inputs the ice-covered image samples into the downsampling network in the transmission line ice-covered recognition model to be trained, extracts the multi-scale feature map of the ice-covered image samples through the multi-scale feature extraction unit of the downsampling network, extracts the semantic feature map of the ice-covered image samples through the semantic feature extraction unit of the downsampling network, and fuses the multi-scale feature map and the semantic feature map based on the gated attention mechanism through the feature fusion unit of the downsampling network to obtain the image features of the ice-covered image samples; the server 104 inputs the image features of the ice-covered image samples into the upsampling network in the transmission line ice-covered recognition model to be trained, and generates an ice-covered prediction image based on the image features through the upsampling network; the server 104 trains the transmission line ice-covered recognition model according to the loss information determined by the actual ice-covered information of the ice-covered image samples and the ice-covered prediction information of the ice-covered prediction image. Terminal 102 may include, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Server 104 may be implemented as a standalone server or a server cluster consisting of multiple servers.

[0050] In an exemplary embodiment, Figure 2 As shown in the figure, a training method for the transmission line ice recognition model is provided, and the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the method, which includes the following steps S202 to S208.

[0051] Step S202: Acquire an ice-covered image sample set for a transmission line; the ice-covered image sample set includes a plurality of ice-covered image samples.

[0052] The ice-covered image samples may be obtained by pre-processing the original ice-covered images of the power transmission lines captured by the image acquisition device.

[0053] Optionally, the server obtains a sample set of ice-covered images of the transmission line.

[0054] In step S204, the ice-covered image sample is input into the downsampling network in the transmission line ice-covered recognition model to be trained, and the multi-scale feature extraction unit of the downsampling network extracts the multi-scale feature map of the ice-covered image sample. The semantic feature extraction unit of the downsampling network extracts the semantic feature map of the ice-covered image sample. The feature fusion unit of the downsampling network fuses the multi-scale feature map and the semantic feature map based on the gated attention mechanism to obtain the image features of the ice-covered image sample.

[0055] Among them, the transmission line ice recognition model to be trained can be a model that adopts a hybrid Transformer-CNN architecture with multi-scale feature fusion and introduces an adversarial training mechanism with physical constraints.

[0056] Among them, the downsampling network can include a parallel multi-scale feature extraction unit, a semantic feature extraction unit and a feature fusion unit, the multi-scale feature extraction unit is used to output a multi-scale feature map, the semantic feature extraction unit is used to output a semantic feature map, and the feature fusion unit is used to fuse the multi-scale feature map output by the multi-scale feature extraction unit and the semantic feature map output by the semantic feature extraction unit.

[0057] In practical applications, the downsampling network can include parallel CNN branches and Transformer branches, as well as feature fusion units using a gated attention mechanism. The CNN branch uses a stack of dilated convolutional layers with a dilation rate of [2, 4, 6] to output a multi-scale feature map { , , The Transformer branch divides the input ice-covered image samples into 16×16 sequences, calculates global dependencies through multi-head self-attention, and outputs semantic feature maps. The feature fusion unit uses the gated attention mechanism to fuse the multi-scale feature map and the semantic feature map. The calculation formula is:

[0058] ,

[0059] ,

[0060] Among them, σ is the sigmoid function, is the learnable parameter matrix, and ⊙ represents element-by-element multiplication.

[0061] Among them, image features can represent the deep image information of ice-covered image samples.

[0062] Optionally, the server inputs the ice-covered image sample into the downsampling network in the transmission line ice-covered recognition model to be trained, extracts the multi-scale feature map of the ice-covered image sample through the multi-scale feature extraction unit of the downsampling network, extracts the semantic feature map of the ice-covered image sample through the semantic feature extraction unit of the downsampling network, and fuses the multi-scale feature map and the semantic feature map based on the gated attention mechanism through the feature fusion unit of the downsampling network to obtain the image features of the ice-covered image sample.

[0063] Step S206: input the image features of the ice-covered image samples into an upsampling network in the transmission line ice-covered recognition model to be trained, and generate an ice-covered prediction image based on the image features through the upsampling network.

[0064] The upsampling network can include a physically driven deconvolution layer, where the discretized form of the ice growth partial differential equation is embedded in the deconvolution kernel weights:

[0065] ;

[0066] in, is a constraint term based on the Navier-Stokes equations (NS equations) of fluid mechanics, and α and β are trainable coefficients.

[0067] Among them, the ice coverage prediction image can be a segmented image with the same resolution as the ice-covered line image sample, and the information of each pixel point indicates whether the location belongs to the ice-covered area or not. The ice coverage prediction image can also be a thickness distribution map with the same resolution as the ice-covered line image sample, and the information of each pixel point indicates the ice thickness at the location.

[0068] Optionally, the server inputs the image features of the ice-covered image samples into an upsampling network in a transmission line ice-covered recognition model to be trained, and generates an ice-covered prediction image based on the image features through the upsampling network.

[0069] In practical applications, the image size is restored through an upsampling network, and deconvolution is used to complete the upsampling, expanding the feature size to the original image size. During the upsampling process, low-level features can be fused to improve image segmentation accuracy or thickness prediction accuracy.

[0070] Step S208 : training a transmission line icing recognition model based on the actual icing information of the icing image samples and the loss information determined by the icing prediction information of the icing prediction images.

[0071] The actual ice coverage information may refer to the actual ice coverage area and actual ice coverage thickness of the ice coverage image sample, which can be determined by pre-marking.

[0072] The ice cover prediction information may refer to the predicted ice cover area and predicted ice cover thickness obtained by the model identifying the ice cover image samples.

[0073] The loss information may be a mean square error.

[0074] In the actual training process, the mean square error (MSE) is selected as the loss function. MSE measures the gap between the model's predicted value and the true value by taking the average of the square of the difference between each predicted value and the true value. The formula is:

[0075] ,

[0076] Among them, the smaller the MSE value, the better the model.

[0077] In the actual training process, the Adam algorithm is used to update the parameters during the training process. The Adam algorithm designs independent adaptive learning rates for different parameters by calculating the first-order moment estimate and the second-order moment estimate of the gradient. The parameter update method is:

[0078] ,

[0079] Among them, β1 and β2 are the first-order moment estimates v t and the second-order moment estimate s t The exponential decay factor, α is the initial learning rate, and ε is a very small positive number to prevent the denominator from being 0. This application sets the parameters , in order to speed up the convergence of the model.

[0080] Optionally, the server trains the transmission line icing recognition model according to mean square error loss information determined based on actual icing information of the icing image samples and icing prediction information of the icing prediction images.

[0081] In the training method of the above-mentioned transmission line icing recognition model, an icing image sample set for the transmission line is obtained; the icing image sample set includes multiple icing image samples; the icing image samples are input into the downsampling network in the transmission line icing recognition model to be trained, and the multi-scale feature map of the icing image samples is extracted by the multi-scale feature extraction unit of the downsampling network, and the semantic feature map of the icing image samples is extracted by the semantic feature extraction unit of the downsampling network, and the multi-scale feature map and the semantic feature are fused by the feature fusion unit of the downsampling network based on the gated attention mechanism. The images are fused to obtain image features of ice-covered image samples; the image features of the ice-covered image samples are input into the upsampling network in the transmission line icing recognition model to be trained, and an icing prediction image is generated based on the image features through the upsampling network; the transmission line icing recognition model is trained according to the loss information determined by the actual icing information of the ice-covered image samples and the icing prediction information of the icing prediction image; in this way, the transmission line icing recognition model for accurately identifying icing conditions can be trained quickly and efficiently, which is conducive to accurately and efficiently identifying icing conditions of transmission lines.

[0082] In an exemplary embodiment, an ice cover image sample set for a transmission line is obtained, including: obtaining an ice cover thickness image set obtained by a multispectral imager for ice cover thickness by simulating an icing environment of the transmission line, and obtaining an ice cover surface image set obtained by a high-speed polarization camera for ice cover surface; the ice cover thickness image set includes ice cover thickness image subsets corresponding to multiple acquisition moments; the ice cover thickness image subset corresponding to any acquisition moment includes ice cover thickness images of different bands acquired at the acquisition moment; the ice cover surface image set includes multiple ice cover surface images; visual processing is performed on the images in the ice cover thickness image set to obtain processed images, and the images in the ice cover surface image set are segmented to obtain segmented images; and an ice cover image sample set is generated based on the processed images and the segmented images.

[0083] Among them, the high-speed polarized light camera can take high-frequency photos of the ice-covered surface, thereby recording the microscopic texture information of the ice-covered surface. After processing and analysis, this information can be used as polarized light texture features.

[0084] The processed image may be an image of ice thickness of the transmission line with higher image quality.

[0085] The segmented image may be an image of the ice-covered surface of the transmission line that retains the features of the ice-covered surface but has a smaller image.

[0086] Optionally, by simulating the icing environment of the transmission line, including precisely controlling the temperature gradient to -10°C to 0°C, the humidity gradient to 85%-95%, and the wind speed to 3-5m / s in an artificial climate chamber, simulating the natural icing formation conditions, using a multispectral imager to synchronously collect ice thickness images in the visible light band and the near-infrared band, and recording the environmental parameter matrix Q = [temperature, humidity, wind speed, icing time] during the shooting, and using a high-speed polarized light camera to capture the dynamic growth process of the microstructure on the ice surface at a rate of 120 frames per second, and associating the environmental parameters with the corresponding timestamps, the server then obtains an ice thickness image set obtained by the multispectral imager for ice thickness collection, and obtains an ice surface image set obtained by the high-speed polarized light camera for ice surface collection, and then performs visualization processing on the images in the ice thickness image set to obtain each processed image, and the images in the ice surface image set are segmented to obtain each segmented image, and an ice image sample set is generated based on each processed image and each segmented image.

[0087] In actual applications, when collecting images, it is necessary to select conductors with a certain ice thickness in a simulated transmission line icing environment, set up a multispectral imager to shoot in the direction of the light, and avoid interfering backgrounds such as buildings, plants, and the sky as much as possible. When using a high-speed polarized light camera, it is necessary to capture the surface conditions of the ice-covered lines.

[0088] In this application, acquiring multiple types of ice image samples facilitates model training. In transmission line icing research, different types of images provide comprehensive icing information. Therefore, collecting images of conductors with a certain ice thickness can clarify the impact of ice on conductor thickness, while collecting images of the iced surface of transmission lines can reveal features such as morphology and texture. Combining this information from different dimensions enriches the sample set of icing images.

[0089] In this embodiment, by simulating the icing environment of the transmission line, an ice thickness image set obtained by a multispectral imager for ice thickness collection is obtained, and an ice surface image set obtained by a high-speed polarization camera for ice surface collection is obtained; the ice thickness image set includes ice thickness image subsets corresponding to multiple collection moments; the ice thickness image subset corresponding to any collection moment includes ice thickness images of different bands collected at the collection moment; the ice surface image set includes multiple ice surface images; the images in the ice thickness image set are visualized to obtain processed images, and the images in the ice surface image set are segmented to obtain segmented images; based on the processed images and the segmented images, an ice image sample set is generated; in this way, the ice image sample set covers more types of ice images, providing a rich data basis for training an accurate transmission line ice recognition model. At the same time, visualizing or segmenting the collected original images is conducive to improving the image quality input to the model, and is conducive to training an accurate transmission line ice recognition model.

[0090] In an exemplary embodiment, images in an ice thickness image set are visualized to obtain processed images, including: using a principal component analysis model to perform principal component analysis on ice thickness images of different bands collected at any acquisition moment to obtain a principal component analysis result corresponding to any acquisition moment; based on the principal component analysis result corresponding to any acquisition moment, ice thickness images of different bands in an ice thickness image subset corresponding to any acquisition moment are fused to obtain a fused ice thickness image corresponding to any acquisition moment; enhancing the contrast between the ice layer and the conductor in the fused ice thickness image corresponding to any acquisition moment by an adaptive histogram equalization method to obtain a contrast-enhanced ice thickness image corresponding to any acquisition moment; and using the contrast-enhanced ice thickness image corresponding to each acquisition moment as each processed image.

[0091] Optionally, the server can use principal component analysis to analyze ice thickness images of different bands collected at any acquisition moment, extract the first three principal component features, and fuse the ice thickness images of different bands in the ice thickness image subset corresponding to any acquisition moment based on the extracted first three principal component features to obtain a fused ice thickness image corresponding to any acquisition moment, and then enhance the contrast between the ice layer and the conductor in the fused ice thickness image corresponding to any acquisition moment through adaptive histogram equalization, so as to obtain a contrast-enhanced ice thickness image corresponding to any acquisition moment, and use the contrast-enhanced ice thickness images corresponding to each acquisition moment as each processed image to realize visualization processing of the ice thickness image set.

[0092] In this embodiment, by adopting the principal component analysis model, principal component analysis is performed on the ice thickness images of different bands collected at any acquisition moment to obtain the principal component analysis results corresponding to any acquisition moment; based on the principal component analysis results corresponding to any acquisition moment, the ice thickness images of different bands in the ice thickness image subset corresponding to any acquisition moment are fused to obtain the fused ice thickness image corresponding to any acquisition moment; the contrast between the ice layer and the conductor in the fused ice thickness image corresponding to any acquisition moment is enhanced by the adaptive histogram equalization method to obtain the contrast-enhanced ice thickness image corresponding to any acquisition moment. The contrast-enhanced ice thickness image corresponding to each acquisition moment is used as each processed image. In this way, principal component analysis can be used to extract principal component features, thereby fusing images of multiple bands. By integrating the information of these different bands, the information of each band can be fully utilized, so that the fused image can more comprehensively represent the characteristics of the ice layer and the conductor than the image of a single band. By enhancing the contrast between the ice layer and the conductor through adaptive histogram equalization, the pixel grayscale value distribution of the ice layer and the conductor can be more finely adjusted, which is conducive to more clearly determining the characteristics of the ice layer and the conductor, and facilitating better distinction between the boundaries and thickness of the ice layer and the conductor.

[0093] In an exemplary embodiment, images in an ice-covered surface image set are segmented to obtain segmented images, including: dividing any ice-covered surface image according to a preset window size and a preset window step size to obtain a set of sub-blocks corresponding to any ice-covered surface image; performing quadtree decomposition on the set of sub-blocks corresponding to any ice-covered surface image to determine a sub-block containing ice edge features in the set of sub-blocks of any ice-covered surface image as a target sub-block; determining a segmentation method for the ice-covered surface image based on the target sub-block corresponding to any ice-covered surface image; and segmenting each ice-covered surface image according to the segmentation method to obtain segmented images.

[0094] The preset window size may be 512×512 pixels.

[0095] The preset window step size may be 256 pixels.

[0096] Quadtree decomposition is a segmentation method used in image processing that decomposes an image layer by layer into multiple sub-blocks. In a quadtree structure, each parent node represents a larger image block, which can be decomposed into four equally sized child nodes (sub-blocks). This decomposition process can continue until a preset stopping condition is met, such as reaching a specific image resolution level or satisfying certain image feature conditions.

[0097] Optionally, the server uses a sliding window method to divide any ice-covered surface image according to a window size of 512×512 pixels and a step size of 256 pixels to obtain a set of sub-blocks corresponding to any ice-covered surface image. Then, the server performs quadtree decomposition on the set of sub-blocks corresponding to any ice-covered surface image, retains the sub-blocks containing ice edge features, determines them as target sub-blocks, and attaches corresponding environmental parameter vectors as metadata tags. The server determines a segmentation method for the ice-covered surface image; and segments each ice-covered surface image according to the segmentation method of each ice-covered surface image to obtain each segmented image.

[0098] In this embodiment, any ice-covered surface image is divided according to a preset window size and a preset window step size to obtain a set of sub-blocks corresponding to any ice-covered surface image; a quadtree decomposition method is used to perform quadtree decomposition on the set of sub-blocks corresponding to any ice-covered surface image, so as to determine sub-blocks in the set of sub-blocks of any ice-covered surface image that contain ice edge features as target sub-blocks; based on the target sub-blocks corresponding to any ice-covered surface image, a segmentation method for the ice-covered surface image is determined; and according to the segmentation method of each ice-covered surface image, each ice-covered surface image is segmented to obtain each segmented image; in this way, the sub-blocks containing ice edge features can be retained, thereby effectively reducing the computational complexity while retaining important information in the image.

[0099] In an exemplary embodiment, the ice cover prediction image is an ice cover area prediction image or an ice cover thickness prediction image; the ice cover area prediction image is determined based on a pixel-level segmentation mask; the pixel-level segmentation mask is used to identify whether each pixel position in the ice cover area prediction image is an ice cover area; the ice cover thickness prediction image is determined based on an ice cover thickness distribution map; the ice cover thickness distribution map is used to identify the ice cover thickness at each pixel position in the ice cover thickness prediction image.

[0100] Among them, the upsampling network of the transmission line ice recognition model is connected to the output layer, which can generate pixel-level segmentation masks , used to accurately identify ice-covered areas and generate ice thickness distribution maps , intuitively showing the distribution of ice thickness.

[0101] in, In the image, M represents the pixel-level segmentation mask, which is used to identify whether each pixel in the image belongs to the ice-covered area; {0, 1} represents the value range, 0 represents that the pixel does not belong to the ice-covered area, and 1 represents that the pixel belongs to the ice-covered area; H and W represent the height and width of the image, respectively. H×W determines the number and layout of image pixels, which means that M is a two-dimensional matrix of size H×W, and each element in the matrix corresponds to the ice-covered identity of a pixel in the image.

[0102] in, In the figure, T represents the ice thickness distribution map, which is used to show the distribution of ice thickness on the transmission line; R represents the real number set, which means that the elements in the T matrix are real numbers. These real numbers are used to quantify the ice thickness corresponding to each pixel position; H×W indicates that the size of the matrix is consistent with the image pixel layout, and each element corresponds to the ice thickness of a pixel position in the image.

[0103] In this embodiment, the ice cover prediction image is an ice cover area prediction image or an ice cover thickness prediction image. The ice cover area prediction image is determined based on the pixel-level segmentation mask, and the ice cover thickness prediction image is determined based on the ice cover thickness distribution map. The model can output a refined ice cover prediction image.

[0104] In an exemplary embodiment, Figure 3 As shown in the figure, a method for identifying ice coating on a transmission line is provided. Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0105] Step S302: Acquire an image of ice coverage to be identified on the transmission line.

[0106] The ice-covered image to be identified may be an image of a power transmission line whose ice-covered area and ice-covered thickness need to be determined.

[0107] Optionally, the server obtains an image of ice coverage to be identified on the transmission line.

[0108] In step S304, the ice-covered image to be identified is input into the downsampling network in the pre-trained transmission line ice-covered identification model. The multi-scale feature extraction unit of the downsampling network extracts the multi-scale feature map of the ice-covered image to be identified. The semantic feature extraction unit of the downsampling network extracts the semantic feature map of the ice-covered image to be identified. The feature fusion unit of the downsampling network fuses the multi-scale feature map and the semantic feature map based on the gated attention mechanism to obtain the image features of the ice-covered image to be identified.

[0109] Optionally, the server inputs the ice-covered image to be identified into the downsampling network in the pre-trained transmission line ice-covered recognition model, extracts the multi-scale feature map of the ice-covered image to be identified through the multi-scale feature extraction unit of the downsampling network, extracts the semantic feature map of the ice-covered image to be identified through the semantic feature extraction unit of the downsampling network, and fuses the multi-scale feature map and the semantic feature map based on the gated attention mechanism through the feature fusion unit of the downsampling network to obtain the image features of the ice-covered image to be identified.

[0110] Step S306: input the image features of the ice-covered image to be identified into the upsampling network in the pre-trained transmission line ice-covered identification model, and generate an ice-covered identification image based on the image features through the upsampling network.

[0111] Optionally, the server inputs the image features of the ice-covered image to be identified into an upsampling network in a pre-trained transmission line ice-covered identification model, and generates an ice-covered identification image based on the image features through the upsampling network.

[0112] Step S308: determining the ice condition of the ice image to be identified based on the ice identification image.

[0113] Optionally, the server determines the ice coverage condition of the ice coverage image to be identified based on the ice coverage identification image.

[0114] In practical applications, icing conditions can include geometric parameters such as the area, perimeter, and center of gravity of the iced area, as well as information on the severity of the iced area (such as light ice, moderate ice, and heavy ice). This detailed output information allows operation and maintenance personnel to more intuitively understand the actual situation of iced areas on transmission lines, providing strong support for the development of targeted de-icing measures and operation and maintenance plans.

[0115] In practical applications, the pre-trained transmission line ice recognition model can be integrated into the real-time transmission line monitoring system to enable dynamic monitoring of iced areas and timely warnings. Integrating geographic information system (GIS) technology, identified iced areas are visually annotated on a transmission line map, making it easier for operations and maintenance personnel to quickly locate and view them. When the area or severity of an iced area exceeds a preset threshold, the system automatically issues a warning, notifying relevant personnel to take appropriate measures, effectively mitigating the damage posed by ice to the transmission line.

[0116] The above-mentioned method for identifying ice on transmission lines obtains an ice image to be identified on the transmission line; inputs the ice image to be identified into a downsampling network in a pre-trained transmission line ice recognition model, extracts a multi-scale feature map of the ice image to be identified through a multi-scale feature extraction unit of the downsampling network, extracts a semantic feature map of the ice image to be identified through a semantic feature extraction unit of the downsampling network, fuses the multi-scale feature map and the semantic feature map based on a gated attention mechanism through a feature fusion unit of the downsampling network to obtain image features of the ice image to be identified; inputs the image features of the ice image to be identified into an upsampling network in a pre-trained transmission line ice recognition model, generates an ice recognition image based on the image features through the upsampling network; and determines the ice condition of the ice image to be identified based on the ice recognition image; in this way, the pre-trained transmission line ice recognition model can efficiently and accurately identify the ice condition in the ice line image to be identified.

[0117] The transmission line icing identification method of the present application corresponds to a transmission line icing identification system, which may include four modules, namely, an image acquisition module, a data processing module, a model training module and an icing identification module.

[0118] The image acquisition module is used to capture images of ice-covered transmission lines. An outdoor experimental test platform is built to simulate the icing environment of transmission lines. Conductors with a certain ice thickness are selected. A camera (e.g., a multispectral imager) is used to capture images in the direction of the sun, minimizing interference from backgrounds such as buildings, plants, and the sky. The camera is rotated at a certain angle to capture different images. These images are then processed for clear visualization, selecting images with high resolution of ice edges and a well-centered field of view. A camera (e.g., a high-speed polarized light camera) is also used to capture the surface conditions of the ice-covered lines. The original 3840×2160 ice-covered line images are segmented to reduce the amount of data to be processed while preserving the characteristics of the ice images and increasing data diversity.

[0119] Among them, the data processing module is used to preprocess the collected image data and produce a transmission line icing dataset. The dataset contains three types of objects: transmission lines, ice, and background, and is divided into 800 training sets, 200 validation sets, and 200 test sets.

[0120] When generating the sample set, this application also constructs a physics-based adversarial sample generation network to generate an enhanced dataset with physical authenticity by solving the ice growth partial differential equation. The equation is expressed as:

[0121] ;

[0122] Where ρ is the ice density, D is the temperature-dependent diffusion coefficient, k is the freezing rate, and φ is the airflow distribution function. By rendering the solution of this equation as a three-dimensional projection of the ice layer structure and mixing it with the real collected data in a ratio of 7:3, an enhanced dataset with physical authenticity can be formed. By introducing innovations such as multimodal data fusion, enhanced physical modeling, and environmental parameter correlation, the sample set of this application significantly surpasses traditional sample sets in terms of data diversity, physical authenticity, and feature richness.

[0123] The model training module is used to train the hybrid Transformer-CNN model with multi-scale feature fusion using the prepared transmission line ice dataset, setting the mean square error as the loss function, the Adam algorithm as the optimizer, and the parameters as , use the directional propagation algorithm for training, and monitor the loss value and accuracy of the validation set during training. When it drops to 0.105, the model converges and training stops. At this time, the accuracy on the test set is 0.98.

[0124] In the model training phase of this application, an adversarial sample generator G is introduced to enhance the robustness and generalization ability of the model. Generator G generates adversarial samples based on the input image data, and its network structure satisfies:

[0125] ,

[0126] Here, G represents the adversarial example generator network model. It generates adversarial examples based on the features of the input image and the results of the coordinate-aware convolutional layer (CoordConv). The CoordConv layer adds coordinate information to the traditional convolutional layer, enabling the model to better capture the spatial relationships between pixels in the image, helping to generate more targeted adversarial examples. This adversarial example generation simulates the complex conditions and noise interference found in actual transmission line ice-covered images. This allows the model to learn how to cope with various changes during training, thereby improving its ability to recognize different ice conditions and ultimately enhancing the performance and accuracy of the entire transmission line ice-covered recognition model.

[0127] The icing recognition module is used to test the test set using the trained model to identify ice on the transmission lines. The model's recognition accuracy is evaluated using the Intersection over Union (MIoU) calculation formula:

[0128] ,

[0129] Among them, k+1 is the number of categories (including background category), represents the number of pixels that belong to class i but are predicted to be class j, Indicates the number of correctly predicted pixels. After testing, the model achieved an MIoU value of 80.1% on the transmission line ice dataset.

[0130] Compared to the FCN detection algorithm, the transmission line ice recognition method of this application improves recognition accuracy for both simple and complex backgrounds. It demonstrates strong robustness against iced conductors in different background categories, achieving a detection accuracy rate exceeding 85.9%. This application, based on a training model using a small number of iced images, enables rapid and accurate detection of iced conductors, improving the accuracy of intelligent transmission line ice recognition.

[0131] This application's transmission line ice recognition method, based on semantic segmentation using deep learning, does not rely on any prior features and can classify every pixel in an image. This method offers significant advantages over traditional digital image processing methods. It accurately identifies even irregular ice patterns, achieving an average intersection-over-union (IoU) ratio of 80.1%. This rapid recognition speed, which expedited computational resources, optimized model training and recognition speed and efficiency, and achieved runtimes that met engineering requirements, demonstrates excellent application value.

[0132] In another embodiment, Figure 4 As shown in the figure, a training method for the transmission line ice recognition model is provided, and the method is applied to Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0133] Step S402: By simulating the icing environment of the transmission line, an ice thickness image set obtained by a multispectral imager for ice thickness is obtained, and an ice surface image set obtained by a high-speed polarization camera for ice surface is obtained; the ice thickness image set includes ice thickness image subsets corresponding to multiple acquisition moments; the ice thickness image subset corresponding to any acquisition moment includes ice thickness images of different bands acquired at the acquisition moment.

[0134] Step S404 : performing visualization processing on the images in the ice thickness image set to obtain processed images, and segmenting the images in the ice surface image set to obtain segmented images.

[0135] Step S406: Generate an ice-covered image sample set based on each processed image and each segmented image.

[0136] Step S408: Input the ice-covered image sample into the downsampling network in the transmission line ice-covered recognition model to be trained, extract the multi-scale feature map of the ice-covered image sample through the multi-scale feature extraction unit of the downsampling network, extract the semantic feature map of the ice-covered image sample through the semantic feature extraction unit of the downsampling network, and fuse the multi-scale feature map and the semantic feature map based on the gated attention mechanism through the feature fusion unit of the downsampling network to obtain the image features of the ice-covered image sample.

[0137] Step S410: input the image features of the ice-covered image sample into an upsampling network in the transmission line ice-covered recognition model to be trained, and generate an ice-covered prediction image based on the image features through the upsampling network.

[0138] Step S412: Training a transmission line icing recognition model based on the actual icing information of the icing image samples and the loss information determined by the icing prediction information of the icing prediction images.

[0139] It should be noted that the specific limitations of the above steps can refer to the specific limitations of the training method for a transmission line icing recognition model mentioned above.

[0140] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0141] Based on the same inventive concept, embodiments of the present application also provide a training device for a transmission line icing recognition model for implementing the aforementioned training method for a transmission line icing recognition model. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the training device for a transmission line icing recognition model provided below can be found in the limitations of the training method for a transmission line icing recognition model described above and will not be further elaborated here.

[0142] In an exemplary embodiment, Figure 5 As shown, a training device for a transmission line ice coating recognition model is provided, comprising: an acquisition module 502, a feature extraction module 504, an image generation module 506 and a training module 508, wherein:

[0143] An acquisition module 502 is configured to acquire an ice-covered image sample set for a transmission line; the ice-covered image sample set includes a plurality of ice-covered image samples;

[0144] A feature extraction module 504 is configured to input the ice-covered image sample into a downsampling network in a transmission line ice-covered recognition model to be trained, extract a multi-scale feature map of the ice-covered image sample through a multi-scale feature extraction unit of the downsampling network, extract a semantic feature map of the ice-covered image sample through a semantic feature extraction unit of the downsampling network, and fuse the multi-scale feature map and the semantic feature map through a feature fusion unit of the downsampling network based on a gated attention mechanism to obtain image features of the ice-covered image sample;

[0145] An image generation module 506 is configured to input image features of the ice image sample into an upsampling network in the transmission line ice recognition model to be trained, and generate an ice prediction image based on the image features through the upsampling network;

[0146] The training module 508 is configured to train a transmission line icing recognition model based on actual icing information of the icing image samples and loss information determined by icing prediction information of the icing prediction images.

[0147] In one embodiment, the acquisition module 502 is specifically used to obtain an ice thickness image set obtained by a multispectral imager for ice thickness collection, and an ice surface image set obtained by a high-speed polarization camera for ice surface collection by simulating an icing environment of a transmission line; the ice thickness image set includes ice thickness image subsets corresponding to multiple collection moments; the ice thickness image subset corresponding to any collection moment includes ice thickness images of different bands collected at the collection moment; the ice surface image set includes multiple ice surface images; the images in the ice thickness image set are visualized to obtain processed images, and the images in the ice surface image set are segmented to obtain segmented images; and an ice image sample set is generated based on the processed images and the segmented images.

[0148] In one embodiment, the acquisition module 502 is specifically used to use a principal component analysis model to perform principal component analysis on ice thickness images of different bands collected at any acquisition moment, and obtain a principal component analysis result corresponding to any acquisition moment; based on the principal component analysis result corresponding to any acquisition moment, the ice thickness images of different bands in the ice thickness image subset corresponding to any acquisition moment are fused to obtain a fused ice thickness image corresponding to any acquisition moment; the contrast between the ice layer and the conductor in the fused ice thickness image corresponding to any acquisition moment is enhanced by an adaptive histogram equalization method to obtain a contrast-enhanced ice thickness image corresponding to any acquisition moment; and the contrast-enhanced ice thickness image corresponding to each acquisition moment is used as each processed image.

[0149] In one embodiment, the acquisition module 502 is specifically used to divide any ice-covered surface image according to a preset window size and a preset window step size to obtain a set of sub-blocks corresponding to any ice-covered surface image; use a quadtree decomposition method to perform quadtree decomposition on the set of sub-blocks corresponding to any ice-covered surface image, so as to determine the sub-blocks containing ice edge features in the set of sub-blocks of any ice-covered surface image as target sub-blocks; based on the target sub-blocks corresponding to any ice-covered surface image, determine a segmentation method for the ice-covered surface image; and segment each ice-covered surface image according to the segmentation method of each ice-covered surface image to obtain each segmented image.

[0150] In one embodiment, the ice cover prediction image is an ice cover area prediction image or an ice cover thickness prediction image; the ice cover area prediction image is determined based on a pixel-level segmentation mask; the pixel-level segmentation mask is used to identify whether each pixel position in the ice cover area prediction image is an ice cover area; the ice cover thickness prediction image is determined based on an ice cover thickness distribution map; the ice cover thickness distribution map is used to identify the ice cover thickness at each pixel position in the ice cover thickness prediction image.

[0151] In an exemplary embodiment, Figure 6 As shown, a transmission line ice coating identification device is provided, comprising: an image acquisition module 602, an image input module 604, an image output module 606 and an identification module 608, wherein:

[0152] An image acquisition module 602 is configured to acquire an image of ice coverage to be identified on a transmission line;

[0153] Image input module 604 is used to input the ice image to be identified into the downsampling network in the pre-trained transmission line ice identification model, extract a multi-scale feature map of the ice image to be identified through the multi-scale feature extraction unit of the downsampling network, extract a semantic feature map of the ice image to be identified through the semantic feature extraction unit of the downsampling network, and fuse the multi-scale feature map and the semantic feature map based on the gated attention mechanism through the feature fusion unit of the downsampling network to obtain image features of the ice image to be identified;

[0154] An image output module 606 is configured to input the image features of the ice image to be identified into an upsampling network in a pre-trained transmission line ice identification model, and generate an ice identification image based on the image features through the upsampling network;

[0155] The recognition module 608 is configured to determine the ice condition of the ice image to be recognized based on the ice recognition image.

[0156] Each module in the aforementioned power line icing identification model training device and power line icing identification device can be implemented in whole or in part via software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0157] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 7As shown. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, memory and input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store training data of a transmission line icing recognition model and transmission line icing recognition data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a training method for a transmission line icing recognition model and a transmission line icing recognition method are implemented.

[0158] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0159] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program. When executed by the processor, the computer program causes the processor to perform the aforementioned method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line. The method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line may be the steps of the method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line in the aforementioned embodiments.

[0160] In one embodiment, a computer-readable storage medium is provided, storing a computer program. When executed by a processor, the computer program causes the processor to perform the aforementioned method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line. The method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line may be the steps of the method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line in the aforementioned embodiments.

[0161] In one embodiment, a computer program product is provided, including a computer program. When executed by a processor, the computer program causes the processor to perform the aforementioned method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line. The method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line may be the steps of the method for training a transmission line icing recognition model and the steps of the method for identifying icing on a transmission line in the aforementioned embodiments.

[0162] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0163] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0164] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A training method for a transmission line ice recognition model, characterized in that: The method comprises: Acquire an ice-covered image sample set for a transmission line; the ice-covered image sample set includes a plurality of ice-covered image samples; Inputting the ice-covered image sample into a downsampling network in a transmission line ice-covered recognition model to be trained, extracting a multi-scale feature map of the ice-covered image sample through a multi-scale feature extraction unit of the downsampling network, extracting a semantic feature map of the ice-covered image sample through a semantic feature extraction unit of the downsampling network, and fusing the multi-scale feature map and the semantic feature map based on a gated attention mechanism through a feature fusion unit of the downsampling network to obtain image features of the ice-covered image sample; Inputting the image features of the ice-covered image samples into an upsampling network in the transmission line ice-covered recognition model to be trained, and generating an ice-covered prediction image based on the image features through the upsampling network; The transmission line icing recognition model is trained according to the actual icing information of the icing image samples and the loss information determined by the icing prediction information of the icing prediction images.

2. The method according to claim 1, characterized in that The step of obtaining a sample set of ice-covered images of a transmission line includes: By simulating the icing environment of the transmission line, an ice thickness image set obtained by a multispectral imager for ice thickness is obtained, and an ice surface image set obtained by a high-speed polarized light camera for ice surface is obtained; the ice thickness image set includes ice thickness image subsets corresponding to multiple acquisition moments; the ice thickness image subset corresponding to any of the acquisition moments includes ice thickness images of different bands acquired at the acquisition moment; the ice surface image set includes multiple ice surface images; Performing visualization processing on the images in the ice thickness image set to obtain processed images, and segmenting the images in the ice surface image set to obtain segmented images; The ice-covered image sample set is generated based on each of the processed images and each of the segmented images.

3. The method according to claim 2, characterized in that The visualizing process of the images in the ice thickness image set to obtain processed images includes: Using a principal component analysis model, principal component analysis is performed on the ice thickness images of different bands collected at any of the collection moments to obtain a principal component analysis result corresponding to any of the collection moments; Based on the principal component analysis result corresponding to any of the acquisition moments, ice thickness images of different bands in the ice thickness image subset corresponding to any of the acquisition moments are fused to obtain a fused ice thickness image corresponding to any of the acquisition moments; enhancing the contrast between the ice layer and the conductor in the fused ice thickness image corresponding to any of the acquisition moments by an adaptive histogram equalization method, thereby obtaining a contrast-enhanced ice thickness image corresponding to any of the acquisition moments; The contrast-enhanced ice thickness image corresponding to each acquisition moment is used as each processed image.

4. The method according to claim 2, characterized in that The segmenting of the images in the ice-covered surface image set to obtain segmented images includes: Dividing any of the ice-covered surface images according to a preset window size and a preset window step size to obtain a set of sub-blocks corresponding to any of the ice-covered surface images; Performing quadtree decomposition on a set of subblocks corresponding to any of the ice-covered surface images using a quadtree decomposition method, so as to determine a subblock containing ice edge features in the set of subblocks of any of the ice-covered surface images as a target subblock; Determining a segmentation method for the ice-covered surface image based on a target sub-block corresponding to any of the ice-covered surface images; According to the segmentation method of each ice-covered surface image, each ice-covered surface image is segmented to obtain each segmented image.

5. The method according to claim 1, wherein The ice cover prediction image is an ice cover area prediction image or an ice cover thickness prediction image; the ice cover area prediction image is determined based on a pixel-level segmentation mask; the pixel-level segmentation mask is used to identify whether each pixel position in the ice cover area prediction image is an ice cover area; the ice cover thickness prediction image is determined based on an ice cover thickness distribution map; the ice cover thickness distribution map is used to identify the ice cover thickness at each pixel position in the ice cover thickness prediction image.

6. A method for identifying ice coating on a transmission line, characterized in that: The method comprises: Acquire an image of ice cover to be identified on a transmission line; Inputting the ice-covered image to be identified into a downsampling network in a pre-trained transmission line ice-covered recognition model, extracting a multi-scale feature map of the ice-covered image to be identified through a multi-scale feature extraction unit of the downsampling network, extracting a semantic feature map of the ice-covered image to be identified through a semantic feature extraction unit of the downsampling network, and fusing the multi-scale feature map and the semantic feature map based on a gated attention mechanism through a feature fusion unit of the downsampling network to obtain image features of the ice-covered image to be identified; Inputting the image features of the ice-covered image to be identified into the upsampling network in the pre-trained transmission line ice-covered identification model, and generating an ice-covered identification image based on the image features through the upsampling network; Based on the ice cover recognition image, an ice cover condition of the ice cover image to be recognized is determined.

7. A training device for a transmission line ice recognition model, characterized in that: The device comprises: An acquisition module, configured to acquire an ice-covered image sample set for a transmission line; the ice-covered image sample set includes a plurality of ice-covered image samples; a feature extraction module, configured to input the ice-covered image sample into a downsampling network in a transmission line ice-covered recognition model to be trained, extract a multiscale feature map of the ice-covered image sample through a multiscale feature extraction unit of the downsampling network, extract a semantic feature map of the ice-covered image sample through a semantic feature extraction unit of the downsampling network, and fuse the multiscale feature map and the semantic feature map based on a gated attention mechanism through a feature fusion unit of the downsampling network to obtain image features of the ice-covered image sample; An image generation module, configured to input the image features of the ice-covered image sample into an upsampling network in the transmission line ice-covered recognition model to be trained, and generate an ice-covered prediction image based on the image features through the upsampling network; A training module is used to train the transmission line icing recognition model based on the actual icing information of the icing image sample and the loss information determined by the icing prediction information of the icing prediction image.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images