Risk category assessment method and device of power transmission tower, terminal equipment and storage medium

Abstract

The invention discloses a risk category assessment method and device for a power transmission tower, terminal equipment and a storage medium, and relates to the technical field of tower assessment, and the method comprises the steps: obtaining real-time data of a target power transmission tower, the real-time data comprising time sequence data and image data; inputting the real-time data into a hybrid model composed of an attention mechanism and a residual network, extracting a time sequence feature vector of the time sequence data through the attention mechanism, and extracting an image feature vector in the image data through the residual network; and fusing the time sequence feature vector and the image feature vector, inputting the fused time sequence feature vector and image feature vector into a classifier, outputting the probability of each risk category, and selecting the category with the highest probability as a risk category label. According to the method, the time sequence feature vector in the time sequence data and the image feature vector in the image data are extracted, and the two different feature vectors are combined for evaluation, so that the accuracy of risk category evaluation of the power transmission tower can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of tower assessment technology, and in particular to a method, apparatus, terminal equipment and storage medium for assessing the risk category of power transmission towers. Background Technology

[0002] Sudden geological disasters such as mudslides and landslides in mountainous and hilly areas pose a huge threat to the safety of transmission line towers. Therefore, it is of great significance to collect and analyze the precursor information of geological disasters such as mudslides and landslides, so as to realize the monitoring and early warning of the status of transmission line towers.

[0003] Existing risk classification assessment methods for power transmission towers typically involve modeling the tower's state time series data using long short-term memory networks, training the model with normal data, and then assessing the risk category of the power transmission tower. However, using only time series data collected by sensors as input to the model to assess the risk category of the power transmission tower can easily lead to misjudgments due to single-mode failure, resulting in low accuracy in assessing the risk category of the power transmission tower. Summary of the Invention

[0004] This invention provides a method, apparatus, terminal equipment, and storage medium for assessing the risk category of power transmission towers. It can solve the technical problem that the existing technology, which only uses time-series data collected by sensors to input models to assess the risk category of power transmission towers, is prone to misjudgment due to single-mode failure, resulting in low accuracy of risk category assessment of power transmission towers.

[0005] This invention provides a method for risk category assessment of power transmission towers, comprising:

[0006] Acquire real-time data of the target transmission tower, wherein the real-time data includes time-series data and image data;

[0007] The real-time data is input into a hybrid model consisting of an attention mechanism and a residual network. The attention mechanism extracts the temporal feature vector of the temporal data, and the residual network extracts the image feature vector of the image data.

[0008] The time-series feature vector and the image feature vector are fused and input into the classifier, which outputs the probability of each risk category and selects the category with the highest probability as the risk category label.

[0009] Furthermore, before inputting the real-time data into the hybrid model consisting of an attention mechanism and a residual network, the following steps are also included:

[0010] The time-series data in the real-time data is subjected to data alignment processing, and an adaptive histogram equalization method is used to enhance the contrast of the crack region in the image data to obtain optimized real-time data.

[0011] Furthermore, the step of extracting the temporal feature vector of the time-series data through the attention mechanism includes:

[0012] The time series data is divided into several windows, and the time series data in each window is used as the input data for the attention mechanism.

[0013] The attention output vector of each window is calculated based on the input data, and the attention output vectors of all windows are fused to obtain the temporal feature vector.

[0014] Furthermore, the step of extracting image feature vectors from the image data through the residual network includes:

[0015] The residual network calculates the channel statistics of the image data based on the height and width of the image data;

[0016] The channel weights of the fully connected layer are determined based on the weight matrix of the fully connected layer in the residual network and the channel statistics of the image data.

[0017] The image data input to the corresponding channel is recalibrated according to the channel weights to obtain the recalibrated feature map of each channel;

[0018] The image feature vector of the image data is obtained by fusing the recalibrated feature maps of all channels.

[0019] Furthermore, the step of fusing the temporal feature vector and the image feature vector and then inputting them into the classifier includes:

[0020] The temporal feature vector and the image feature vector are fused using the following formula to obtain a fused feature vector, which is then input into the classifier:

[0021] Fused Feature=σ(W t ·T)·T+σ(W i ·I)·I

[0022] Where Fused Feature is the fused feature vector, I is the image feature vector, T is the temporal feature vector, and W is the fused feature vector. t W represents the learnable weights of the temporal feature vector. i σ represents the learnable weights of the image feature vector, and σ is the activation function.

[0023] Furthermore, the time-series data includes vibration signals, tilt angle change rate, and soil temperature and humidity data, and the image data includes surface crack image data.

[0024] This invention provides a risk category assessment device for power transmission towers, comprising:

[0025] A real-time data acquisition module is used to acquire real-time data of the target transmission tower, wherein the real-time data includes time-series data and image data;

[0026] The feature vector extraction module is used to input the real-time data into a hybrid model composed of an attention mechanism and a residual network, extract the temporal feature vector of the temporal data through the attention mechanism, and extract the image feature vector of the image data through the residual network;

[0027] The risk category assessment module is used to fuse the time-series feature vector and the image feature vector and input them into the classifier, output the probability of each risk category, and select the category with the highest probability as the risk category label.

[0028] Furthermore, the step of extracting image feature vectors from the image data through the residual network includes:

[0029] The residual network calculates the channel statistics of the image data based on the height and width of the image data;

[0030] The channel weights of the fully connected layer are determined based on the weight matrix of the fully connected layer in the residual network and the channel statistics of the image data.

[0031] The image data input to the corresponding channel is recalibrated according to the channel weights to obtain the recalibrated feature map of each channel;

[0032] The image feature vector of the image data is obtained by fusing the recalibrated feature maps of all channels.

[0033] The present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the risk category assessment method for power transmission towers as described above.

[0034] The present invention provides a computer-readable storage medium, comprising: a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the risk category assessment method for power transmission towers as described above.

[0035] The following benefits can be obtained by implementing the present invention:

[0036] This invention extracts time-series feature vectors from time-series data and image feature vectors from image data, fuses the two feature vectors, and inputs them into a classifier. The classifier then classifies the risk category of the transmission tower. By combining two different feature vectors for evaluation, it can effectively avoid misjudgments caused by single-mode failure, thereby effectively improving the accuracy of transmission tower risk category assessment.

[0037] Furthermore, this invention adaptively recalibrates channel weights through residual networks, enabling the model to focus more on the key visual features of transmission tower defects (such as corrosion, cracks, and tilting), thereby suppressing background interference and effectively improving the accuracy of risk category assessment. Moreover, the dynamic adjustment mechanism of channel statistics reduces the dependence on large-scale defect samples, ensuring the accuracy of risk categories in scenarios with scarce data (such as rare tower types or damage from extreme weather). Attached Figure Description

[0038] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0039] Figure 1 This is a flowchart illustrating the risk category assessment method for power transmission towers provided in an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram illustrating the principle of the sliding window attention mechanism provided in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the depth-separable convolution principle provided in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the channel attention mechanism provided in an embodiment of the present invention;

[0043] Figure 5 This is another flowchart illustrating a risk category assessment method for power transmission towers provided in an embodiment of the present invention;

[0044] Figure 6 This is a schematic diagram of the structure of the risk category assessment device for power transmission towers provided in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0047] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0048] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0049] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0050] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0051] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0052] See Figure 1 To address the technical problem that existing technologies, which rely solely on time-series data collected by sensors to input models for assessing the risk category of transmission towers, are prone to misjudgments due to single-mode failures, resulting in low accuracy in risk category assessment, an embodiment of the present invention provides a method for assessing the risk category of transmission towers, comprising:

[0053] S1. Acquire real-time data of the target transmission tower, including time-series data and image data;

[0054] In this embodiment of the invention, the target transmission tower is the transmission tower to be evaluated. Real-time data around the target transmission tower can be obtained with the transmission tower as the center and a preset distance as the radius. The time-series data includes vibration signals, tilt angle change rate, and soil temperature and humidity data, while the image data includes surface crack image data. The time-series data can be acquired by sensors, while the image data can be acquired by a wide-angle camera.

[0055] S2. Input real-time data into a hybrid model consisting of an attention mechanism and a residual network. Extract the temporal feature vector of the temporal data through the attention mechanism and extract the image feature vector of the image data through the residual network.

[0056] S3. After fusing the time-series feature vector and the image feature vector, input them into the classifier, output the probability of each risk category, and select the category with the highest probability as the risk category label.

[0057] In this embodiment of the invention, the classifier outputs a binary classification probability bar based on the input temporal feature vector and image feature vector. The classifier determines whether a target transmission tower poses a risk based on this probability bar; for example, if the binary classification probability is higher than a preset value, the target transmission tower is determined to pose a risk. Furthermore, the risk category is determined based on the output risk categories and their corresponding probabilities. Risk categories include debris flow, landslide, and structural failure. The probability corresponding to each risk category is compared, and the category with the highest probability is selected as the risk category label. For example, if the probability of debris flow is higher than the probabilities corresponding to landslide and structural failure, the current transmission tower is determined to pose a debris flow risk. In this embodiment of the invention, when a target transmission tower is determined to pose a risk, the risk-related information is uploaded to the cloud via a BeiDou or 4G module for early warning.

[0058] This invention extracts time-series feature vectors from time-series data and image feature vectors from image data, fuses the two feature vectors, and inputs them into a classifier. The classifier then classifies the risk category of the transmission tower. By combining two different feature vectors for evaluation, it can effectively avoid misjudgments caused by single-mode failure, thereby effectively improving the accuracy of transmission tower risk category assessment.

[0059] In one embodiment, before step S2, inputting real-time data into the hybrid model consisting of the attention mechanism and the residual network, the method further includes:

[0060] Data alignment is performed on the time-series data in the real-time data, and an adaptive histogram equalization method is used to enhance the contrast of the crack region in the image data to obtain optimized real-time data.

[0061] In this embodiment of the invention, a MEMS vibration sensor can be used to capture abnormal ground vibrations to obtain vibration signals, a dual-axis tilt sensor (accuracy ±0.01°, temperature compensation range -40 to 85°C) can be used to monitor the rate of change of the tower tilt angle, a soil moisture meter (range 0 to 100% VWC, accuracy ±3%) can be used to detect the risk of landslides caused by sudden changes in groundwater level to obtain soil temperature and humidity data, and a low-power wide-angle camera can be used to capture the trend of surface crack expansion to obtain surface crack image data.

[0062] In this embodiment of the invention, a hardware architecture for edge node computing can be used to preprocess real-time data, including synchronizing multi-source data timestamps based on the PTP protocol to align time-series data; and using adaptive histogram equalization (AHE) to enhance the contrast of crack regions to preprocess image data.

[0063] The embodiments of the present invention can eliminate the time difference of multi-source sensors through data alignment, avoiding the risk of misjudgment or omission; adaptive histogram equalization significantly enhances the contrast of crack areas, making early micro-cracks easier to detect and improving the accuracy of risk category assessment of transmission towers.

[0064] In one embodiment, step S2, extracting the temporal feature vector of the time series data through an attention mechanism, includes:

[0065] S201. Divide the time series data into several windows, and use the time series data in each window as the input data for the attention mechanism;

[0066] Please see Figure 2 In this embodiment of the invention, the 12-layer encoder of the attention mechanism can be reduced to 4 layers. The temporal data is input into the model and enters the sliding window attention. The temporal data is divided into k local windows (with a window size of w) through the sliding window.

[0067] S202. Calculate the attention output vector of each window based on the input data, and fuse the attention output vectors of all windows to obtain the temporal feature vector.

[0068] In this embodiment of the invention, the attention output vector is calculated independently within each window:

[0069]

[0070] Among them, Attention(Q i ,K i V i Let Q be the attention output vector of the i-th window, Softmax(·) be the normalization exponential function, and Q be the expression for Q. i ,K i V i ∈R w×d It is the query, key, and value matrix of the i-th window. Window overlap can be achieved by sliding a step size s. V is the scaling factor. i The value matrix is ​​used. In this embodiment of the invention, the window size w = 64 and the step size s = 32 are set to process vibration spectrum and tilt rate of change data, thereby balancing local details and long-range dependencies, and reducing the complexity from O(n) of standard attention to... 2 The computational complexity is reduced to O(n·w), which can adapt to the computing power limitations of edge devices.

[0071] To enhance the model's generalization ability and adaptability to long-term time-series data features, relative position encoding (RPE) is added:

[0072]

[0073] Among them, e ij Let Q be the query vector at position i. i The key vector K at position j j The original attention scores between Q. i Let K be the query vector at position i. j Let b be the key vector at position j. i-j The learnable bias term for the relative position (ij) between positions i and j can be obtained from the embedding table; d is the dimension of the query / key vector. This is the scaling factor.

[0074] The window-based attention mechanism of this invention can quickly lock onto instantaneous danger signals such as loose bolts and sudden increases in wind vibration, avoiding long sequences from overwhelming critical periods, and can reduce the computational complexity of time-series data while retaining the ability to model long-range dependencies.

[0075] In one embodiment, step S2, extracting image feature vectors from image data using a residual network, includes:

[0076] S211. The residual network calculates the channel statistics of the image data based on the height and width of the image data.

[0077] In this embodiment of the invention, the image data can be surface crack image data (256×256 pixels).

[0078] Please see Figure 3-4 The surface crack image data is input into the improved residual network. The principle diagram of the channel attention mechanism in this residual network is shown below. Figure 4 As shown. Among them, Figure 4 In the diagram, X represents the input feature map, H′, W′, and C′ represent the height, width, and number of channels of the input feature map, respectively. U represents the convolutional feature map, where H, W, and C represent the height, width, and number of channels of the convolutional feature map, respectively. F... sq (·) represents global information aggregation, F ex (·, W) represents the channel weights learned, F scale (·,·) represents coordinate recalibration. The output feature map is recalibrated (weighted) for the c-th channel. In this embodiment of the invention, the residual network removes the last two stages, retaining the first three stages (output feature map size 28×28). Each stage contains a set of three residual blocks, and each residual block contains a convolution using a combination of 3×3 depthwise convolutions and 1×1 pointwise convolutions. Figure 3 The depth shown can be separated into convolutional and pooling layers as well as fully connected layers. Channel attention (SE Block) is added between each residual block to enhance the feature weights of crack edges.

[0079] In this embodiment of the invention, the surface crack image data first enters the residual block through depthwise separable convolution, which compresses the model size to 1.8MB while preserving the crack morphology features. The input is then directly passed to the subsequent channel attention (SEBlock) via a skip connection. The channel attention (SEBlock) obtains channel statistics through compression.

[0080]

[0081] Among them, z c x represents the global average pooling result (i.e., channel statistics) for the c-th channel; c (i,j) represents the value of the c-th channel of the input feature map at spatial position (i,j); H represents the height of the convolutional feature map; W represents the width of the convolutional feature map; and c represents the index of the current channel.

[0082] S212. Determine the channel weights of the fully connected layer based on the weight matrix of the fully connected layer in the residual network and the channel statistics of the image data.

[0083] In this embodiment of the invention, channel weights can be learned through a fully connected layer:

[0084] s=σ(W2·δ(W1·z)) (4)

[0085] Where s is the learned channel weight vector (each element corresponds to the scaling factor of a channel); z is the channel statistics vector obtained from the compression operation (z = [z1, z2, ..., z2]). C W1 is the weight matrix of the first fully connected layer, with dimensions [missing information]. Used for dimensionality reduction; W2 is the weight matrix of the second fully connected layer, with dimensions of [missing information]. Used for dimensionality reduction; δ is the ReLU activation function (Rectified Linear Unit); σ is the Sigmoid activation function (compresses the output to the range [0,1]); C is the total number of channels in the feature map; r is the compression ratio, used to control the degree of dimensionality reduction (for example, r=16 means that the intermediate layer dimension is C / 16). R is the set of real numbers.

[0086] S213. Recalibrate the image data input to the corresponding channel according to the channel weight to obtain the recalibrated feature map of each channel;

[0087] In this embodiment of the invention, the recalibration expression is as follows:

[0088]

[0089] in, The output feature map after recalibration (weighting) of the c-th channel; s is the learned weight corresponding to the c-th channel (from the c-th element of vector s); x c This is the original input feature map for the c-th channel.

[0090] S214. Fuse the recalibrated feature maps of all channels to obtain the image feature vector of the image data.

[0091] This invention can also reduce the number of parameters by adjusting the compression ratio r (set to 16) to adapt to edge computing resource constraints, increase the weight of key channels such as crack edges, suppress background noise, and automatically focus on crack areas in mountain crack images, resulting in a 9.2% improvement in the image branch F1-score.

[0092] The embodiments of the present invention adaptively recalibrate the channel weights through residual networks, enabling the model to focus more on the key visual features of transmission tower defects (such as corrosion, cracks, and tilting), thereby suppressing background interference and effectively improving the accuracy of risk category assessment.

[0093] The embodiments of the present invention reduce the reliance on large-scale defect samples through a dynamic adjustment mechanism of channel statistics, and ensure the accuracy of risk categories in scenarios with scarce data (such as rare tower types or extreme weather damage).

[0094] In one embodiment, step S3, fusing the temporal feature vector and the image feature vector and then inputting it into the classifier, includes:

[0095] The temporal feature vector and the image feature vector are fused using the following formula to obtain a fused feature vector, which is then input into the classifier:

[0096] Fused Feature=σ(W t ·T)·T+σ(W i ·I)·I

[0097] Where Fused Feature is the fused feature vector, I is the image feature vector, T is the temporal feature vector, and W is the fused feature vector. t W represents the learnable weights of the temporal feature vector. i σ represents the learnable weights of the image feature vector, and σ is the activation function.

[0098] In one embodiment, the fusion weights can be dynamically adjusted based on the real-time confidence of temporal and image features (such as vibration spectrum energy and crack clarity) to further avoid misjudgments caused by single-mode failure.

[0099] Please see Figure 5 This is another flowchart illustrating a risk category assessment method for power transmission towers provided in an embodiment of the present invention. Figure 5 In this process, data is collected through a sensor array and preprocessed using an edge computing node PK3588. The preprocessed data includes time-series data such as vibration and tilt angle, as well as image data captured by a camera. The time-series data is processed using a Transformer (attention mechanism) model, with feature extraction based on sliding window attention and relative position encoding to obtain a time-series feature vector. The image data is processed using a ResNet (residual network) model, with feature extraction through depthwise separable convolution and channel attention to obtain an image feature vector. A dynamic fusion mechanism layer fuses the time-series and image feature vectors, outputting a binary classification probability bar and risk category label, and then uploading the abnormal data via BeiDou / 4G.

[0100] Implementing the embodiments of the present invention has the following beneficial effects:

[0101] This invention extracts time-series feature vectors from time-series data and image feature vectors from image data, fuses the two feature vectors, and inputs them into a classifier. The classifier then classifies the risk category of the transmission tower. By combining two different feature vectors for evaluation, it can effectively avoid misjudgments caused by single-mode failure, thereby effectively improving the accuracy of transmission tower risk category assessment.

[0102] Furthermore, this embodiment of the invention adaptively recalibrates channel weights through a residual network, enabling the model to focus more on the key visual features of transmission tower defects (such as corrosion, cracks, and tilting), thereby suppressing background interference and effectively improving the accuracy of risk category assessment. Moreover, the dynamic adjustment mechanism of channel statistics reduces the dependence on large-scale defect samples, ensuring the accuracy of risk categories in scenarios with scarce data (such as rare tower types or damage from extreme weather).

[0103] like Figure 6 As shown, based on the above method embodiments, corresponding apparatus embodiments are provided;

[0104] An embodiment of the present invention provides a risk category assessment device for power transmission towers, comprising:

[0105] The real-time data acquisition module 10 is used to acquire real-time data of the target transmission tower, including time-series data and image data.

[0106] The feature vector extraction module 20 is used to input real-time data into a hybrid model composed of an attention mechanism and a residual network, extract the temporal feature vector of the temporal data through the attention mechanism, and extract the image feature vector of the image data through the residual network;

[0107] The risk category assessment module 30 is used to fuse the time-series feature vector and the image feature vector and input them into the classifier, output the probability of each risk category, and select the category with the highest probability as the risk category label.

[0108] In one embodiment, before feeding real-time data into a hybrid model consisting of an attention mechanism and a residual network, the method further includes:

[0109] Data alignment is performed on the time-series data in the real-time data, and an adaptive histogram equalization method is used to enhance the contrast of the crack region in the image data to obtain optimized real-time data.

[0110] In one embodiment, extracting the temporal feature vector from the temporal data using an attention mechanism includes:

[0111] The time series data is divided into several windows, and the time series data in each window is used as the input data for the attention mechanism.

[0112] The attention output vector for each window is calculated based on the input data, and the attention output vectors of all windows are fused to obtain the temporal feature vector.

[0113] In one embodiment, extracting image feature vectors from image data using a residual network includes:

[0114] The residual network calculates the channel statistics of the image data based on the height and width of the image data;

[0115] The channel weights of the fully connected layer are determined based on the weight matrix of the fully connected layer in the residual network and the channel statistics of the image data.

[0116] The image data input to the corresponding channel is recalibrated according to the channel weight to obtain the recalibrated feature map of each channel;

[0117] The image feature vector of the image data is obtained by fusing the recalibrated feature maps of all channels.

[0118] In one embodiment, the fused temporal feature vector and image feature vector are input into the classifier, including:

[0119] The temporal feature vector and the image feature vector are fused using the following formula to obtain a fused feature vector, which is then input into the classifier:

[0120] Fused Feature=σ(W t ·T)·T+σ(W i ·I)·I

[0121] Where Fused Feature is the fused feature vector, I is the image feature vector, T is the temporal feature vector, and W is the fused feature vector. t W represents the learnable weights of the temporal feature vector. i σ represents the learnable weights of the image feature vector, and σ is the activation function.

[0122] In one embodiment, the time-series data includes vibration signals, tilt rate of change, and soil temperature and humidity data, and the image data includes surface crack image data.

[0123] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the risk category assessment method for power transmission towers provided by any of the above-described method embodiments of the present invention.

[0124] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0125] Based on the above-described embodiments of the risk category assessment method for transmission towers, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the risk category assessment method for transmission towers according to any embodiment of the present invention.

[0126] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0127] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0128] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0129] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the risk category assessment method for power transmission towers described in any of the above-described method embodiments of the present invention.

[0130] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0131] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for assessing the risk category of power transmission towers, characterized in that, include: Acquire real-time data of the target transmission tower, wherein the real-time data includes time-series data and image data; The real-time data is input into a hybrid model consisting of an attention mechanism and a residual network. The attention mechanism is used to extract the temporal feature vector of the time-series data, and the residual network is used to extract the image feature vector of the image data. The time-series feature vector and the image feature vector are fused and input into the classifier, which outputs the probability of each risk category and selects the category with the highest probability as the risk category label.

2. The risk category assessment method for transmission towers as described in claim 1, characterized in that, Before inputting the real-time data into the hybrid model consisting of an attention mechanism and a residual network, the following steps are also included: The time-series data in the real-time data is subjected to data alignment processing, and an adaptive histogram equalization method is used to enhance the contrast of the crack region in the image data to obtain optimized real-time data.

3. The risk category assessment method for transmission towers as described in claim 1, characterized in that, The step of extracting the temporal feature vector of the time-series data through the attention mechanism includes: The time series data is divided into several windows, and the time series data in each window is used as the input data for the attention mechanism. The attention output vector of each window is calculated based on the input data, and the attention output vectors of all windows are fused to obtain the temporal feature vector.

4. The risk category assessment method for transmission towers as described in claim 1, characterized in that, The step of extracting image feature vectors from the image data using the residual network includes: The residual network calculates the channel statistics of the image data based on the height and width of the image data; The channel weights of the fully connected layer are determined based on the weight matrix of the fully connected layer in the residual network and the channel statistics of the image data. The image data input to the corresponding channel is recalibrated according to the channel weights to obtain the recalibrated feature map of each channel; The image feature vector of the image data is obtained by fusing the recalibrated feature maps of all channels.

5. The risk category assessment method for transmission towers as described in claim 1, characterized in that, The step of fusing the temporal feature vector and the image feature vector and then inputting them into the classifier includes: The temporal feature vector and the image feature vector are fused using the following formula to obtain a fused feature vector, which is then input into the classifier: Fused Feature=σ(W t ·T)·T+σ(W i ·I)·I Where Fused Feature is the fused feature vector, I is the image feature vector, T is the temporal feature vector, and W is the fused feature vector. t W represents the learnable weights of the temporal feature vector. i σ represents the learnable weights of the image feature vector, and σ is the activation function.

6. The risk category assessment method for transmission towers as described in claim 1, characterized in that, The time-series data includes vibration signals, tilt angle change rate, and soil temperature and humidity data, and the image data includes surface crack image data.

7. A risk category assessment device for power transmission towers, characterized in that, include: A real-time data acquisition module is used to acquire real-time data of the target transmission tower, wherein the real-time data includes time-series data and image data; The feature vector extraction module is used to input the real-time data into a hybrid model composed of an attention mechanism and a residual network, extract the temporal feature vector of the time-series data through the attention mechanism, and extract the image feature vector of the image data through the residual network; The risk category assessment module is used to fuse the time-series feature vector and the image feature vector and input them into the classifier, output the probability of each risk category, and select the category with the highest probability as the risk category label.

8. The risk category assessment device for transmission towers as described in claim 7, characterized in that, The step of extracting image feature vectors from the image data using the residual network includes: The residual network calculates the channel statistics of the image data based on the height and width of the image data; The channel weights of the fully connected layer are determined based on the weight matrix of the fully connected layer in the residual network and the channel statistics of the image data. The image data input to the corresponding channel is recalibrated according to the channel weights to obtain the recalibrated feature map of each channel; The image feature vector of the image data is obtained by fusing the recalibrated feature maps of all channels.

9. A terminal device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the risk category assessment method for transmission towers as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the risk category assessment method for transmission towers as described in any one of claims 1-6.

Images