Power distribution network telegraph pole inclination detection method, system, equipment and medium

Abstract

The invention relates to the technical field of power systems, and discloses a power distribution network telegraph pole inclination detection method, system and device and a medium. According to the method, an image data set containing annotation information of a telegraph pole is input into a telegraph pole image segmentation model for telegraph pole image segmentation; detecting the position information of the main axis of the telegraph pole, calculating the inclination of the telegraph pole according to the position information of the main axis of the telegraph pole, and training a deep neural network according to the image data set and the inclination of the telegraph pole of each piece of image data; through the telegraph pole inclination detection model, telegraph pole inclination detection is carried out on the image data of the current detection time period to obtain the telegraph pole inclination, so that the telegraph pole inclination state is detected through the telegraph pole inclination detection model, and the detection precision and the real-time response capability of the model to the telegraph pole inclination state are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method, system, equipment and medium for detecting the inclination of electric poles in a distribution network. Background Art

[0002] In the power industry, the stability of distribution network poles is crucial for ensuring the safety and reliability of power supply. However, due to both natural factors (such as strong winds, heavy rain, and earthquakes) and human influences, poles may tilt or even collapse, which can not only cause power outages but also lead to serious safety accidents. Traditional methods for detecting pole tilt rely primarily on manual inspections or simple sensor monitoring. These methods suffer from low detection efficiency, insufficient accuracy, and delayed response. This makes it difficult to accurately monitor the status of poles in real time, especially in areas with complex terrain or harsh environments. This hinders the timely warning and handling of potential risks. Summary of the Invention

[0003] In view of this, the present invention provides a method, system, device and medium for detecting the tilt of power poles in a distribution network, which solves the problems of low detection efficiency, insufficient accuracy, delayed response and other problems existing in the method for detecting the tilt of power poles in a distribution network. In particular, in areas with complex terrain or harsh environment, it is difficult to achieve real-time and accurate monitoring of the status of power poles, thereby affecting the timely warning and handling of potential risks. Technical problems.

[0004] A first aspect of the present invention provides a method for detecting the inclination of a power pole in a distribution network, comprising:

[0005] Acquire multiple image data of the power distribution network area to form an image dataset; wherein the image dataset includes annotation information of utility poles;

[0006] Inputting the image data set into a utility pole image segmentation model to obtain utility pole image segmentation results for each of the image data; wherein the utility pole image segmentation model is used to segment the image data according to the annotation information of the utility poles and locate the positions of the utility poles in the image data;

[0007] Detecting the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculating the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data;

[0008] Training a deep neural network based on the image dataset and the inclination of the utility poles in each of the image data to obtain a utility pole inclination detection model;

[0009] The utility pole tilt detection model is used to perform utility pole tilt detection on the image data of the current detection period to obtain the utility pole tilt corresponding to the image data of the current detection period.

[0010] Preferably, acquiring a plurality of image data of the distribution network area to form an image data set includes:

[0011] Collecting a plurality of image data of the power distribution network area, wherein the image data includes regional image information of utility poles;

[0012] performing denoising processing on the plurality of image data;

[0013] Segmenting and labeling the denoised image data to obtain labeling information of the utility pole;

[0014] The image data set is constructed based on a plurality of labeled image data.

[0015] Preferably, inputting the image dataset into a utility pole image segmentation model to obtain utility pole image segmentation results for each image data comprises:

[0016] For each image data in the image dataset, performing a first dilated convolution operation on the image data to obtain a first utility pole feature map;

[0017] The first utility pole feature map is calculated through two branches respectively, wherein, in one branch, the first utility pole feature map is subjected to multiple parallel first dilated residual convolution operations to obtain multiple second utility pole feature maps; in the other branch, the first utility pole feature map is subjected to a global context aggregation operation to obtain a third utility pole feature map;

[0018] Performing a feature addition operation on the plurality of second utility pole feature maps to obtain a fourth utility pole feature map, and performing a second dilated residual convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map;

[0019] Performing an upsampling operation on the fifth utility pole feature map to obtain a sixth utility pole feature map;

[0020] Performing a channel superposition operation on the third utility pole characteristic map and the sixth utility pole characteristic map to obtain a seventh utility pole characteristic map;

[0021] Performing a third dilated residual convolution operation on the seventh utility pole feature map to obtain an eighth utility pole feature map, and performing an upsampling operation on the eighth utility pole feature map to obtain a ninth utility pole feature map;

[0022] The ninth utility pole feature map is input into an image segmentation classifier for classification to obtain the utility pole image segmentation result.

[0023] Preferably, the process of the dilated residual convolution operation includes:

[0024] Performing a second dilated convolution operation on the first utility pole feature map to obtain a first utility pole sub-feature map;

[0025] performing a batch normalization operation on the first utility pole sub-feature map to obtain a second utility pole feature map;

[0026] Performing nonlinear processing on the second utility pole characteristic graph through a Swish function to obtain a third utility pole characteristic graph;

[0027] Performing a channel superposition operation on the third utility pole sub-characteristic map and the first utility pole sub-characteristic map to obtain a fourth utility pole sub-characteristic map;

[0028] Performing a third dilated convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map;

[0029] The fifth utility pole sub-feature map is normalized to obtain a sixth utility pole sub-feature map, and the sixth utility pole sub-feature map is nonlinearly processed using a Swish function to obtain the second utility pole feature map.

[0030] Preferably, the process of the global context aggregation operation includes:

[0031] The first utility pole feature map is calculated in parallel on three sub-branches, wherein in the three sub-branches, a multi-level dilated convolution operation is performed on the first utility pole feature map, and then a channel superposition operation is performed on the utility pole sub-feature maps obtained by the multi-level dilated convolution operation to obtain seventh utility pole sub-feature maps corresponding to the three sub-branches respectively; wherein the dilated convolutions in the multi-level dilated convolution operation have different scales;

[0032] Performing a feature addition operation on the three seventh utility pole sub-feature maps to obtain an eighth utility pole feature map;

[0033] A fourth dilated convolution operation is performed on the eighth utility pole sub-feature map to obtain the third utility pole feature map.

[0034] Preferably, the loss function of the image segmentation classifier is expressed as:

[0035]

[0036] Where, is the loss value, 、 are all random numbers between 0 and 1. Pixel The predicted probability, N is the number of pixels, Pixel The true value of .

[0037] Preferably, the detecting the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculating the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data includes:

[0038] For each of the utility pole image segmentation results, extracting a connected region of pixels marked as a utility pole according to the utility pole image segmentation result;

[0039] Performing foreground detection on the connected pixel region to obtain the regional position of the utility pole, and cropping the connected pixel region according to the regional position of the utility pole to obtain a utility pole regional image;

[0040] Performing edge detection on the utility pole area image to obtain edge contours of the utility poles;

[0041] Performing Hough line detection on the edge contour of the utility pole to obtain the main axis position information of the utility pole; wherein the main axis position information includes the top center point position and the bottom center point position of the utility pole;

[0042] Coordinate calculation is performed based on the top center point position and the bottom center point position of the utility pole to obtain the inclination of the utility pole.

[0043] In a second aspect, the present invention provides a distribution network electric pole tilt detection system, comprising:

[0044] An image acquisition module is used to acquire multiple image data of the power distribution network area to form an image data set; wherein the image data set includes annotation information of the utility poles;

[0045] An image segmentation module is configured to input the image data set into a utility pole image segmentation model to obtain a utility pole image segmentation result for each of the image data; wherein the utility pole image segmentation model is configured to segment the image data and locate the utility pole position in the image data based on the annotation information of the utility pole;

[0046] an inclination calculation module, configured to detect the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculate the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data;

[0047] a tilt detection training module, configured to train a deep neural network based on the image dataset and the tilt of the utility poles in each of the image data to obtain a utility pole tilt detection model;

[0048] The inclination detection module is used to perform electric pole inclination detection on the image data of the current detection period using the electric pole inclination detection model to obtain the electric pole inclination corresponding to the image data of the current detection period.

[0049] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for detecting the inclination of power poles in a distribution network as described in the first aspect.

[0050] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of the method for detecting the inclination of power poles in a distribution network as described in the first aspect.

[0051] It can be seen from the above technical solutions that the present invention performs telephone pole image segmentation by inputting an image dataset containing labeled information of telephone poles into a telephone pole image segmentation model, detects the main axis position information of the telephone pole according to the telephone pole image segmentation result, and calculates the inclination of the telephone pole according to the main axis position information of the telephone pole, and then trains a deep neural network based on the image dataset and the telephone pole inclination of each image data to obtain a telephone pole inclination detection model, and performs telephone pole inclination detection on the image data of the current detection period through the telephone pole inclination detection model to obtain the telephone pole inclination, thereby detecting the inclination state of the telephone pole through the telephone pole inclination detection model, greatly improving the detection accuracy and real-time response capability of the model for the inclination state of the telephone pole, ensuring the immediate recognition and accurate monitoring of the inclination changes of the telephone pole, thereby optimizing the operation and maintenance efficiency of the power grid and ensuring the safety and stability of the power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A diagram illustrating an application environment of a method for detecting the tilt of electric poles in a distribution network provided by an embodiment of the present invention;

[0054] Figure 2A flow chart of a method for detecting the inclination of electric poles in a distribution network provided by an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of the structure of a utility pole image segmentation model provided by an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of the structure of the ARCM module provided in an embodiment of the present invention;

[0057] Figure 5 A schematic diagram of the structure of a GCAM module provided in an embodiment of the present invention;

[0058] Figure 6 A schematic diagram of utility pole tilt detection provided by an embodiment of the present invention;

[0059] Figure 7 A schematic structural diagram of a power distribution network pole tilt detection system provided by an embodiment of the present invention;

[0060] Figure 8 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0062] The method for detecting the inclination of power poles in a distribution network provided by the embodiment of the present application can be applied to Figure 1In the application environment shown. The terminal 101 communicates with the server 102 via a network. The data storage system can store data that the server 102 needs to process. The data storage system can be integrated on the server 102, or placed on the cloud or other network servers. The terminal 101 or the server 102 forms an image dataset by acquiring multiple image data of the distribution network area; wherein the image dataset contains annotation information of the utility poles; the image dataset is input into the utility pole image segmentation model to obtain the utility pole image segmentation results of each image data; wherein the utility pole image segmentation model is used to segment the image data according to the annotation information of the utility poles and locate the position of the utility poles in the image data; according to each utility pole image segmentation result, the main axis position information of the utility pole is detected, and the inclination of the utility pole is calculated according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data; a deep neural network is trained according to the image dataset and the inclination of the utility pole of each image data to obtain a utility pole inclination detection model; the utility pole inclination detection model is used to detect the inclination of the image data of the current detection period to obtain the inclination of the utility pole corresponding to the image data of the current detection period.

[0063] The terminal 101 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, and the like.

[0064] The server 102 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0065] like Figure 2 As shown, the embodiment of the present application provides a method for detecting the inclination of power poles in a power distribution network, and the method is applied to Figure 1 The terminal 101 or the server 102 in the embodiment is used as an example to illustrate the method, which includes the following steps S1 to S5.

[0066] Step S1: Acquire multiple image data of a distribution network area to form an image dataset; wherein the image dataset includes annotation information of utility poles.

[0067] For example, drones, fixed cameras or mobile inspection equipment are used to collect images of utility poles in the distribution network area in various environments and at different angles to improve the generalization ability of the model. Various environments include daytime, nighttime, rainy and snowy weather, etc. Different angles refer to front, side, and overhead views.

[0068] By automatically labeling the utility poles in the image data, the labeling information of the utility poles is obtained.

[0069] Step S2: input the image data set into the utility pole image segmentation model to obtain the utility pole image segmentation results of each image data; wherein the utility pole image segmentation model is used to segment the image data according to the annotation information of the utility poles and locate the positions of the utility poles in the image data.

[0070] Among them, by constructing a distribution network pole image segmentation model (DNPM), the pole image of the distribution network area is input into the DNPM model. The model automatically segments and locates the pole area position in the image, thereby obtaining the pole image segmentation result.

[0071] The result of the utility pole image segmentation is to separate the utility pole area in the image from the background and retain only the utility pole part. The utility pole image segmentation result not only contains the location information of the utility pole, but also ensures the integrity and accuracy of the utility pole area, effectively avoiding detection errors caused by background interference or missing utility pole parts.

[0072] Step S3: Detect the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculate the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data.

[0073] The main axis position information of the utility pole is a geometric feature of the utility pole in the image. It describes the linear extension direction of the utility pole in the image. The main axis position information includes the coordinates of its starting point and ending point in the image, which correspond to the top and bottom center points of the utility pole, respectively.

[0074] After obtaining the main axis position information, the angle between the main axis and the horizontal axis of the image is calculated through trigonometric operations to determine the pole's tilt. This accurately reflects the pole's tilt in the image. The calculated tilt provides a quantitative pole tilt indicator for each image data point, which is crucial for subsequent pole health assessments and tilt warnings.

[0075] Step S4: training a deep neural network based on the image data set and the inclination of the utility poles in each image data to obtain a utility pole inclination detection model.

[0076] After obtaining the image dataset and the utility pole tilt values ​​for each image, the dataset is divided into training and validation sets according to a specific ratio. All parameters of the neural network model are initialized, and hyperparameters related to model training, such as the learning rate, batch size, and number of iterations, are configured. The training and validation data sets are then divided into multiple batches. During each iteration, a batch of training data is fed into the model for forward and backward propagation. The training loss for that batch is calculated, and the model parameters are updated to obtain the initial utility pole tilt detection model.

[0077] Simultaneously, the validation set data is fed into the initial utility pole tilt detection model in batches, and validation loss is calculated to evaluate model performance. The utility pole tilt detection model automatically adjusts its parameters based on the training loss and monitors the model's generalization ability using validation loss. This training process continues until both the training and validation losses stabilize, indicating that the model has reached convergence and training is complete, resulting in a utility pole tilt detection model.

[0078] Step S5: Performing a pole tilt detection on the image data of the current detection period using a pole tilt detection model to obtain the pole tilt corresponding to the image data of the current detection period.

[0079] Among them, after the training of the utility pole tilt detection model is completed, the utility pole tilt detection model is applied to the real-time monitoring and early warning of the status of utility poles in the distribution network.

[0080] First, ensure that the model can receive real-time images of utility poles from distribution stations and set reasonable alarm thresholds to ensure that the detection confidence level of the pole's tilt status meets the required level. Once the utility pole tilt detection model detects a tilted pole and the confidence level exceeds the preset alarm threshold, the system automatically triggers an alarm mechanism. This mechanism records key information such as the specific time of tilt, the pole's location, and the tilt angle. It then immediately notifies the pole's operation and maintenance personnel and security personnel via multiple channels such as email, text messages, or system notifications, ensuring that potential safety hazards are promptly identified and addressed. This effectively improves the ability to monitor the status of distribution network poles, avoids safety incidents that may be caused by tilt issues, and further ensures the stable operation and safety of the distribution network.

[0081] It should be noted that the embodiment of the present application performs telephone pole image segmentation by inputting an image dataset containing labeled information of telephone poles into a telephone pole image segmentation model, detects the main axis position information of the telephone pole based on the telephone pole image segmentation result, and calculates the inclination of the telephone pole based on the main axis position information of the telephone pole, and then trains a deep neural network based on the image dataset and the telephone pole inclination of each image data to obtain a telephone pole inclination detection model, and performs telephone pole inclination detection on the image data of the current detection period through the telephone pole inclination detection model to obtain the telephone pole inclination, thereby detecting the inclination state of the telephone pole through the telephone pole inclination detection model, greatly improving the detection accuracy and real-time response capability of the model for the inclination state of the telephone pole, ensuring the immediate recognition and accurate monitoring of the inclination changes of the telephone pole, thereby optimizing the operation and maintenance efficiency of the power grid and ensuring the safety and stability of the power grid operation.

[0082] In some embodiments, acquiring a plurality of image data of a power distribution network area to form an image dataset includes:

[0083] Step S101: Collect multiple image data of the power distribution network area, wherein the image data includes regional image information of utility poles.

[0084] Step S102: performing denoising processing on the plurality of image data.

[0085] The collected images were subjected to image denoising to improve subsequent image segmentation and construct a distribution network utility pole image dataset. The image data was expanded or compressed into feature maps of size H×W×3, with a length of H pixels, a width of W pixels, and three channels. The reason for this is that the images are in a three-channel format: red, green, and blue (RGB).

[0086] Step S103: segment and label the denoised image data to obtain labeling information of the utility poles.

[0087] You can use the image segmentation and annotation tool Super Annotate to annotate the image data. The categories are "telephone poles" and "background." Specifically, select the pixel areas in the image that contain only telephone poles and label them as "telephone poles." Pixel areas without telephone poles are labeled "background."

[0088] Step S104: construct an image data set based on the multiple labeled image data.

[0089] In some embodiments, the image dataset is input into a utility pole image segmentation model to obtain utility pole image segmentation results for each image data, including:

[0090] Step S201: For each image data in the image data set, perform a first dilated convolution operation on the image data to obtain a first utility pole feature map;

[0091] Step S202: Calculate the first utility pole feature map through two branches respectively. In one branch, perform multiple parallel first dilated residual convolution operations on the first utility pole feature map to obtain multiple second utility pole feature maps. In the other branch, perform a global context aggregation operation on the first utility pole feature map to obtain a third utility pole feature map.

[0092] Step S203: performing a feature addition operation on the plurality of second utility pole feature maps to obtain a fourth utility pole feature map, and performing a second dilated residual convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map;

[0093] Step S204: performing an upsampling operation on the fifth utility pole feature map to obtain a sixth utility pole feature map;

[0094] Step S205: performing a channel superposition operation on the third utility pole characteristic map and the sixth utility pole characteristic map to obtain a seventh utility pole characteristic map;

[0095] Step S206: performing a third dilated residual convolution operation on the seventh utility pole feature map to obtain an eighth utility pole feature map, and performing an upsampling operation on the eighth utility pole feature map to obtain a ninth utility pole feature map;

[0096] Step S207: input the ninth utility pole feature image into an image segmentation classifier for classification to obtain a utility pole image segmentation result.

[0097] like Figure 3 As shown, each image data in the image dataset is input into the DNPM model, and the model automatically segments and locates the location of the utility pole area in the image. The specific calculation process is as follows:

[0098] Input the utility pole feature map with the size of H×W×3 The image data is fed into an Adaptive Convolution (AConv) module with C convolution kernels, a size of 3×3, and a stride of 2 to perform the first dilated convolution operation, resulting in a utility pole feature map of size H / 2×W / 2×C. (First utility pole feature map).

[0099] right Calculation is performed on two branches:

[0100] In the first branch, Input three parallel first void residual convolution operation (Atrous Residual Convolution Module, ARCM) modules to obtain the utility pole feature map of the same size of H / 4×W / 4×C 、 and (Characteristics of the second utility pole). 、 and Perform feature addition (ADD) to obtain a pole feature map of size H / 4×W / 4×C. (Characteristics of the fourth utility pole). Input the ARCM module to obtain the characteristic map of the utility pole with the size of H / 8×W / 8×C (Fifth utility pole characteristic diagram). Perform a four-fold upsampling operation to obtain a utility pole feature map of size H / 2×W / 2×C (Characteristic diagram of the sixth utility pole);

[0101] In the second branch, Input the Global Context Aggregation Module (GCAM) module to obtain a pole feature map with a size of H / 2×W / 2×C. (Third pole characteristic diagram), the characteristic diagram of the pole with the same dimensions of H / 2×W / 2×C and Perform channel superposition (Concat) operation to obtain the pole feature map of size H / 2×W / 2×2C (Characteristic diagram of the seventh electric pole). Input the ARCM module to obtain the characteristic map of the utility pole with the size of H / 4×W / 4×C (Characteristics of the eighth electric pole). Perform a four-fold upsampling operation to obtain a utility pole feature map of size H×W×C (Characteristics of the ninth electric pole). Input to the image segmentation classifier and use the difficult-target loss (Difficult-target Loss, ) as the segmentation loss function, and obtain and output the segmentation result of the utility pole image. Utility pole image segmentation results. The utility pole image segmentation results include the category, pixel-level mask, and confidence level of the utility poles in the image. The category refers to the category label of the utility pole, which can be classified as "utility pole" or "background."

[0102] AConv is a publicly available deep learning convolutional algorithm that can reduce model parameters and improve operational efficiency. The channel concatenation (Concat) operation is used to concatenate two or more feature matrices along a certain dimension to generate a larger feature matrix. For example, a feature matrix of size H × W1 and a feature matrix of size H × W2 can be concat-ed to obtain a feature matrix of size H × (W1 + W2).

[0103] Among them, the process of the dilated residual convolution operation includes:

[0104] Step S2011: performing a second dilated convolution operation on the first utility pole feature map to obtain a first utility pole sub-feature map;

[0105] Step S2012: performing a batch normalization operation on the first utility pole feature map to obtain a second utility pole feature map;

[0106] Step S2013: performing nonlinear processing on the second utility pole characteristic graph through a Swish function to obtain a third utility pole characteristic graph;

[0107] Step S2014: performing a channel superposition operation on the third utility pole feature map and the first utility pole feature map to obtain a fourth utility pole feature map;

[0108] Step S2015: performing a third dilated convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map;

[0109] Step S2016: normalize the fifth utility pole sub-characteristic map to obtain a sixth utility pole sub-characteristic map, and perform nonlinear processing on the sixth utility pole sub-characteristic map using a Swish function to obtain a second utility pole characteristic map.

[0110] It should be noted that the dilated residual convolution operation is applicable to all dilated residual convolution operations such as the first dilated residual convolution operation, the second dilated residual convolution operation, and the third dilated residual convolution operation in the embodiment of the present application.

[0111] The embodiment of the present application designs an atrous residual convolution ARCM module. In the ARCM module, a lightweight atrous convolution (Atrous Convolution) is combined with a residual connection (Residual Connection) to quickly generate multi-scale utility pole features, reducing the number of model parameters and thus improving the overall operation speed of the model. The ARCM module can expand the receptive field of the detail features of the utility pole image, thereby better capturing the key information in the image and enhancing the model's image segmentation capability for utility poles. The architecture of the ARCM module is as follows: Figure 4 The operation process of the ARCM module is as follows:

[0112] The characteristic map of the utility pole with dimensions of H×W×C The first utility pole feature map is fed into a dilated convolution (3×3AConv) with C convolution kernels, a size of 3×3, and a stride of 2, resulting in a utility pole feature map of size H / 2×W / 2×C. (Characteristics of the first electric pole). Perform batch normalization (BatchNorm, BN) operation to obtain the pole feature map of size H / 2×W / 2×C (Feature diagram of the second electric pole). Then The nonlinear processing is performed by Swish function calculation to obtain the characteristic map of the utility pole with the size of H / 2×W / 2×C. (The third utility pole feature diagram). The utility pole feature diagram with the same dimensions of H / 2×W / 2×C and Perform channel superposition (Concat) operation to obtain the pole feature map of size H×W×C (Characteristic image of the fourth electric pole).

[0113] Will Input into the dilated convolution (3×3AConv) with C convolution kernels, size 3×3, and stride 2, and obtain the pole feature map of size H / 2×W / 2×C. (Characteristics of the fifth electric pole). Perform a BatchNorm operation to obtain a utility pole feature map of size H / 2×W / 2×C. (Characteristic diagram of the sixth electric pole). Then The nonlinear processing is performed by Swish function calculation to obtain and output the characteristic map of the utility pole with the size of H / 2×W / 2×C. (Feature diagram of the second utility pole).

[0114] It is understood that the atrous residual convolution ARCM module designed in the embodiments of this application uses atrous convolution and residual connections to rapidly generate multi-scale utility pole feature images, reducing the number of model parameters and thereby improving the model's overall computational speed. By expanding the receptive field of detail features in the utility pole image, the ARCM module can better capture key information in the image and enhance the model's ability to segment utility pole images.

[0115] In some embodiments, the process of global context aggregation operation includes:

[0116] Step S2021: Parallel calculations are performed on the first utility pole feature map on the three sub-branches. In each of the three sub-branches, a multi-level dilated convolution operation is performed on the first utility pole feature map. Then, a channel superposition operation is performed on the utility pole sub-feature maps obtained by the multi-level dilated convolution operation to obtain seventh utility pole sub-feature maps corresponding to the three sub-branches. The scales of the dilated convolutions in the multi-level dilated convolution operation are different.

[0117] Step S2022: performing a feature addition operation on the three seventh utility pole feature maps to obtain an eighth utility pole feature map;

[0118] Step S2023: Perform a fourth dilated convolution operation on the eighth utility pole sub-feature map to obtain a third utility pole feature map.

[0119] This paper designs a Global Context Aggregation Module (GCAM). GCAM effectively captures global context information in the image through multi-level dilated convolution feature extraction and then recalibrates the channel weights of the feature map through an attention mechanism, thereby enhancing the model's sensitivity to important features. GCAM not only improves the model's ability to perceive key details in utility pole images, but also maintains a high level of feature expression while reducing computational complexity, further optimizing the accuracy and efficiency of utility pole image segmentation. Figure 5 The specific calculation process is as follows:

[0120] Input the utility pole feature map with the size of H×W×C (The first pole characteristic diagram) is calculated on three branches (three sub-branches) respectively, such as Figure 5 As shown:

[0121] In the first branch (first sub-branch), Input into the dilated convolution (3×1 AConv) with C convolution kernels, size 3×1, and stride 1, and obtain the pole feature map of size H×W×C. .Will Input into the dilated convolution (1×3AConv) with C convolution kernels, 1×3 size, and 1 stride, and obtain the pole feature map of size H×W×C. .Will Input into the dilated convolution (3×3 AConv) with C convolution kernels, size 3×3, and stride 1, and obtain the pole feature map of size H×W×C. The characteristic diagram of the utility pole with the same size of H×W×C and Perform channel superposition (Concat) operation to obtain the pole feature map of size H×W×2C .

[0122] In the second branch (second sub-branch), Input into the dilated convolution (5×1 AConv) with C convolution kernels, size 5×1, and stride 1, and obtain the pole feature map of size H×W×C. .Will Input into the dilated convolution (1×5AConv) with C convolution kernels, 1×5 size, and 1 stride, and obtain the pole feature map of size H×W×C. .Will Input into the dilated convolution (3×3 AConv) with C convolution kernels, size 3×3, and stride 1, and obtain the pole feature map of size H×W×C. The characteristic diagram of the utility pole with the same size of H×W×C and Perform channel superposition (Concat) operation to obtain the pole feature map of size H×W×2C .

[0123] In the third branch (third sub-branch), Input into the dilated convolution (5×1 AConv) with C convolution kernels, size 5×1, and stride 1, and obtain the pole feature map of size H×W×C. .Will Input into the dilated convolution (1×5 AConv) with C convolution kernels, 1×5 size, and 1 stride, and obtain the pole feature map of size H×W×C. .Will Input into the dilated convolution (3×3AConv) with C convolution kernels, size 3×3, and stride 1, and obtain the pole feature map of size H×W×C. The characteristic diagram of the utility pole with the same size of H×W×C and Perform channel superposition (Concat) operation to obtain the pole feature map of size H×W×2C .

[0124] The characteristic diagram of the utility pole with the same size of H×W×2C 、 and Perform feature addition (ADD) operation to obtain the pole feature map of size H×W×2C .Will Input into the dilated convolution (3×3 AConv) with C convolution kernels, size 3×3, and stride 1, and obtain and output the pole feature map of size H×W×C. , and output the pole feature map (Third utility pole feature map).

[0125] In some embodiments, the loss function of the image segmentation classifier is expressed as:

[0126]

[0127] Where, is the loss value, 、 are all random numbers between 0 and 1. Pixel The predicted probability, N is the number of pixels, Pixel The true value of .

[0128] The Dt-Loss loss function dynamically adjusts sample weights to reduce the contribution of easily classified samples to gradient updates, thereby focusing more attention on difficult-to-classify samples. This mechanism is particularly suitable for scenarios with class imbalance and can effectively improve the model's segmentation performance for minority class samples. The Dt-Loss loss function optimizes the model's segmentation accuracy and robustness by measuring the similarity between the predicted results and the true labels and evaluating the overlap of binary images or binary masks. In addition, the Dt-Loss loss function introduces a random number mechanism during the weight adjustment process, further enhancing its adaptability to sample distribution and the model's generalization ability.

[0129] In some embodiments, based on the segmentation results of each utility pole image, detecting the main axis position information of the utility pole, and calculating the inclination of the utility pole based on the main axis position information of the utility pole to obtain the inclination of the utility pole for each image data includes:

[0130] Step S301: for each utility pole image segmentation result, extract the pixel connected area marked as a utility pole according to the utility pole image segmentation result.

[0131] The pixel-connected regions of the utility pole category are extracted from the utility pole image segmentation results. These regions are formed by connecting all pixels labeled as utility poles. These regions reflect the actual location and shape of the utility poles in the image.

[0132] Step S302: Perform foreground detection on the connected pixel area to obtain the area position of the utility pole, and crop the connected pixel area according to the area position of the utility pole to obtain an image of the utility pole area.

[0133] Contour detection and image processing technology are used to locate the specific position of the utility pole in the image. Based on the positioning results, the image of the local area where the utility pole is located is cropped.

[0134] Contour detection image processing technology extracts edge or boundary information from an image. It helps identify an object's shape, size, and location. Contours are extracted by setting a threshold to separate the image into the desired foreground and background. The specific process involves setting a global threshold and treating pixels above the threshold as foreground.

[0135] Step S303: Perform edge detection on the electric pole area image to obtain the edge contour of the electric pole.

[0136] Edge detection uses the Sobel edge detector, a discrete differential operator that combines Gaussian smoothing and differential derivation to approximate the first-order gradient of an image's grayscale. The Sobel operator effectively suppresses noise during edge detection and is particularly effective for images with grayscale gradients and high levels of noise. By performing Sobel edge detection on the utility pole area image, the edge contours of the poles can be clearly determined, providing accurate edge information for subsequent main axis position detection. The specific process is as follows:

[0137] The pole area image G is processed by the sobel operator and The gradient magnitudes in the directions are:

[0138]

[0139] Where, The image G of the electric pole area under the sobel operator The gradient magnitude in the direction, The image G of the electric pole area under the sobel operator The magnitude of the gradient in the direction.

[0140] The gradient amplitude of the pixel point of the utility pole area image G is the contour of the connecting line in the utility pole area image G. The gradient amplitude is:

[0141]

[0142] Where, is the gradient amplitude.

[0143] Step S304: Perform Hough line detection on the edge contour of the utility pole to obtain the main axis position information of the utility pole; wherein the main axis position information includes the top center point position and the bottom center point position of the utility pole.

[0144] The Hough Line Detection Algorithm (HLDA) is a widely used geometric shape detection technique in computer vision and image processing, primarily used to detect straight lines in images. The core idea of ​​the Hough transform is to convert geometric shapes in an image from a Cartesian coordinate system to a parameter space, thereby enabling efficient shape detection. For line detection, the Hough transform uses polar coordinates to represent lines, and its parametric equation is:

[0145]

[0146] Where, is the perpendicular distance from the origin to the line, Is a vertical line with The angle between the axes.

[0147] Step S305: performing coordinate calculation based on the top center point position and the bottom center point position of the utility pole to obtain the inclination of the utility pole.

[0148] Among them, the detected straight line parameters are inversely transformed back to the image space and the straight line is drawn, such as Figure 6 As shown in the figure, the inclination angle of the pole is calculated based on the coordinates of the top center point and the bottom center point of the pole. The specific formula is as follows:

[0149]

[0150] Where, and are the coordinates of the top and bottom center points of the pole, is the tilt angle (also the angle between the vertical line and angle between the axes).

[0151] In an exemplary embodiment, in order to more clearly illustrate a method for detecting the tilt of a power pole in a distribution network provided in an embodiment of the present application, the method for detecting the tilt of a power pole in a distribution network is specifically described below using a specific example.

[0152] In this example, a fixed camera collects 2,400 images of utility poles in a power distribution network area. The data covers various environmental conditions (such as daytime, nighttime, rainy, and snowy weather) and angles (front, side, and overhead). The captured images are then preprocessed using Gaussian filtering to reduce noise and improve image quality. The preprocessed images are then expanded or compressed into a 1280×1280×3 feature map, with a length of 1280 pixels, a width of 1280 pixels, and three channels.

[0153] The image segmentation and annotation tool Super Annotate was used to annotate the images in the distribution network utility pole image dataset. The images were categorized as "utility poles" or "background." The specific annotation method was to select the pixel areas of the utility poles in the image and label them as "utility poles," and label the pixel areas of the non-utility poles in the image as "background."

[0154] The constructed distribution network utility pole tilt detection image dataset was divided into a training set and a validation set at a ratio of 5:1. Specifically, the training set contained 2,400 utility pole images, while the validation set contained 2,000 images, and the validation set contained 400 images. This training and validation set was used to train the distribution network utility pole tilt detection model designed by the present invention. First, the model training parameters and hyperparameters were initialized.

[0155] This example highlights several key parameters: Batch Gradient Descent (BGD) is used as the optimizer, the total number of training epochs is set to 700, the batch size is 32, and the initial learning rate is 0.001. These parameters are dynamically adjusted based on the model's performance over multiple training sessions to ensure optimal performance. During training, the model gradually improves its segmentation capabilities by learning the characteristics of utility pole images, ultimately achieving high-precision detection of utility pole tilt. The training goal is to achieve high segmentation accuracy and tilt detection accuracy on the validation set, thereby ensuring the model's reliability and practicality in real-world scenarios.

[0156] After completing the basic parameter initialization of the model training, the model training begins. Input the utility pole feature map with a size of 1280×1280×3 The convolution kernel is 128, the size is 3×3, and the stride is 2, and the feature map of the electric pole with a size of 640×640×128 is obtained. .Will Perform two branch calculations. In the first branch, Input three parallel ARCM modules to obtain the utility pole feature map with the same size of 320×320×128 、 and .Will 、 and Perform feature addition (ADD) operation to obtain a utility pole feature map with a size of 320×320×128 .Will Input the ARCM module to obtain the utility pole feature map with a size of 160×160×128 .Will Perform a four-fold upsampling operation to obtain a utility pole feature map with a size of 640×640×128 In the first branch, Input the GCAM module to obtain the utility pole feature map with a size of 640×640×128 , the utility pole feature map with the same size of 640×640×128 and Perform channel superposition (Concat) operation to obtain the pole feature map of size 640×640×256 .Will Input the ARCM module to obtain the utility pole feature map with a size of 320×320×128 .Will Perform a four-fold upsampling operation to obtain a utility pole feature map with a size of 1280×1280×128 .Will Input to the image segmentation classifier and use As the loss function of segmentation, the image segmentation result of the utility pole is obtained and output The segmentation results include the category of the utility pole in the image, a pixel-level mask, and a confidence score. The category refers to the pole's label, which can be classified as "pole" or "background." The pixel-level mask is the predicted category for each pixel, represented as a two-dimensional matrix. The confidence score is the confidence score of the model's prediction for each pixel, indicating the reliability of the prediction.

[0157] The distribution network pole tilt detection model training in this example is passed The backpropagation mechanism of the loss function updates internal parameters. During algorithm training, gradient backpropagation gradually updates model weights in each iteration. With each round, the model converges and segmentation accuracy improves. After all training rounds, the algorithm uses the divided validation set data for dual validation using Intersection over Union (IoU) and Dice coefficient. Overall IoU and average Dice coefficient serve as evaluation criteria for the effectiveness of the algorithm training. Ultimately, the model outputs a segmentation result, including the area of ​​the utility pole in the image and its corresponding confidence level.

[0158] From the utility pole image segmentation results of the distribution network utility pole image segmentation model, pixel-connected regions classified as utility poles are extracted. Contour detection image processing techniques are then used to locate the specific locations of the utility poles within the image. Based on the localization results, the image of the local area containing the utility poles is cropped. Canny edge detection is performed on the cropped area to extract the edge contours of the utility poles. The Hough transform algorithm is then used to detect the main axis of the utility poles and to extract the top and bottom ends of the poles. Based on the top and bottom ends of the poles, the inclination angle of the poles is calculated. For example, the top endpoints of the poles are detected to be (100, 50) and the bottom endpoints are (120, 250).

[0159]

[0160] The results showed that the pole was tilted about 5.7 degrees to the right.

[0161] After successfully training the distribution network pole tilt detection model, it was integrated into the existing distribution network monitoring system, ensuring that the system could receive real-time video frames from the distribution network site. An alarm threshold of 0.5 was set in the system to capture the segmentation confidence level of the tilted pole area. Once the real-time segmentation confidence level of the tilted pole area exceeded the system's preset threshold, the monitoring system automatically locked onto and identified the target tilted pole area.

[0162] The system then activates a continuous monitoring mechanism to continuously track and analyze the tilt of the power poles. Furthermore, an intelligent alarm function is incorporated into the distribution network power pole tilt detection system. When the system detects that the tilt angle of a power pole exceeds the warning threshold, or if the tilt trend continues to intensify, the system immediately triggers an alarm, notifying relevant personnel to take timely intervention measures, effectively preventing potential safety incidents. This not only greatly improves the efficiency of on-site safety monitoring of the distribution network, but also ensures more comprehensive operational safety protection for power grid facilities.

[0163] Based on the same inventive concept, an embodiment of the present application further provides a distribution network pole tilt detection system for implementing the above-mentioned distribution network pole tilt detection method.

[0164] The implementation solution provided by the system to solve the problem is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more distribution network pole tilt detection system embodiments provided below can refer to the limitations of the distribution network pole tilt detection method above and will not be repeated here.

[0165] like Figure 7 As shown, an embodiment of the present application provides a power distribution network pole tilt detection system, comprising:

[0166] The image acquisition module 100 is used to acquire multiple image data of the power distribution network area to form an image data set; wherein the image data set includes annotation information of the utility poles;

[0167] The image segmentation module 200 is used to input the image data set into the utility pole image segmentation model to obtain the utility pole image segmentation results of each image data; wherein the utility pole image segmentation model is used to segment the image data according to the annotation information of the utility poles and locate the positions of the utility poles in the image data;

[0168] The inclination calculation module 300 is used to detect the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculate the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data;

[0169] The tilt detection training module 400 is used to train a deep neural network based on the image data set and the tilt of the utility poles in each image data to obtain a utility pole tilt detection model;

[0170] The inclination detection module 500 is used to perform electric pole inclination detection on the image data of the current detection period using the electric pole inclination detection model to obtain the electric pole inclination corresponding to the image data of the current detection period.

[0171] In some embodiments, the image acquisition module 100 is configured to:

[0172] Collecting multiple image data of the power distribution network area, wherein the image data includes regional image information of utility poles;

[0173] Performing denoising on multiple image data;

[0174] Segment and label the denoised image data to obtain the labeling information of the utility poles;

[0175] An image dataset is constructed based on multiple labeled image data.

[0176] In some embodiments, the image segmentation module 200 is configured to:

[0177] For each image data in the image dataset, perform a first dilated convolution operation on the image data to obtain a first utility pole feature map;

[0178] The first utility pole feature map is calculated through two branches respectively. In one branch, the first utility pole feature map is subjected to multiple parallel first hole residual convolution operations to obtain multiple second utility pole feature maps; in the other branch, the first utility pole feature map is subjected to a global context aggregation operation to obtain a third utility pole feature map;

[0179] Performing a feature addition operation on the plurality of second utility pole feature maps to obtain a fourth utility pole feature map, and performing a second dilated residual convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map;

[0180] Performing an upsampling operation on the fifth utility pole feature map to obtain a sixth utility pole feature map;

[0181] Perform a channel superposition operation on the third utility pole characteristic map and the sixth utility pole characteristic map to obtain a seventh utility pole characteristic map;

[0182] Performing a third dilated residual convolution operation on the seventh utility pole feature map to obtain an eighth utility pole feature map, and performing an upsampling operation on the eighth utility pole feature map to obtain a ninth utility pole feature map;

[0183] The ninth utility pole feature map is input into the image segmentation classifier for classification to obtain the utility pole image segmentation result.

[0184] In some embodiments, the process of the dilated residual convolution operation includes:

[0185] Perform a second dilated convolution operation on the first utility pole feature map to obtain a first utility pole sub-feature map;

[0186] Performing a batch normalization operation on the first utility pole feature map to obtain a second utility pole feature map;

[0187] The second electric pole characteristic graph is subjected to nonlinear processing by a Swish function to obtain a third electric pole characteristic graph;

[0188] Performing a channel superposition operation on the third utility pole feature map and the first utility pole feature map to obtain a fourth utility pole feature map;

[0189] Performing a third dilated convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map;

[0190] The fifth utility pole sub-feature map is normalized to obtain a sixth utility pole sub-feature map, and the sixth utility pole sub-feature map is nonlinearly processed using a Swish function to obtain a second utility pole feature map.

[0191] In some embodiments, the process of global context aggregation operation includes:

[0192] The first utility pole feature map is calculated in parallel on the three sub-branches. In each of the three sub-branches, a multi-level dilated convolution operation is performed on the first utility pole feature map. Then, a channel superposition operation is performed on the utility pole sub-feature maps obtained by the multi-level dilated convolution operation to obtain the seventh utility pole sub-feature maps corresponding to the three sub-branches. The scales of the dilated convolutions in the multi-level dilated convolution operation are different.

[0193] Performing a feature addition operation on the three seventh utility pole feature maps to obtain an eighth utility pole feature map;

[0194] The fourth dilated convolution operation is performed on the eighth utility pole sub-feature map to obtain the third utility pole feature map.

[0195] In some embodiments, the loss function of the image segmentation classifier is expressed as:

[0196]

[0197] Where, is the loss value, 、 are all random numbers between 0 and 1. Pixel The predicted probability, N is the number of pixels, Pixel The true value of .

[0198] In some embodiments, the inclination calculation module 300 is configured to:

[0199] For each utility pole image segmentation result, extract the pixel connected area marked as the utility pole according to the utility pole image segmentation result;

[0200] Perform foreground detection on the connected pixel area to obtain the regional position of the utility pole, and then crop the connected pixel area according to the regional position of the utility pole to obtain the utility pole area image;

[0201] Perform edge detection on the electric pole area image to obtain the edge contour of the electric pole;

[0202] Perform Hough line detection on the edge contour of the utility pole to obtain the main axis position information of the utility pole; wherein the main axis position information includes the top center point position and the bottom center point position of the utility pole;

[0203] The inclination of the pole is obtained by performing coordinate calculation based on the top center point position and the bottom center point position of the pole.

[0204] like Figure 8 As shown, an embodiment of the present application provides an electronic device, the electronic device 10 includes a memory 20 and a processor 30, the memory 20 stores a computer program, and when the computer program is executed by the processor 30, the processor 30 performs the steps of the distribution network pole tilt detection method in the above embodiment.

[0205] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the steps of the method for detecting the inclination of power poles in a distribution network as described in the above embodiment are implemented.

[0206] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, electronic devices, and computer storage media can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0207] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0208] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0209] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, electronic devices, computer storage media and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0210] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0211] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0212] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the method described in each embodiment of the present invention via a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0213] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting the inclination of electric poles in a distribution network, characterized in that: include: Acquire multiple image data of the power distribution network area to form an image dataset; wherein the image dataset includes annotation information of utility poles; Inputting the image data set into a utility pole image segmentation model to obtain utility pole image segmentation results for each of the image data; wherein the utility pole image segmentation model is used to segment the image data according to the annotation information of the utility poles and locate the positions of the utility poles in the image data; Detecting the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculating the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data; Training a deep neural network based on the image dataset and the inclination of the utility poles in each of the image data to obtain a utility pole inclination detection model; The utility pole tilt detection model is used to perform utility pole tilt detection on the image data of the current detection period to obtain the utility pole tilt corresponding to the image data of the current detection period.

2. The method for detecting the inclination of power poles in a power distribution network according to claim 1, characterized in that: The step of acquiring a plurality of image data of the distribution network area to form an image data set includes: Collecting a plurality of image data of the power distribution network area, wherein the image data includes regional image information of utility poles; performing denoising processing on the plurality of image data; Segmenting and labeling the denoised image data to obtain labeling information of the utility pole; The image data set is constructed based on a plurality of labeled image data.

3. The method for detecting the inclination of power poles in a power distribution network according to claim 1, characterized in that: Inputting the image data set into the utility pole image segmentation model to obtain the utility pole image segmentation results of each image data includes: For each image data in the image dataset, performing a first dilated convolution operation on the image data to obtain a first utility pole feature map; The first utility pole feature map is calculated through two branches respectively, wherein, in one branch, the first utility pole feature map is subjected to multiple parallel first dilated residual convolution operations to obtain multiple second utility pole feature maps; in the other branch, the first utility pole feature map is subjected to a global context aggregation operation to obtain a third utility pole feature map; Performing a feature addition operation on the plurality of second utility pole feature maps to obtain a fourth utility pole feature map, and performing a second dilated residual convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map; Performing an upsampling operation on the fifth utility pole feature map to obtain a sixth utility pole feature map; Performing a channel superposition operation on the third utility pole characteristic map and the sixth utility pole characteristic map to obtain a seventh utility pole characteristic map; Performing a third dilated residual convolution operation on the seventh utility pole feature map to obtain an eighth utility pole feature map, and performing an upsampling operation on the eighth utility pole feature map to obtain a ninth utility pole feature map; The ninth utility pole feature map is input into an image segmentation classifier for classification to obtain the utility pole image segmentation result.

4. The method for detecting the inclination of power poles in a power distribution network according to claim 3, characterized in that: The process of the dilated residual convolution operation includes: Performing a second dilated convolution operation on the first utility pole feature map to obtain a first utility pole sub-feature map; performing a batch normalization operation on the first utility pole sub-feature map to obtain a second utility pole feature map; Performing nonlinear processing on the second utility pole characteristic graph through a Swish function to obtain a third utility pole characteristic graph; Performing a channel superposition operation on the third utility pole sub-characteristic map and the first utility pole sub-characteristic map to obtain a fourth utility pole sub-characteristic map; Performing a third dilated convolution operation on the fourth utility pole feature map to obtain a fifth utility pole feature map; The fifth utility pole sub-feature map is normalized to obtain a sixth utility pole sub-feature map, and the sixth utility pole sub-feature map is nonlinearly processed using a Swish function to obtain the second utility pole feature map.

5. The method for detecting the inclination of power poles in a power distribution network according to claim 3, characterized in that: The process of the global context aggregation operation includes: The first utility pole feature map is calculated in parallel on three sub-branches, wherein in the three sub-branches, a multi-level dilated convolution operation is performed on the first utility pole feature map, and then a channel superposition operation is performed on the utility pole sub-feature maps obtained by the multi-level dilated convolution operation to obtain seventh utility pole sub-feature maps corresponding to the three sub-branches respectively; wherein the dilated convolutions in the multi-level dilated convolution operation have different scales; Performing a feature addition operation on the three seventh utility pole sub-feature maps to obtain an eighth utility pole feature map; A fourth dilated convolution operation is performed on the eighth utility pole sub-feature map to obtain the third utility pole feature map.

6. The method for detecting the inclination of power poles in a power distribution network according to claim 3, characterized in that: The loss function of the image segmentation classifier is expressed as: Where, is the loss value, 、 are all random numbers between 0 and 1. Pixel The predicted probability, N is the number of pixels, Pixel The true value of .

7. The method for detecting the inclination of power poles in a distribution network according to any one of claims 1 to 6, characterized in that: The detecting the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculating the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data includes: For each of the utility pole image segmentation results, extracting a connected region of pixels marked as a utility pole according to the utility pole image segmentation result; Performing foreground detection on the connected pixel region to obtain the regional position of the utility pole, and cropping the connected pixel region according to the regional position of the utility pole to obtain a utility pole regional image; Performing edge detection on the utility pole area image to obtain edge contours of the utility poles; Performing Hough line detection on the edge contour of the utility pole to obtain the main axis position information of the utility pole; wherein the main axis position information includes the top center point position and the bottom center point position of the utility pole; Coordinate calculation is performed based on the top center point position and the bottom center point position of the utility pole to obtain the inclination of the utility pole.

8. A distribution network electric pole tilt detection system, characterized in that: include: An image acquisition module is used to acquire multiple image data of the power distribution network area to form an image data set; wherein the image data set includes annotation information of the utility poles; An image segmentation module is configured to input the image data set into a utility pole image segmentation model to obtain a utility pole image segmentation result for each of the image data; wherein the utility pole image segmentation model is configured to segment the image data and locate the utility pole position in the image data based on the annotation information of the utility pole; an inclination calculation module, configured to detect the main axis position information of the utility pole according to the segmentation result of each utility pole image, and calculate the inclination of the utility pole according to the main axis position information of the utility pole to obtain the inclination of the utility pole of each image data; a tilt detection training module, configured to train a deep neural network based on the image dataset and the tilt of the utility poles in each of the image data to obtain a utility pole tilt detection model; The inclination detection module is used to perform electric pole inclination detection on the image data of the current detection period using the electric pole inclination detection model to obtain the electric pole inclination corresponding to the image data of the current detection period.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, wherein a computer program is stored in the memory. When the computer program is executed by the processor, the processor performs the steps of the method for detecting the inclination of a power pole in a distribution network according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the steps of the method for detecting the inclination of power poles in a distribution network as claimed in any one of claims 1 to 7 are implemented.

Images