Methods, devices, computer equipment, readable storage media, and program products for predicting potential hazards in power transmission channels.

Abstract

This application relates to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for predicting potential hazards in power transmission channels. The method includes: acquiring optical remote sensing information and synthetic aperture radar (SAR) information collected from the environment where the power transmission channel is located; performing multiple convolutional feature extractions on the SAR information and optical remote sensing information respectively to obtain multiple radar feature information and multiple optical remote sensing feature information of different sizes; for each size, fusing the radar feature information and optical remote sensing feature information at that size to obtain fused feature information corresponding to that size; performing multi-level feature extraction on the fused feature information corresponding to each size to obtain target feature information corresponding to each size; and predicting potential hazards based on the target feature information to obtain the hazard prediction result of the power transmission channel in the environment. This method can improve the accuracy of hazard prediction.

Description

Technical Field

[0001] This application relates to the field of power technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for predicting potential hazards in power transmission channels. Background Technology

[0002] With the ever-increasing demand for electricity, the safe operation of power transmission lines has become crucial. However, transmission channels often traverse complex terrains such as forests, mountains, and farmland. External obstacles (such as fallen trees, construction sites, and dense vegetation) can pose hidden dangers to power facilities, leading to power outages or even major accidents. Therefore, monitoring and intelligent identification of potential hazards in power transmission channels has become a key research focus for the power sector.

[0003] In traditional technology, the inspection of transmission lines mainly relies on manual inspection. However, due to the complex environment of some transmission lines, especially in high-altitude or remote areas, it is difficult to carry out inspections, and therefore, the accuracy of the inspection results cannot be guaranteed. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product for predicting potential hazards in power transmission channels that can improve the accuracy of inspections, in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a method for predicting potential hazards in power transmission channels, including:

[0006] Acquire optical remote sensing information and synthetic aperture radar information about the environment where the power transmission channel is located;

[0007] Multiple convolutional feature extractions are performed on the synthetic aperture radar information and the optical remote sensing information to obtain multiple radar feature information of different sizes and multiple optical remote sensing feature information of different sizes.

[0008] For each of the aforementioned dimensions, the radar feature information and the optical remote sensing feature information at that dimension are fused to obtain the fused feature information corresponding to that dimension.

[0009] Multi-level feature extraction is performed on the fused feature information corresponding to each of the dimensions to obtain the target feature information corresponding to each of the dimensions.

[0010] Based on the target feature information, hazard prediction is performed to obtain the hazard prediction result of the power transmission channel in the environment.

[0011] In one embodiment, the synthetic aperture radar information undergoes multiple convolutional feature extractions to obtain multiple radar feature information of different sizes, including:

[0012] Perform a convolution operation on the synthetic aperture radar information to obtain convolutional radar information;

[0013] The convolutional radar information is activated to obtain the initial radar feature information corresponding to the convolutional radar information.

[0014] The initial radar feature information is subjected to multiple rounds of downsampling processing, and the processing result of each round is determined as multiple radar feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operation and activation operation; the convolution object of the first round is the initial radar feature information; the convolution object of any round other than the first round is the radar feature information obtained in the previous round.

[0015] In one embodiment, the activation operation on the convolutional radar information to obtain the initial radar feature information corresponding to the convolutional radar information includes:

[0016] Obtain the activation function consisting of hyperparameters and adaptive parameters;

[0017] The initial radar feature information corresponding to the convolutional radar information is obtained by performing an activation operation on the convolutional radar information based on the activation function.

[0018] In one embodiment, the optical remote sensing information is subjected to multiple convolutional feature extractions to obtain multiple optical remote sensing feature information of different sizes, including:

[0019] The optical remote sensing information is convolved to obtain convolutional remote sensing information;

[0020] Multi-scale analysis is performed on the convolutional remote sensing information to obtain the initial remote sensing feature information corresponding to the convolutional remote sensing information;

[0021] The initial remote sensing feature information is subjected to multiple rounds of downsampling processing, and the processing result of each round is determined as multiple optical remote sensing feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operation and multi-scale analysis; the convolution object of the first round is the initial remote sensing feature information; the convolution object of any round other than the first round is the optical remote sensing feature information obtained in the previous round.

[0022] In one embodiment, the step of performing multi-scale analysis on the convolutional remote sensing information to obtain initial remote sensing feature information corresponding to the convolutional remote sensing information includes:

[0023] Convolutional remote sensing information is performed on convolutional kernels of different sizes to obtain multiple convolutional remote sensing feature information.

[0024] The convolutional remote sensing features are combined with the optical remote sensing information to obtain the combined feature information;

[0025] The merged feature information is subjected to activation, convolution, and channel weighting operations to obtain initial remote sensing feature information.

[0026] In one embodiment, the step of performing multi-level feature extraction on the fused feature information corresponding to each of the dimensions to obtain the target feature information corresponding to each of the dimensions includes:

[0027] From the fusion feature information corresponding to each of the aforementioned sizes, determine the maximum fusion feature information corresponding to the largest size;

[0028] The maximum fused feature information is enhanced to obtain enhanced fused feature information;

[0029] Multi-level feature extraction is performed on the enhanced fusion feature information and the other fusion feature information besides the enhanced fusion feature information to obtain the target feature information corresponding to each of the dimensions.

[0030] In one embodiment, the step of enhancing the maximum fused feature information to obtain enhanced fused feature information includes:

[0031] The maximum fused feature information is subjected to global average pooling to obtain the maximum fused feature matrix;

[0032] The maximum fusion feature matrix is ​​sequentially subjected to convolution, batch normalization, activation, and secondary convolution operations to obtain the attention weights corresponding to the maximum fusion feature matrix.

[0033] The enhanced fusion feature information is obtained by calculating the maximum fusion feature information based on the attention weight.

[0034] Secondly, this application also provides a power transmission channel hazard prediction device, comprising:

[0035] The information acquisition module is used to acquire optical remote sensing information and synthetic aperture radar information collected from the environment where the power transmission channel is located;

[0036] The convolutional feature extraction module is used to perform multiple convolutional feature extractions on the synthetic aperture radar information and the optical remote sensing information respectively, to obtain multiple radar feature information of different sizes and multiple optical remote sensing feature information of different sizes.

[0037] The information fusion module is used to fuse the radar feature information and the optical remote sensing feature information for each of the dimensions to obtain the fused feature information corresponding to the dimension.

[0038] A multi-level feature extraction module is used to perform multi-level feature extraction on the fused feature information corresponding to each of the dimensions to obtain the target feature information corresponding to each of the dimensions.

[0039] The hazard prediction module is used to predict hazards based on the characteristic information of each target, and to obtain the hazard prediction result of the power transmission channel in the environment.

[0040] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0041] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described above.

[0042] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method described above.

[0043] The aforementioned method, apparatus, computer equipment, computer-readable storage medium, and computer program product for predicting potential hazards in power transmission channels acquire optical remote sensing information and synthetic aperture radar (SAR) information collected about the environment where the power transmission channel is located. This allows for the determination of the specific conditions of the environment. Subsequently, multiple convolutional feature extractions are performed on the SAR information and optical remote sensing information to obtain multiple radar feature information and multiple optical remote sensing feature information of different sizes. For each size, the radar feature information and optical remote sensing feature information at that size are fused to obtain the fused feature information corresponding to that size. Multi-level feature extraction is then performed on the fused feature information corresponding to each size to obtain the target feature information corresponding to each size. Based on the target feature information, hazard prediction is performed to obtain the hazard prediction result of the power transmission channel in the environment. This method can effectively fuse optical remote sensing information and SAR information, combining the advantages of both to achieve more comprehensive identification of potential hazards in power transmission channels, thereby improving the accuracy of hazard prediction for power transmission channels. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is an application environment diagram of the power transmission channel hidden danger prediction method in one embodiment;

[0046] Figure 2 This is a flowchart illustrating a method for predicting potential hazards in power transmission channels in one embodiment;

[0047] Figure 3 This is a flowchart illustrating the activation operation steps in one embodiment;

[0048] Figure 4 This is a flowchart illustrating the multi-scale analysis steps in one embodiment;

[0049] Figure 5 This is a flowchart illustrating the enhanced fusion steps in one embodiment;

[0050] Figure 6 This is a flowchart illustrating the steps for predicting potential hazards in a power transmission channel in one embodiment;

[0051] Figure 7 This is a flowchart illustrating a method for predicting potential hazards in power transmission channels in another embodiment;

[0052] Figure 8 This is a structural block diagram of a power transmission channel hidden danger prediction device in one embodiment;

[0053] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0055] The power transmission channel hazard prediction method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located on the cloud or other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Specifically, during the process of predicting potential hazards in the power transmission channel, server 104 acquires optical remote sensing information and synthetic aperture radar (SAR) information collected from terminal 102 regarding the environment where the power transmission channel is located. It then performs multiple convolutional feature extractions on the SAR and optical remote sensing information to obtain multiple radar feature information and multiple optical remote sensing feature information of different sizes. For each size, it fuses the radar and optical remote sensing feature information to obtain the fused feature information corresponding to that size. It then performs multi-level feature extraction on the fused feature information corresponding to each size to obtain the target feature information corresponding to each size. Based on the target feature information, it performs hazard prediction to obtain the hazard prediction result of the power transmission channel in the environment.

[0056] In one exemplary embodiment, such as Figure 2 As shown, a method for predicting potential hazards in power transmission channels is provided, and this method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S202 to S210. Wherein:

[0057] Step S202: Acquire optical remote sensing information and synthetic aperture radar information collected for the environment where the power transmission channel is located.

[0058] A power transmission channel refers to the set of all nodes and edges in a network through which current flows as it is injected from power source nodes and flows out from load nodes. These nodes and edges together constitute the path for power transmission. With the continuous growth of societal demand for electricity, the safe operation of power transmission lines has become crucial. However, power transmission channels often traverse complex terrains, such as forests, mountains, and farmland. External obstacles (such as fallen trees, construction sites, and dense vegetation) can pose hidden dangers to power facilities, leading to power outages or even major accidents. Therefore, the monitoring and intelligent identification of hidden dangers in power transmission channels has become a key research direction for the power sector.

[0059] Optical remote sensing information is acquired by using optical remote sensing equipment on space vehicles such as optical remote sensing satellites, space stations, or space shuttles, including space cameras, scanners, and imaging spectrometers, to detect ground or celestial bodies. These devices primarily use visible light, ultraviolet light, and infrared light for detection, and possess high-resolution imaging and "what you see is what you get" capabilities. Synthetic Aperture Radar (SAR) information is acquired using synthetic aperture radar technology. SAR is a high-resolution imaging radar that can obtain high-resolution radar images similar to optical photography even under extremely low visibility weather conditions. It utilizes the relative motion between the radar and the target, and through data processing methods, synthesizes a large equivalent antenna aperture to obtain high-resolution radar images.

[0060] Specifically, optical remote sensing information provides the possibility of monitoring large areas. Optical remote sensing and synthetic aperture radar (SAR) information, as two important information sources, each possess unique advantages: optical remote sensing information has high resolution and is intuitive, clearly presenting surface features and the status of power transmission lines, but it is easily affected by cloud cover and weather conditions. SAR information, acquired through radar images using different polarization methods, can penetrate clouds and vegetation, possessing all-weather operation capabilities, and is particularly advantageous in detecting structural features and the composition of target materials. Therefore, it is possible to acquire both optical remote sensing and SAR information collected on the environment where power transmission channels are located, and combine these two sources to analyze potential hazards in power transmission channels.

[0061] In step S204, multiple convolutional feature extractions are performed on the synthetic aperture radar information and optical remote sensing information to obtain multiple radar feature information of different sizes and multiple optical remote sensing feature information of different sizes.

[0062] Among them, radar feature information refers to the information after feature extraction from synthetic aperture radar information. Optical remote sensing feature information refers to the information after feature extraction from optical remote sensing information.

[0063] Specifically, convolutional feature extraction can effectively extract key features from input information. These features include local features such as edges and textures, as well as more complex patterns obtained by combining these features at deeper levels. Furthermore, convolutional feature extraction reduces model complexity through parameter sharing. In convolutional layers, convolutional kernels (filters) of different sizes move across the input information, resulting in feature information of different sizes. Small convolutional kernels capture local details (such as road edges), while large convolutional kernels are used to model large-scale structures (such as forest cover and farmland patch features). Therefore, performing multiple convolutional feature extractions on synthetic aperture radar (SAR) information and optical remote sensing information respectively, obtaining multiple radar feature information and multiple optical remote sensing feature information of different sizes, can improve the accuracy of key feature extraction and facilitate subsequent feature extraction and fusion processes of feature information of different sizes.

[0064] In some specific embodiments, the server can perform convolution operations on the synthetic aperture radar information to obtain convolutional radar information, perform activation operations on the convolutional radar information to obtain initial radar feature information corresponding to the convolutional radar information, and perform multiple rounds of downsampling processing on the initial radar feature information, determining the processing result of each round as multiple radar feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operations and activation operations; the convolution object of the first round is the initial radar feature information; the convolution object of any round other than the first round is the radar feature information obtained in the previous round.

[0065] In other specific embodiments, the server can also perform convolution operations on the optical remote sensing information to obtain convolutional remote sensing information, perform multi-scale analysis on the convolutional remote sensing information to obtain initial remote sensing feature information corresponding to the convolutional remote sensing information, and perform multiple rounds of downsampling processing on the initial remote sensing feature information, determining the processing result of each round as multiple optical remote sensing feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operations and multi-scale analysis; the convolution object of the first round is the initial remote sensing feature information; the convolution object of any round other than the first round is the optical remote sensing feature information obtained in the previous round.

[0066] Step S206: For each size, the radar feature information and optical remote sensing feature information of the size are fused to obtain the fused feature information corresponding to the size.

[0067] Among them, the fused feature information is the information obtained by fusing radar feature information and optical remote sensing feature information.

[0068] Specifically, for each size, the radar feature information and the optical remote sensing feature information at that size have the same dimension. Therefore, the radar feature information and the optical remote sensing feature information can be weighted and fused, that is, half of each value is added together to obtain the fused feature information corresponding to that size.

[0069] Step S208: Perform multi-level feature extraction on the fused feature information corresponding to each size to obtain the target feature information corresponding to each size.

[0070] Among them, target feature information refers to the feature information that is ultimately used for hazard prediction.

[0071] Specifically, after obtaining the fused feature information corresponding to each size, multi-level feature extraction can be performed on each fused feature information to obtain the target feature information corresponding to each size. This facilitates subsequent hazard prediction for power transmission channels. In this process, through multi-level sub-feature extraction, obstacles (such as fallen trees and engineering equipment) near transmission towers and power lines can be accurately identified, even if the target size is small and the background is complex. When it is necessary to monitor changes in ground features (such as landslides and changes in vegetation growth), subtle changes can also be captured. This is suitable for post-disaster inspection of power transmission lines, enabling efficient response by detecting minor anomalies such as fallen trees or damaged equipment. Optical remote sensing information is affected by external factors such as weather and seasons. Through multi-scale and non-local feature modeling, the model maintains high robustness and strong generalization ability under different data distributions.

[0072] In some specific embodiments, the server can determine the maximum fusion feature information corresponding to the largest size from the fusion feature information corresponding to each size, perform feature enhancement on the maximum fusion feature information to obtain enhanced fusion feature information, and perform multi-level feature extraction on the enhanced fusion feature information and the remaining fusion feature information to obtain the target feature information corresponding to each size. In other specific embodiments, a multi-level feature extraction model can be preset, and multi-level feature extraction can be performed on the fusion feature information corresponding to each size based on the multi-level feature extraction model to obtain the target feature information corresponding to each size.

[0073] Step S210: Based on the characteristic information of each target, perform hazard prediction to obtain the hazard prediction results of the power transmission channel in the environment.

[0074] The hazard prediction results refer to the predictions of potential hazards that may exist in the environment of the power transmission channel. For example, the hazard prediction results may include specific hazard types and locations.

[0075] Specifically, after determining the characteristic information of each target, the server can predict potential hazards based on this information, thus obtaining the predicted hazard information for the power transmission channel in the environment. For example, the characteristic information of each target is as follows: , , The server can , , The inputs are fed into the target prediction head, and the output is a tensor containing prediction information. Each line corresponds to a prediction, including bounding box coordinates, class label, and confidence score.

[0076] The Head is responsible for directly predicting the object's class, location (including bounding box coordinates), and presence confidence from the target feature information. The output of the detection head is a three-dimensional tensor, where each row contains the class probability, bounding box coordinates, and object presence confidence for each predicted box at each spatial location (relative to the target feature information).

[0077] Bounding box prediction process: Each cell is responsible for predicting the coordinates and confidence scores of several bounding boxes. The coordinates of the bounding boxes include: center point... (Position relative to the cell containing the target feature information), width and height (Scaled relative to the width and height of the entire image), these coordinates are passed through a sigmoid function to ensure that the output value is between 0 and 1, which allows for more stable training.

[0078] (2) Category and confidence prediction process: In addition to coordinate prediction, each bounding box also has a confidence prediction, which is used to indicate the probability that there is an object of a specific category in the box.

[0079] The aforementioned method for predicting potential hazards in power transmission channels acquires optical remote sensing and synthetic aperture radar (SAR) information about the environment in which the transmission channel is located, thus determining the specific conditions of the environment. Then, multiple convolutional feature extractions are performed on the SAR and optical remote sensing information to obtain multiple radar and optical remote sensing feature information of different sizes. For each size, the radar and optical remote sensing feature information are fused to obtain the corresponding fused feature information. Multi-level feature extraction is then performed on the fused feature information corresponding to each size to obtain the target feature information corresponding to each size. Based on the target feature information, hazard prediction is performed to obtain the hazard prediction result of the power transmission channel in the environment. This method effectively fuses optical remote sensing and SAR information, combining the advantages of both to achieve more comprehensive identification of potential hazards in power transmission channels, thereby improving the accuracy of hazard prediction.

[0080] In an exemplary embodiment, multiple convolutional feature extractions are performed on synthetic aperture radar (SAR) information to obtain multiple radar feature information of different sizes. This includes: performing a convolution operation on the SAR information to obtain convolutional radar information; performing an activation operation on the convolutional radar information to obtain initial radar feature information corresponding to the convolutional radar information; and performing multiple rounds of downsampling processing on the initial radar feature information, determining the result of each round as multiple radar feature information of different sizes. Each round of downsampling processing includes sequential convolution and activation operations. The convolution object in the first round is the initial radar feature information; the convolution object in any round other than the first round is the radar feature information obtained in the previous round.

[0081] Convolutional radar information refers to the information obtained after convolving synthetic aperture radar information. Initial radar feature information refers to the information obtained by activating the convolutional radar information, which still requires further processing.

[0082] Specifically, the server can generate synthetic aperture radar images. The input is fed into a complex convolution (ComplexConv), and... Using the real and imaginary data as input, the real and imaginary parts are processed separately through complex convolution to obtain synthetic aperture radar information. ;right The input is fed into a Conv2d_BN_SiLU module with a kernel size of 3×3 and a stride of 4. That is, a Conv2d layer, a BN layer, and a SiLU layer with a kernel size of 3×3 and a stride of 4 are concatenated to obtain intermediate convolutional radar information. Next, the SARFB module is used to... To process this, specifically, first, input the information. Perform a Conv2d (convolution) operation with a kernel size of 3×3 and a stride of 1 to obtain convolutional radar information. ,right Perform activation operations to obtain the initial radar feature information corresponding to the convolutional radar information. For example, such as Figure 3 As shown, the above activation operation includes: Input pairs Activate using the Cselu activation function to obtain the first convolutional radar information. ;right The information is input into the nonlocal module to obtain the second convolutional radar information. ;right The input is fed into the SE attention mechanism to obtain the initial radar feature information (Output). .

[0083] Then, the initial radar characteristic information Multiple rounds of downsampling processing are performed, that is, multiple processing steps are performed using the Conv2d module and the SARFB module to obtain candidate radar feature information. , , , , , .in, , This refers to multiple radar feature information of different sizes. Specifically, the initial radar feature information is processed using the Conv2d module. Perform a convolution operation to obtain Through the SARFB module Perform the activation operation to obtain radar characteristic information. The subsequent operations follow the same process as described above, and will not be repeated here.

[0084] In this embodiment, features are extracted layer by layer by stacking convolutional modules. In particular, for the processing of synthetic aperture radar information, this structured stacking can significantly improve the accuracy of information feature extraction.

[0085] In an exemplary embodiment, an activation operation is performed on the convolutional radar information to obtain initial radar feature information corresponding to the convolutional radar information, including: obtaining an activation function composed of hyperparameters and adaptive parameters; and performing an activation operation on the convolutional radar information based on the activation function to obtain initial radar feature information corresponding to the convolutional radar information.

[0086] Hyperparameters are parameters manually set before the machine learning algorithm runs, used to control the model's behavior and performance. These parameters are not automatically learned from training data but must be set by the user based on experience and domain knowledge. Adaptive parameter technology is a method for optimizing and adjusting system parameters. It ensures optimal system performance by automatically adjusting system parameters under different operating conditions and environments. The activation function, in this application, refers to the Cselu activation function, and its calculation formula is as follows:

[0087]

[0088] in, , These are hyperparameters, usually positive values. , For adaptive parameters, where Adjust according to training results , Size, For input, This is the output result.

[0089] In this embodiment, the Cselu activation function captures details and important information at different scales. The Cselu activation function not only avoids gradient vanishing but also ensures greater model robustness through dynamic adjustment of adaptive parameters m and n. The Cselu activation function can extract multi-level responses, helping the model identify potential hazards.

[0090] In an exemplary embodiment, multiple convolutional feature extractions are performed on optical remote sensing information to obtain multiple optical remote sensing feature information of different sizes. This includes: performing a convolution operation on the optical remote sensing information to obtain convolutional remote sensing information; performing multi-scale analysis on the convolutional remote sensing information to obtain initial remote sensing feature information corresponding to the convolutional remote sensing information; and performing multiple rounds of downsampling processing on the initial remote sensing feature information, determining the processing result of each round as multiple optical remote sensing feature information of different sizes. Each round of downsampling processing includes sequential convolution operations and multi-scale analysis. The convolution object in the first round is the initial remote sensing feature information. The convolution object in any round other than the first round is the optical remote sensing feature information obtained in the previous round.

[0091] Convolutional remote sensing information refers to the information obtained by performing a convolution operation on optical remote sensing information. Initial remote sensing feature information refers to the information obtained by performing activation operations on the convolutional remote sensing information, which still requires further operations.

[0092] Specifically, the server can transmit optical remote sensing information. The input is fed into the Conv2d module for convolution to obtain intermediate convolutional remote sensing information. Then The input is fed into a Conv2d_BN_SiLU module with a kernel size of 3×3 and a stride of 2. That is, a Conv2d layer, a BN layer, and a SiLU layer with a kernel size of 3×3 and a stride of 2 are concatenated to obtain convolutional remote sensing information. Next, the convolutional remote sensing information is processed through the OPTICSFB module. Multi-scale analysis was performed to obtain initial remote sensing feature information. .

[0093] Subsequently, the initial remote sensing feature information Multiple rounds of downsampling processing are performed, that is, multiple processing steps are performed using the Conv2d module and the OPTICSFB module to obtain candidate remote sensing feature information. , , , , , , .in, , , This refers to multiple remote sensing feature information of different sizes. Specifically, the initial radar feature information is processed using the Conv2d module. Perform a convolution operation to obtain Through the SARFB module Perform the activation operation to obtain radar characteristic information. The subsequent operations follow the same process as described above, and will not be repeated here.

[0094] In this embodiment, in optical remote sensing, the textures and geometric features of different land features have different scales. The OPTICSFB module extracts multi-scale features in parallel using 3×3, 5×5, and 7×7 convolutional kernels. Small convolutional kernels capture local details (such as road edges), while large convolutional kernels are used to model large-scale structures (such as forest cover and farmland block features). Spectral information and spatial structure are often complementary. The OPTICSFB module introduces a fusion strategy to jointly model spectral dimensions and spatial features. 1×1 convolutions are used to reduce spectral dimensions, thereby reducing computational overhead and avoiding information redundancy. Detailed edges in optical remote sensing images (such as building outlines or power tower structures) are crucial. OPTICSFB enhances the model's sensitivity to edge information by adding edge detection convolutional layers.

[0095] In optical remote sensing information, OPTICSFB can accurately identify obstacles (such as fallen trees and engineering equipment) near power transmission towers and power lines, even if the target is small and the background is complex. When it is necessary to monitor changes in ground features (such as landslides and changes in vegetation growth), the OPTICSFB module can capture subtle changes. It is suitable for post-disaster power line inspections, enabling efficient response by detecting minor anomalies such as fallen trees or damaged equipment. Optical remote sensing information is affected by external factors such as weather and seasons. The OPTICSFB module uses multi-scale and non-local feature modeling to ensure that the model maintains high robustness and strong generalization ability under different data distributions.

[0096] In an exemplary embodiment, multi-scale analysis is performed on convolutional remote sensing information to obtain initial remote sensing feature information corresponding to the convolutional remote sensing information, including: performing convolution operations on the convolutional remote sensing information based on convolutional kernels of different sizes to obtain multiple convolutional remote sensing feature information; merging each convolutional remote sensing feature information with optical remote sensing information to obtain merged feature information; and performing activation operations, convolution operations, and channel weighting operations on the merged feature information to obtain initial remote sensing feature information.

[0097] Among them, convolutional remote sensing feature information refers to the feature information obtained by performing convolution operations on convolutional remote sensing information based on convolutional kernels of different sizes. Merged feature information is the information obtained by merging each convolutional remote sensing feature information with optical remote sensing information.

[0098] Specifically, such as Figure 4 As shown, in the OPTICSFB module, firstly... The inputs were fed into three different kernel sizes: a depthwise separable convolutional (DWConv) layer with kernel sizes of 5×5 and 3×3, and a Conv2d layer with a kernel size of 1×1. Using different kernel sizes allows for the capture of features at different scales. The 5×5 kernel extracts a wider range of spatial information, while the 3×3 kernel focuses more on capturing details within a medium range, resulting in multiple convolutional remote sensing feature information. , , Then , , , Perform the Add operation (merge operation) to obtain the merged feature information. By integrating original features with deeply extracted features, the expressive power of optical feature maps of potential hazards in power transmission channels is enhanced. Then... The input is fed into the Cselu activation layer for activation, yielding the first remote sensing feature information. Then The input is fed into a Conv2d layer with a kernel size of 1×1 for convolution operation, resulting in richer and more refined second remote sensing feature information. Then The input is fed into the SE attention mechanism for channel weighting to obtain the initial remote sensing feature information (Output). .

[0099] In this embodiment, an OPTICSFB module is designed, which includes a multi-scale convolution submodule, a spectral-spatial feature fusion submodule, an attention mechanism submodule, and an edge perception submodule, to address the diversity, complexity, and interference issues in optical remote sensing data.

[0100] In an exemplary embodiment, multi-level feature extraction is performed on the fusion feature information corresponding to each size to obtain target feature information corresponding to each size, including: determining the maximum fusion feature information corresponding to the largest size from the fusion feature information corresponding to each size; performing feature enhancement on the maximum fusion feature information to obtain enhanced fusion feature information; and performing multi-level feature extraction on the enhanced fusion feature information and the remaining fusion feature information other than the enhanced fusion feature information to obtain target feature information corresponding to each size.

[0101] Among them, enhanced fusion feature information refers to the feature information after feature enhancement of the maximum fusion feature information.

[0102] Specifically, from the fusion feature information corresponding to each size, the maximum fusion feature information corresponding to the largest size is determined. ,Will The input is fed into the LYAttention module for feature enhancement, resulting in enhanced fused feature information. ;Will The input is fed into the SPPF module to obtain the first fused feature information. ;right Upsampling is performed to obtain the second fused feature information. ;Will and Perform a Concat operation to obtain the third fused feature information. ;Will The input is fed into the C3K2 module to obtain the fourth fused feature information. ;right Perform an upsampling operation to obtain the fifth fused feature information. ;Will and Perform the Concat operation to obtain the sixth fused feature information. ;Will The input is fed into the C3K2 module to obtain the seventh fused feature information. ;Will The input is fed into the Conv2d module (3×3 kernel size, stride 2) to obtain the eighth fused feature information. ;Will and Perform the Concat operation to obtain the ninth fused feature information. ;Will The input is fed into the C3K2 module to obtain the tenth fusion feature information. ;Will The input is fed into the Conv2d module (3×3 kernel size, stride 2) to obtain the eleventh fused feature information. ;Will and Perform the Concat operation to obtain the twelfth fusion feature information. ;Will The input is fed into the C3K2 module to obtain the thirteenth fusion feature information. .

[0103] In this embodiment, a LYAttention module is designed, such as... Figure 5 As shown, the maximum fused feature information of the input is first... Global average pooling is performed, followed by processing through a Conv2d layer, a BN layer, a ReLU activation function, and another Conv2d layer. This is then processed by a Cselu activation function to obtain the attention weights, which are then... Element-wise multiplication outputs new enhanced fusion feature information. .

[0104] For the maximum fused feature information of the input ,in It is the number of channels for integrating feature information. and These represent the height and width of the feature map corresponding to the fused feature information, respectively. First, for Perform global average pooling, retaining each channel and height during the pooling process, to obtain a matrix. ,

[0105]

[0106] in, Representation matrix The value of the element in the c-th row and h-th column;

[0107] Then, it is processed by a Conv2d layer, a BN layer, a ReLU activation function (Rectified Linear Unit), and another Conv2d layer to capture cross-channel interactions, i.e.;

[0108]

[0109] in It is a Cselu function, and BN represents batch normalization. This represents the attention weights calculated at different channels and z-positions. The Conv2d layer calculates attention weights by aggregating the z-positions of adjacent channels, and stacking two Conv2d layers enhances the module's learning ability. The attention weights are then... Applied to By multiplying element by element, new enhanced fusion feature information is obtained. .

[0110] In one embodiment, such as Figure 6 As shown, a method for predicting potential hazards in power transmission channels in practical application scenarios is also provided, including:

[0111] Step 1: Construct the original remote sensing image dataset D1 of potential hazards in power transmission channels, or use an existing publicly available original remote sensing image dataset of potential hazards in power transmission channels. Organize the acquired information to ensure that the optical remote sensing information and synthetic aperture radar information can be "one-to-one correspondence" (i.e., the acquisition time is the same and the scanned area is the same, which is called a set of data. Both optical remote sensing information and synthetic aperture radar information include image data). Manually annotate the acquired image data, and use the annotation tool to select the location of the hazard (i.e., determine the horizontal and vertical coordinates of the upper left and lower right corners of the box) and label it.

[0112] Example: Polarimetric SAR data can be acquired via Sentinel-1 satellite; optical remote sensing data can be acquired via Sentinel-2 satellite.

[0113] Step 2: Perform preprocessing on D1, including but not limited to cropping and data augmentation, to obtain the processed remote sensing image dataset D2 of power transmission channel hazards. Then, divide the remote sensing image dataset of power transmission channel hazards into sub-databases to train the intelligent identification model of power transmission channel hazards that fuses polarimetric SAR and optical remote sensing data.

[0114] D2 contains multiple sets of power transmission channel hazard data, which includes a set of polarimetric SAR data and optical remote sensing data.

[0115] Note: In this invention, the original remote sensing image dataset of power transmission channel hazards was first cropped to the same size. Since the original dataset of power transmission channel hazards was small, in order to ensure that the dataset had enough samples for training, validation and testing, and to enhance the robustness of the intelligent identification model of power transmission channel hazards and reduce the sensitivity of the model to images, the number of samples was increased through data augmentation, so as to better train the deep learning model. The resulting data-augmented remote sensing image dataset of power transmission channel hazards was divided into training set, validation set and test set in a ratio of 8:1:1.

[0116] Step 3: Innovation: A smart identification model for potential hazards in power transmission channels was constructed, integrating polarimetric SAR and optical remote sensing data. This model is used to identify potential hazards in power transmission channels. The model inputs hazard data and outputs the hazard category and location.

[0117] Intelligent identification model for potential hazards in power transmission channels, such as Figure 6 As shown.

[0118] Step 3.1: Combine synthetic aperture radar images The input is fed into a complex convolution (ComplexConv), and... Using the real and imaginary data as input, the real and imaginary parts are processed separately through complex convolution to obtain synthetic aperture radar information. ;right The input is fed into a Conv2d_BN_SiLU module with a kernel size of 3×3 and a stride of 4. That is, a Conv2d layer, a BN layer, and a SiLU layer with a kernel size of 3×3 and a stride of 4 are concatenated to obtain intermediate convolutional radar information. Next, the SARFB module is used to... To process it, specifically, first input... Perform a Conv2d (convolution) operation with a kernel size of 3×3 and a stride of 1 to obtain convolutional radar information. ,right Perform activation operations to obtain the initial radar feature information corresponding to the convolutional radar information. For example, such as Figure 3 As shown, the above activation operation includes: […]. Activate using the Cselu activation function to obtain the first convolutional radar information. ;right The information is input into the nonlocal module to obtain the second convolutional radar information. ;right The initial radar feature information is obtained by inputting it into the SE attention mechanism. .

[0119] Then, the initial radar characteristic information Multiple rounds of downsampling processing are performed, that is, multiple processing steps are performed using the Conv2d module and the SARFB module to obtain candidate radar feature information. , , , , , , .in, , This refers to multiple radar feature information of different sizes. Specifically, the initial radar feature information is processed using the Conv2d module. Perform a convolution operation to obtain Through the SARFB module Perform the activation operation to obtain radar characteristic information. The subsequent operations follow the same process as described above, and will not be repeated here.

[0120] Note: When inputting SAR data (synthetic aperture radar image) into ComplexConv, it is usually input in complex form. This is because SAR data carries amplitude and phase information, which is suitable for expression using complex numbers.

[0121] Innovation: This invention designs a SARFB (SAR data processing) module. The high noise and complex scattering characteristics of SAR data increase the difficulty of feature extraction. This module is specifically optimized for feature extraction from polarimetric SAR images. This module can deeply mine multi-scale information from SAR data and enhance the representation of key regions through non-local feature capture and SE attention mechanisms.

[0122] Note: The high noise (spot noise) and complex scattering characteristics of SAR data (including the reflection of signals by the geometry, azimuth, and electromagnetic properties of ground objects) increase the difficulty of feature extraction.

[0123] Power transmission lines often traverse diverse terrains, such as mountains, forests, and farmland. Polarimetric SAR data in these complex scenarios can be affected by terrain effects and multiple scattering. The SARFB module effectively filters noise and highlights key features through nonlocal feature capture and SE attention mechanisms. In power transmission line hazard identification, the model needs to accurately identify obstacles (such as trees and foreign objects) and anomalies in the tower structure beneath the transmission lines. The SARFB module enhances the model's ability to identify small targets and avoids misjudgments caused by image scale variations through multi-scale convolution and the Cselu activation function. The SARFB module's nonlocal feature capture and adaptive parameters enable the model to maintain high generalization performance on diverse SAR data, reducing identification failures due to differences in data distribution. This design results in higher accuracy, robustness, and stability in practical applications.

[0124] In the SARFB module, first input Perform a Conv2d (convolution) operation with a kernel size of 3×3 and a stride of 1 to obtain convolutional radar information. ,right Perform activation operations to obtain the initial radar feature information corresponding to the convolutional radar information. For example, such as Figure 3 As shown, the above activation operation includes: […]. Activate using the Cselu activation function to obtain the first convolutional radar information. ;right The information is input into the nonlocal module to obtain the second convolutional radar information. ;right The initial radar feature information is obtained by inputting it into the SE attention mechanism. .

[0125] Innovation: The formula for calculating the Cselu activation function is as follows:

[0126]

[0127] in, , These are hyperparameters, usually positive values. , For adaptive parameters, where Adjust according to training results , Size, For input, This is the output result.

[0128] In this embodiment, the Cselu activation function captures details and important information at different scales. The Cselu activation function not only avoids gradient vanishing but also ensures greater model robustness through dynamic adjustment of adaptive parameters m and n. The Cselu activation function can extract multi-level responses, helping the model identify potential hazards.

[0129] Step 3.2: Transfer optical remote sensing information The input is fed into the Conv2d module for convolution to obtain intermediate convolutional remote sensing information. Then The input is fed into a Conv2d_BN_SiLU module with a kernel size of 3×3 and a stride of 2. That is, a Conv2d layer, a BN layer, and a SiLU layer with a kernel size of 3×3 and a stride of 2 are concatenated to obtain convolutional remote sensing information. Next, the convolutional remote sensing information is processed through the OPTICSFB module. Multi-scale analysis was performed to obtain initial remote sensing feature information. .

[0130] Subsequently, the initial remote sensing feature information Multiple rounds of downsampling processing are performed, that is, multiple processing steps are performed using the Conv2d module and the OPTICSFB module to obtain candidate remote sensing feature information. , , , , , , .in, , , This refers to multiple remote sensing feature information of different sizes. Specifically, the initial radar feature information is processed using the Conv2d module. Perform a convolution operation to obtain Through the SARFB module Perform the activation operation to obtain radar characteristic information. The subsequent operations follow the same process as described above, and will not be repeated here.

[0131] In this embodiment, in optical remote sensing, the textures and geometric features of different land features have different scales. The OPTICSFB (Multi-Scale Data Processing) module extracts multi-scale features in parallel using 3×3, 5×5, and 7×7 convolutional kernels. Small convolutional kernels capture local details (such as road edges), while large convolutional kernels are used to model large-scale structures (such as forest cover and farmland block features). Spectral information and spatial structure are often complementary. The OPTICSFB module introduces a fusion strategy to jointly model spectral dimensions and spatial features. 1×1 convolutions are used to reduce spectral dimensions, thereby reducing computational overhead and avoiding information redundancy. Detailed edges in optical remote sensing images (such as building outlines or power tower structures) are crucial. OPTICSFB enhances the model's sensitivity to edge information by adding edge detection convolutional layers.

[0132] In optical remote sensing information, OPTICSFB can accurately identify obstacles (such as fallen trees and engineering equipment) near power transmission towers and power lines, even if the target is small and the background is complex. When it is necessary to monitor changes in ground features (such as landslides and changes in vegetation growth), the OPTICSFB module can capture subtle changes. It is suitable for post-disaster power line inspections, enabling efficient response by detecting minor anomalies such as fallen trees or damaged equipment. Optical remote sensing information is affected by external factors such as weather and seasons. The OPTICSFB module uses multi-scale and non-local feature modeling to ensure that the model maintains high robustness and strong generalization ability under different data distributions.

[0133] In the OPTICSFB module, firstly... The data were fed into three different convolutional kernel sizes: 5×5 and 3×3 depthwise separable convolutional (DWConv) layers, and a 1×1 Conv2d layer. Using different kernel sizes allows for the capture of features at different scales. The 5×5 kernel extracts a wider range of spatial information, while the 3×3 kernel focuses more on capturing details within a medium range, resulting in multiple convolutional remote sensing feature information. , , Then , , , Perform the Add operation (merge operation) to obtain the merged feature information. By integrating original features with deeply extracted features, the expressive power of optical feature maps of potential hazards in power transmission channels is enhanced. Then... The input is fed into the Cselu activation layer for activation, yielding the first remote sensing feature information. Then The input is fed into a Conv2d layer with a kernel size of 1×1 for convolution operation, resulting in richer and more refined second remote sensing feature information. Then The input is fed into the SE attention mechanism for channel weighting to obtain initial remote sensing feature information. .

[0134] Step 3.3: For and Weighted fusion is performed where the two feature maps have the same dimension, and half of each value is added together to obtain the fused feature information. ;right and Weighted fusion is performed to obtain fused feature information. ;right and Weighted fusion is performed to obtain fused feature information. ;

[0135] From the fusion feature information corresponding to each size, determine the maximum fusion feature information corresponding to the largest size. ,Will The input is fed into the LYAttention (feature enhancement module) for feature enhancement, resulting in enhanced and fused feature information. ;Will The input is fed into the SPPF module to obtain the first fused feature information. ;right Upsampling is performed to obtain the second fused feature information. ;Will and Perform a Concat operation to obtain the third fused feature information. ;Will The input is fed into the C3K2 module to obtain the fourth fused feature information. ;right Perform an upsampling operation to obtain the fifth fused feature information. ;Will and Perform the Concat operation to obtain the sixth fused feature information. ;Will The input is fed into the C3K2 module to obtain the seventh fused feature information. ;Will The input is fed into the Conv2d module (3×3 kernel size, stride 2) to obtain the eighth fused feature information. ;Will and Perform the Concat operation to obtain the ninth fused feature information. ;Will The input is fed into the C3K2 module to obtain the tenth fusion feature information. ;Will The input is fed into the Conv2d module (3×3 kernel size, stride 2) to obtain the eleventh fused feature information. ;Will and Perform the Concat operation to obtain the twelfth fusion feature information. ;Will The input is fed into the C3K2 (feature extraction module) to obtain the thirteenth fused feature information. .

[0136] Innovation: A LYAttention module was designed, such as... Figure 5 As shown, the maximum fused feature information of the input is first... Global average pooling is performed, followed by processing through a Conv2d layer, a BN layer, a ReLU activation function, and another Conv2d layer. This is then processed by a Cselu activation function to obtain the attention weights, which are then... Element-wise multiplication outputs new enhanced fusion feature information. .

[0137] For the maximum fused feature information of the input ,in It is the number of channels for integrating feature information. and These represent the height and width of the feature map corresponding to the fused feature information, respectively. First, for Perform global average pooling, retaining each channel and height during the pooling process, to obtain a matrix. ,

[0138]

[0139] in, Representation matrix The value of the element in the c-th row and h-th column;

[0140] Then, it is processed by a Conv2d layer, a BN layer, a ReLU activation function, and another Conv2d layer to capture cross-channel interactions, i.e.;

[0141]

[0142] in It is a Cselu function, and BN represents batch normalization. This represents the attention weights calculated at different channels and z-positions. The Conv2d layer calculates attention weights by aggregating the z-positions of adjacent channels, and stacking two Conv2d layers enhances the module's learning ability. The attention weights are then... Applied to By multiplying element by element, new enhanced fusion feature information is obtained. .

[0143] Step 3.4: The feature information of each target is as follows , , The server can , , The inputs are fed into the target prediction head, and the output is a tensor containing prediction information. Each line corresponds to a prediction, including bounding box coordinates, class label, and confidence score.

[0144] The Head is responsible for directly predicting the object's class, location (including bounding box coordinates), and presence confidence from the target feature information. The output of the detection head is a three-dimensional tensor, where each row contains the class probability, bounding box coordinates, and object presence confidence for each predicted box at each spatial location (relative to the target feature information).

[0145] Bounding box prediction process: Each cell is responsible for predicting the coordinates and confidence scores of several bounding boxes. The coordinates of the bounding boxes include: center point... (Position relative to the cell containing the target feature information), width and height (Scaled relative to the width and height of the entire image), these coordinates are passed through a sigmoid function to ensure that the output value is between 0 and 1, which allows for more stable training.

[0146] (2) Category and confidence prediction process: In addition to coordinate prediction, each bounding box also has a confidence prediction, which is used to indicate the probability that there is an object of a specific category in the box.

[0147] Example:

[0148] Taking the case where optical remote sensing information and synthetic aperture radar information include images as an example, let the size of the feature map corresponding to the optical remote sensing information and synthetic aperture radar information be H×W×C, where H is the height of the feature map, W is the width of the feature map, and C is the number of channels of the feature map.

[0149] The image size of the synthetic aperture radar information is 512×512×2. The data is input into the intelligent identification model for potential hazards in power transmission channels, and firstly... After passing through the ComplexConv layer, synthetic aperture radar information with 64 channels is obtained. The size is 512×512×64; for The input is fed into the Conv2d_BN_SiLU module to obtain intermediate convolutional radar information with 128 channels. The size is 128×128×128; The input is fed into the SARFB module to obtain initial radar feature information with 256 channels. The size is 128×128×256; by stacking the Conv2d module and the SARFB module three times in sequence, candidate radar feature information with a size of 64×64×512 can be obtained respectively. Candidate radar feature information with a size of 64×64×512 Candidate radar feature information with a size of 32×32×512 Candidate radar feature information with a size of 32×32×512 Candidate radar feature information with a size of 16×16×512 Candidate radar feature information with a size of 16×16×512 ;

[0150] The image size of the optical remote sensing information is 512×512×2. The data is input into the intelligent identification model for potential hazards in power transmission channels. First, it passes through a Conv2d layer with a kernel size of 3×3 and a stride of 2 to obtain intermediate convolutional remote sensing information with 64 channels. The size is 256×256×64; The input is fed into a Conv2d_BN_SiLU module with a kernel size of 3×3 and a stride of 2, resulting in convolutional remote sensing information with 128 channels. The size is 128×128×128; The input is fed into the OPTICSFB module to obtain initial remote sensing feature information with 256 channels. The size is 128×128×256; by stacking the Conv2d module and the OPTICSFB module three times in sequence, candidate remote sensing feature information with a size of 64×64×512 can be obtained. Candidate remote sensing feature information with a size of 64×64×512 Candidate remote sensing feature information with a size of 32×32×512 Candidate remote sensing feature information with a size of 32×32×512 Candidate remote sensing feature information with a size of 16×16×512 Candidate remote sensing feature information with a size of 16×16×512 ;

[0151] right and Weighted fusion is performed on the two feature maps, which have the same dimension. Half of the values ​​from each map are added together to obtain the fused feature information. The size is 64×64×512; for and Weighted fusion is performed to obtain fused feature information. The size is 32×32×512; for and Weighted fusion is performed to obtain fused feature information. The size is 16×16×512;

[0152] Will After being input into the LYAttention module, enhanced fusion feature information of size 16×16×512 is obtained. Then The input is fed into the SPPF module, resulting in the first fused feature information with a size of 16×16×512. ;right Perform an upsampling operation to obtain a second fused feature information of size 32×32×512. ;Will and Performing the Concat operation yields a third fused feature information of size 32×32×1024. ;Will The input is fed into the C3K2 module, resulting in a fourth fused feature information of size 32×32×512. ;right Performing an upsample operation yields a fifth fused feature information with a size of 64×64×512. ;Will and Performing the Concat operation yields a sixth fused feature information of size 64×64×1024. ;Will The input is fed into the C3K2 module, resulting in the seventh fusion feature information with a size of 64×64×256. ;Will The input is fed into the Conv2d module, resulting in the eighth fused feature information with a size of 32×32×256. ;Will and Performing the Concat operation yields the ninth fused feature information with a size of 32×32×768. ;Will The input is fed into the C3K2 module, resulting in the tenth fusion feature information with a size of 32×32×512. ;Will The input is fed into the Conv2d module, resulting in the eleventh fused feature information with a size of 16×16×512. ;Will and Performing the Concat operation yields the twelfth fused feature information with a size of 16×16×1024. ;Will The input is fed into the C3K2 module, resulting in the thirteenth fused feature information with a size of 16×16×512. ;Will , , The inputs are fed into the target prediction head, and the output is a tensor containing prediction information. Each line corresponds to a prediction, including bounding box coordinates, class label, and confidence score.

[0153] Step 4: Use the power transmission channel hazard dataset to train the above model to obtain the trained power transmission channel hazard intelligent identification model.

[0154] The parameters of each layer are trained and updated on the intelligent identification model for hidden dangers in power transmission channels. First, all neural network parameters are initialized, and hyperparameters related to the intelligent identification model for hidden dangers in power transmission channels are set, including but not limited to the number of training epochs, batch size, selection of optimizer, and learning rate.

[0155] After initializing the parameters, the training and validation sets are divided into multiple batches. Each batch of training data is input into the intelligent identification model for power transmission channel hazards for training, resulting in a training loss value (loss) for that batch. After one round of training on all batches of data in the entire training set, the validation set is input into the intelligent identification model for power transmission channel hazards according to batches, resulting in a corresponding batch loss value (batch_loss). The validation set loss value is mainly used to monitor whether the intelligent identification model for power transmission channel hazards is overfitting and to adjust the training strategy, such as terminating training early or adjusting the learning rate. During training and validation, the intelligent identification model for power transmission channel hazards will automatically learn and adjust its parameters based on each loss and batch_loss. The training process ends when the batch_loss value converges after one or more rounds.

[0156] Step 5: After the model training is completed, the trained intelligent identification model for potential hazards in power transmission channels is applied to identify and analyze the remote sensing data of the current power transmission channels. The final output includes the specific hazard type and location, providing inspection personnel with accurate early warning information and subsequent handling suggestions.

[0157] In a specific embodiment, such as Figure 7 As shown, a method for predicting potential hazards in power transmission channels is also provided, including:

[0158] Step S701: Obtain optical remote sensing information and synthetic aperture radar information collected for the environment where the power transmission channel is located; perform convolution operation on the synthetic aperture radar information to obtain convolution radar information.

[0159] Step S702: Obtain the activation function composed of hyperparameters and adaptive parameters, and perform activation operation on the convolutional radar information based on the activation function to obtain the initial radar feature information corresponding to the convolutional radar information.

[0160] Step S703: Perform multiple rounds of downsampling processing on the initial radar feature information, and determine the processing result of each round as multiple radar feature information of different sizes;

[0161] Each round of downsampling processing includes sequential convolution and activation operations; the convolution object in the first round is the initial radar feature information; the convolution object in any round other than the first round is the radar feature information obtained in the previous round.

[0162] Step S704: Perform convolution operation on the optical remote sensing information to obtain convolutional remote sensing information. Perform convolution operation on the convolutional remote sensing information based on convolution kernels of different sizes to obtain multiple convolutional remote sensing feature information.

[0163] Step S705: Merge each convolutional remote sensing feature information with the optical remote sensing information to obtain merged feature information;

[0164] Step S706: Perform activation, convolution, and channel weighting operations on the merged feature information to obtain initial remote sensing feature information;

[0165] Step S707: Perform multiple rounds of downsampling processing on the initial remote sensing feature information, and determine the processing result of each round as multiple optical remote sensing feature information of different sizes;

[0166] Each round of downsampling processing includes sequential convolution operations and multi-scale analysis; the convolution object in the first round is the initial remote sensing feature information; the convolution object in any round other than the first round is the optical remote sensing feature information obtained in the previous round.

[0167] Step S708: For each size, the radar feature information and optical remote sensing feature information under the size are fused to obtain the fused feature information corresponding to the size;

[0168] Step S709: From the fusion feature information corresponding to each size, determine the maximum fusion feature information corresponding to the largest size, and perform feature enhancement on the maximum fusion feature information to obtain enhanced fusion feature information;

[0169] Step S710: Perform multi-level feature extraction on the enhanced fusion feature information and the remaining fusion feature information other than the enhanced fusion feature information to obtain the target feature information corresponding to each size.

[0170] Step S711: Based on the characteristic information of each target, perform hazard prediction to obtain the hazard prediction results of the power transmission channel in the environment.

[0171] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0172] Based on the same inventive concept, this application also provides a transmission channel hazard prediction device for implementing the above-mentioned transmission channel hazard prediction method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more transmission channel hazard prediction device embodiments provided below can be found in the limitations of the transmission channel hazard prediction method above, and will not be repeated here.

[0173] In one exemplary embodiment, such as Figure 8 As shown, a power transmission channel hazard prediction device 800 is provided, including: an information acquisition module 802, a convolutional feature extraction module 804, an information fusion module 806, a multi-level feature extraction module 808, and a hazard prediction module 810, wherein:

[0174] The information acquisition module 802 is used to acquire optical remote sensing information and synthetic aperture radar information collected from the environment where the power transmission channel is located;

[0175] The convolutional feature extraction module 804 is used to perform multiple convolutional feature extractions on synthetic aperture radar information and optical remote sensing information respectively, to obtain multiple radar feature information of different sizes and multiple optical remote sensing feature information of different sizes.

[0176] The information fusion module 806 is used to fuse radar feature information and optical remote sensing feature information for each size to obtain fused feature information corresponding to the size.

[0177] The multi-level feature extraction module 808 is used to perform multi-level feature extraction on the fused feature information corresponding to each size to obtain the target feature information corresponding to each size.

[0178] The hazard prediction module 810 is used to predict hazards based on the characteristic information of each target, and obtain the hazard prediction results of the power transmission channel in the environment.

[0179] In one exemplary embodiment, the convolutional feature extraction module 804 includes:

[0180] The first convolution operation unit is used to perform convolution operations on synthetic aperture radar information to obtain convolution radar information.

[0181] The activation operation unit is used to activate the convolutional radar information to obtain the initial radar feature information corresponding to the convolutional radar information.

[0182] The first downsampling processing unit is used to perform multiple rounds of downsampling processing on the initial radar feature information, and to determine the processing result of each round as multiple radar feature information of different sizes. The downsampling process of each round includes sequential convolution and activation operations. The convolution object of the first round is the initial radar feature information. The convolution object of any round other than the first round is the radar feature information obtained in the previous round.

[0183] In an exemplary embodiment, the activation operation unit is specifically used for:

[0184] Obtain the activation function consisting of hyperparameters and adaptive parameters;

[0185] The initial radar feature information corresponding to the convolutional radar information is obtained by activating the convolutional radar information using an activation function.

[0186] In one exemplary embodiment, the convolutional feature extraction module 804 includes:

[0187] The second convolution operation unit is used to perform convolution operations on optical remote sensing information to obtain convolutional remote sensing information.

[0188] The multi-scale analysis unit is used to perform multi-scale analysis on convolutional remote sensing information to obtain the initial remote sensing feature information corresponding to the convolutional remote sensing information.

[0189] The second downsampling processing unit is used to perform multiple rounds of downsampling processing on the initial remote sensing feature information, and to determine the processing result of each round as multiple optical remote sensing feature information of different sizes. The downsampling process of each round includes sequential convolution operations and multi-scale analysis. The convolution object of the first round is the initial remote sensing feature information. The convolution object of any round other than the first round is the optical remote sensing feature information obtained in the previous round.

[0190] In one exemplary embodiment, the multi-scale analysis unit is specifically used for:

[0191] Convolutional remote sensing information is performed using convolutional kernels of different sizes to obtain multiple convolutional remote sensing feature information.

[0192] The convolutional remote sensing features are combined with the optical remote sensing information to obtain the combined feature information;

[0193] The merged feature information is subjected to activation, convolution, and channel weighting operations to obtain the initial remote sensing feature information.

[0194] In an exemplary embodiment, the multi-level feature extraction module 808 is specifically used for:

[0195] From the fusion feature information corresponding to each size, determine the maximum fusion feature information corresponding to the largest size;

[0196] Feature enhancement is performed on the maximum fused feature information to obtain enhanced fused feature information;

[0197] Multi-level feature extraction is performed on the enhanced fusion feature information and the remaining fusion feature information other than the enhanced fusion feature information to obtain the target feature information corresponding to each size.

[0198] Each module in the aforementioned power transmission channel hazard prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0199] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for predicting potential hazards in power transmission channels. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0200] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0201] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0202] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0203] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0204] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0205] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory 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 many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0206] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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 application.

[0207] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting hidden dangers in power transmission channels, characterized in that, The method includes: Acquire optical remote sensing information and synthetic aperture radar information about the environment where the power transmission channel is located; Perform a convolution operation on the synthetic aperture radar information to obtain convolutional radar information; The convolutional radar information is activated to obtain the initial radar feature information corresponding to the convolutional radar information. The initial radar feature information is subjected to multiple rounds of downsampling processing, and the processing result of each round is determined as multiple radar feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operation and activation operation; the convolution object of the first round is the initial radar feature information; the convolution object of any round other than the first round is the radar feature information obtained in the previous round; The optical remote sensing information is convolved to obtain convolutional remote sensing information; Multi-scale analysis is performed on the convolutional remote sensing information to obtain the initial remote sensing feature information corresponding to the convolutional remote sensing information; The initial remote sensing feature information is subjected to multiple rounds of downsampling processing, and the processing result of each round is determined as multiple optical remote sensing feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operation and multi-scale analysis; the convolution object of the first round is the initial remote sensing feature information; the convolution object of any round other than the first round is the optical remote sensing feature information obtained in the previous round. For each of the aforementioned dimensions, the radar feature information and the optical remote sensing feature information at that dimension are fused to obtain the fused feature information corresponding to that dimension. Multi-level feature extraction is performed on the fused feature information corresponding to each of the dimensions to obtain the target feature information corresponding to each of the dimensions. Based on the target feature information, hazard prediction is performed to obtain the hazard prediction result of the power transmission channel in the environment.

2. The method according to claim 1, characterized in that, The activation operation on the convolutional radar information to obtain the initial radar feature information corresponding to the convolutional radar information includes: Obtain the activation function consisting of hyperparameters and adaptive parameters; The initial radar feature information corresponding to the convolutional radar information is obtained by performing an activation operation on the convolutional radar information based on the activation function.

3. The method according to claim 1, characterized in that, The step of performing multi-scale analysis on the convolutional remote sensing information to obtain the initial remote sensing feature information corresponding to the convolutional remote sensing information includes: Convolutional remote sensing information is performed on convolutional kernels of different sizes to obtain multiple convolutional remote sensing feature information. The convolutional remote sensing features are combined with the optical remote sensing information to obtain the combined feature information; The merged feature information is subjected to activation, convolution, and channel weighting operations to obtain initial remote sensing feature information.

4. The method according to claim 1, characterized in that, The step of performing multi-level feature extraction on the fused feature information corresponding to each of the aforementioned dimensions to obtain the target feature information corresponding to each of the aforementioned dimensions includes: From the fusion feature information corresponding to each of the aforementioned sizes, determine the maximum fusion feature information corresponding to the largest size; The maximum fused feature information is enhanced to obtain enhanced fused feature information; Multi-level feature extraction is performed on the enhanced fusion feature information and the other fusion feature information besides the enhanced fusion feature information to obtain the target feature information corresponding to each of the dimensions.

5. A device for predicting potential hazards in power transmission channels, characterized in that, The device includes: The information acquisition module is used to acquire optical remote sensing information and synthetic aperture radar information collected from the environment where the power transmission channel is located; The convolutional feature extraction module includes: The first convolution operation unit is used to perform a convolution operation on the synthetic aperture radar information to obtain convolution radar information. An activation operation unit is used to activate the convolutional radar information to obtain the initial radar feature information corresponding to the convolutional radar information. The first downsampling processing unit is used to perform multiple rounds of downsampling processing on the initial radar feature information, and to determine the processing result of each round as multiple radar feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operation and activation operation; the convolution object of the first round is the initial radar feature information; the convolution object of any round other than the first round is the radar feature information obtained in the previous round. The second convolution operation unit is used to perform convolution operations on the optical remote sensing information to obtain convolutional remote sensing information. A multi-scale analysis unit is used to perform multi-scale analysis on the convolutional remote sensing information to obtain the initial remote sensing feature information corresponding to the convolutional remote sensing information. The second downsampling processing unit is used to perform multiple rounds of downsampling processing on the initial remote sensing feature information, and to determine the processing result of each round as multiple optical remote sensing feature information of different sizes; wherein, each round of downsampling processing includes sequential convolution operations and multi-scale analysis; the convolution object of the first round is the initial remote sensing feature information; the convolution object of any round other than the first round is the optical remote sensing feature information obtained in the previous round; The information fusion module is used to fuse the radar feature information and the optical remote sensing feature information for each of the dimensions to obtain the fused feature information corresponding to the dimension. A multi-level feature extraction module is used to perform multi-level feature extraction on the fused feature information corresponding to each of the dimensions to obtain the target feature information corresponding to each of the dimensions. The hazard prediction module is used to predict hazards based on the characteristic information of each target, and to obtain the hazard prediction result of the power transmission channel in the environment.

6. The apparatus according to claim 5, characterized in that, The activation operation unit is specifically used for: Obtain the activation function consisting of hyperparameters and adaptive parameters; The initial radar feature information corresponding to the convolutional radar information is obtained by performing an activation operation on the convolutional radar information based on the activation function.

7. The apparatus according to claim 5, characterized in that, The multi-scale analysis unit is specifically used for: Convolutional remote sensing information is performed on convolutional kernels of different sizes to obtain multiple convolutional remote sensing feature information. The convolutional remote sensing features are combined with the optical remote sensing information to obtain the combined feature information; The merged feature information is subjected to activation, convolution, and channel weighting operations to obtain initial remote sensing feature information.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images