A transmission line defect point monitoring system based on deep learning

Abstract

A power transmission line defect point monitoring system based on deep learning relates to the field of power grid inspection technology, comprising a monitoring center, wherein the monitoring center is communicatively connected with a power transmission line topology map module, a simulation module, a data acquisition module, an image analysis module and a defect point detection module; a power transmission line topology map of a target area and a meteorological feature prediction model are constructed, and meteorological feature labels are added to the power transmission line topology map; a transmission line dynamic resolution comparison table is constructed; video data of each line segment of a power transmission line to be detected is collected through an unmanned aerial vehicle monitoring terminal; several feature areas in the image of each line segment are extracted for displacement vector analysis; a line segment feature recognition model is constructed, feature areas corresponding to background displacement vectors are eliminated, and whether there are defect points in the transmission line is judged according to local displacement vectors corresponding to other feature areas and the line segment feature recognition model, so as to improve the safety and reliability of the line.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid inspection, and in particular to a transmission line defect point monitoring system based on deep learning. Background Art

[0002] The prior art CN117723876A "A method and system for detecting hidden dangers of overhead transmission lines" performs corona discharge detection on hidden dangers of transmission lines; uses frequency domain feature extraction technology to detect the discharge type at hidden dangers; uses a traveling wave positioning model to locate hidden dangers of transmission lines; and uses high-frequency current sensors to perform online monitoring of hidden dangers. The present invention identifies the type of defective discharge of transmission lines through comprehensive analysis of various time domain and frequency domain parameters, performs pattern recognition of line hidden dangers based on a large amount of statistical analysis based on the waveform characteristics of a single discharge pulse signal, determines the precise location of the hidden danger point through traveling wave positioning, and studies the pre-discharge monitoring and early warning technology of insulation defects of transmission lines to discover hidden dangers of defects in line operation in advance, eliminate the hidden dangers of insulation defects in time, and improve the active protection level of transmission lines.

[0003] The prior art CN115314406B "An intelligent defect detection method for transmission lines based on image analysis" obtains several groups of image information collection points with more similar factors by grouping the environment and specific positions of the transmission lines. Within the same group, the number of defects detected at the corresponding transmission line positions in normal and abnormal environments is counted to obtain the impact of the abnormal environment on the probability of defect occurrence, and the abnormal positions caused by the particularity of construction quality, product quality or installation position are found, and the sampling time interval is adjusted to ensure that positions with high probability of abnormal occurrence and important positions can obtain higher frequency image information sampling, and positions with low probability of abnormal occurrence and less important positions are assigned to lower frequency image information sampling. Under the condition of limited resources, the timeliness of defect monitoring is guaranteed, and the data processing capability and data transmission capability of the system are fully utilized.

[0004] With the continuous deepening of research on power technology, UAV intelligent inspection mode, image processing, deep learning algorithm, etc. in my country, intelligence has become the focus of future development and attention of transmission lines. At present, most power companies are still using manual inspection to monitor the defects of transmission lines and their power devices. This method has high requirements on the work efficiency, workload and misjudgment rate of inspectors. Especially under harsh conditions, manual inspection results often have problems such as low accuracy and poor reliability, which cannot meet the "intelligent" diagnosis design needs of transmission line defects and faults. Summary of the invention

[0005] In order to solve the above technical problems, the object of the present invention is to provide a transmission line defect point monitoring system based on deep learning, including a monitoring center, wherein the monitoring center is communicatively connected with a transmission line topology module, a simulation module, a data acquisition module, an image analysis module and a defect point detection module;

[0006] The transmission line topology map module is used to construct a transmission line topology map of the target area and a meteorological feature prediction model, and add meteorological feature labels to the transmission line topology map;

[0007] The simulation module is used to construct a dynamic resolution comparison table for transmission lines;

[0008] The data acquisition module is used to collect video data of each line segment of the power transmission line to be detected through the unmanned aerial vehicle monitoring terminal;

[0009] The image analysis module is used to extract several characteristic areas in each line segment image to perform displacement vector analysis and obtain background displacement vectors and local displacement vectors;

[0010] The defect point detection module is used to construct a line segment feature recognition model, remove the feature area corresponding to the background displacement vector, and determine whether there are defect points in the transmission line based on the local displacement vectors corresponding to other feature areas and the line segment feature recognition model.

[0011] Furthermore, the transmission line topology map module constructs a transmission line topology map of the target area and a meteorological feature prediction model, and the process of adding meteorological feature labels to the transmission line topology map includes:

[0012] The positions of power grid lines in the target area are obtained by GIS means, and a transmission line topology map in the target area is constructed according to the positions of the power grid lines. The positions of several pole towers in the power grid lines are used as the point structure of the transmission line topology map, and the transmission lines between the several pole tower positions are used as the edge structure of the transmission line topology map. A meteorological feature prediction model is constructed based on deep learning, and the meteorological features corresponding to each transmission line in the transmission line topology map are obtained through the meteorological feature prediction model, and corresponding meteorological feature labels are added to the corresponding edge structures of each transmission line in the transmission line topology map.

[0013] Furthermore, the process of constructing the dynamic resolution comparison table of the power transmission line by the simulation module includes:

[0014] Based on the idea of ​​imitation learning, a simulator is used to generate a simulated transmission line with several defect points under different meteorological characteristics, and the video data of the simulated transmission line is obtained by the UAV monitoring terminal using different shooting resolutions. The defect points of the simulated transmission line are obtained according to the video data, and the defect points are compared with several defect points of the simulated transmission line to obtain the defect detection accuracy corresponding to different shooting resolutions used by the UAV monitoring terminal under different meteorological characteristics. According to the defect detection accuracy corresponding to different shooting resolutions used by the UAV monitoring terminal under different meteorological characteristics, a dynamic resolution comparison table of the transmission line is constructed.

[0015] Furthermore, the process of the data acquisition module collecting video data of each line segment of the power transmission line to be detected through the unmanned aerial vehicle monitoring terminal includes:

[0016] The UAV monitoring terminal determines the shooting resolution of several line segments contained in each transmission line to be detected according to the transmission line dynamic resolution comparison table and the meteorological feature labels, and collects the video data of each line segment of the transmission line to be detected along the extension direction of the transmission line to be detected according to the shooting resolution.

[0017] Furthermore, the image analysis module extracts several characteristic regions in each line segment image to perform displacement vector analysis, and the process of obtaining background displacement vectors and local displacement vectors includes:

[0018] The video data of the route segment is converted into multiple grayscale images. The gradient features of each local area in each grayscale image are obtained by sliding windows. A two-dimensional coordinate system is constructed. The gradient features in each local area in each grayscale image are mapped in the two-dimensional coordinate system. The gradient features in each local area in each grayscale image are matched consistently. Local areas with common gradient features between each grayscale image are marked as feature areas. The displacement vectors of several feature areas between adjacent grayscale images are obtained. The occurrence probabilities of different displacement vectors in multiple grayscale images are obtained. The displacement vector corresponding to the highest occurrence probability is extracted and marked as the background displacement vector, and the other displacement vectors are marked as local displacement vectors.

[0019] Furthermore, the defect point detection module constructs a line segment feature recognition model, removes the feature area corresponding to the background displacement vector, and determines whether there are defect points in the transmission line according to the local displacement vectors corresponding to other feature areas and the line segment feature recognition model. The process includes:

[0020] Acquire a feature area corresponding to a background displacement vector in multiple frames of grayscale images, remove the feature area from each frame of grayscale images, acquire local displacement vectors corresponding to other feature areas in the multiple frames of grayscale images, and construct displacement trajectories of other feature areas in the multiple frames of grayscale images according to the local displacement vectors corresponding to other feature areas, construct a line segment feature recognition model based on deep learning, input the displacement trajectories corresponding to other feature areas into the line segment feature recognition model, output the feature area included in the transmission line in each frame of grayscale images and the vibration frequency and amplitude of the transmission line according to the line segment feature recognition model, preset a vibration frequency threshold and an amplitude threshold of the transmission line, compare the vibration frequency and amplitude of the transmission line with the vibration frequency threshold and the amplitude threshold respectively, if the vibration frequency is greater than the vibration frequency threshold, generate vibration frequency alarm information of the line segment, if the amplitude is greater than the amplitude threshold, generate amplitude alarm information of the line segment, if the vibration frequency is less than or equal to the vibration frequency threshold and the amplitude is less than or equal to the amplitude threshold, perform surface defect detection of the feature area included in the transmission line;

[0021] If the feature area included in the power transmission line does not exist in each frame of the grayscale image output by the line segment feature recognition model, the fracture alarm information of the line segment is generated.

[0022] Furthermore, the process of the defect point detection module performing surface defect detection on the characteristic area of ​​the power transmission line includes:

[0023] The grayscale co-occurrence matrix technology is used to obtain the grayscale co-occurrence matrix of the feature area included in the transmission line in each frame of grayscale image, and the texture eigenvalue is extracted according to the grayscale co-occurrence matrix of the feature area. The standard texture eigenvalue of the line segment and the similarity coefficient threshold are preset, and the texture eigenvalue of the feature area is compared with the standard texture eigenvalue to obtain the similarity coefficient of the feature area. If the similarity coefficient of the feature area is less than the similarity coefficient threshold, the surface defect alarm information of the line segment is generated.

[0024] Furthermore, the process of building a meteorological feature prediction model and a line segment feature recognition model based on deep learning includes:

[0025] The historical meteorological characteristic data corresponding to each transmission line in the target area is obtained by means of GIS, and the historical meteorological characteristic data is used as the test set and training set of the meteorological characteristic prediction model, and the historical line segment video data collected by the unmanned aerial vehicle monitoring terminal is used as the test set and training set of the line segment feature recognition model;

[0026] The training set of the meteorological feature prediction model and the training set of the line segment feature recognition model are respectively input into the meteorological feature prediction model and the line segment feature recognition model for training until the loss function training is stable, and the model parameters are saved. The meteorological feature prediction model and the line segment feature recognition model are respectively tested by the test set of the meteorological feature prediction model and the test set of the line segment feature recognition model until they meet the preset requirements, and the meteorological feature prediction model and the line segment feature recognition model are output.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. According to different meteorological characteristics, the UAV monitoring terminal uses the defect detection accuracy corresponding to different shooting resolutions to construct a dynamic resolution comparison table for transmission lines. The UAV monitoring terminal determines the shooting resolution of several line segments contained in each transmission line to be detected according to the dynamic resolution comparison table for transmission lines and meteorological characteristic labels, and collects video data of each line segment of the transmission line to be detected along the extension direction of the transmission line to be detected according to the shooting resolution, so as to ensure the stable operation of the transmission line monitoring system under harsh environmental conditions and avoid the adverse effects of system failures or false alarms on the operation of the power grid.

[0029] 2. Since drones need to perform transmission line monitoring tasks in various environments, foreground moving objects often appear in the background of the collected video images, such as vehicles, pedestrians, plants, birds, animals, etc. on the road. Therefore, the movement in the video sequence collected by the drone is a combination of global movement caused by background changes, local movement caused by foreground changes in the scene, and local movement of the transmission line. Since the speeds of global movement and local movement are usually different, the background area corresponding to the global movement is eliminated when calculating the local movement trajectory of the transmission line to avoid the influence of background information on the analysis of the local movement trajectory of the transmission line. At the same time, since the movement law of the foreground moving object is difficult to predict, the local movement caused by the foreground moving object cannot be eliminated according to the analysis speed, which will lead to a decrease in the analysis accuracy of the local movement of the transmission line, thereby affecting the accuracy and reliability of the local movement detection of the transmission line. Therefore, the present invention constructs a line segment feature recognition model based on deep learning, identifies the local movement trajectory through the line segment feature recognition model to distinguish the transmission line, and eliminates the influence of the foreground moving object on the local movement detection of the transmission line, thereby improving the safety and reliability of the line. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a schematic diagram of a power transmission line defect point monitoring system based on deep learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0032] like Figure 1 As shown, a power transmission line defect point monitoring system based on deep learning includes a monitoring center, and the monitoring center is communicatively connected with a transmission line topology module, a simulation module, a data acquisition module, an image analysis module and a defect point detection module;

[0033] The transmission line topology map module is used to construct a transmission line topology map of the target area and a meteorological feature prediction model, and add meteorological feature labels to the transmission line topology map;

[0034] The simulation module is used to construct a dynamic resolution comparison table for transmission lines;

[0035] The data acquisition module is used to collect video data of each line segment of the power transmission line to be detected through the unmanned aerial vehicle monitoring terminal;

[0036] The image analysis module is used to extract several characteristic areas in each line segment image to perform displacement vector analysis and obtain background displacement vectors and local displacement vectors;

[0037] The defect point detection module is used to construct a line segment feature recognition model, remove the feature area corresponding to the background displacement vector, and determine whether there are defect points in the transmission line based on the local displacement vectors corresponding to other feature areas and the line segment feature recognition model.

[0038] It should be further explained that, in the specific implementation process, the transmission line topology map module constructs the transmission line topology map of the target area and the meteorological feature prediction model, and the process of adding meteorological feature labels to the transmission line topology map includes:

[0039] The positions of power grid lines in the target area are obtained by GIS means, and a transmission line topology map in the target area is constructed according to the positions of the power grid lines. The positions of several towers in the power grid lines are used as the point structure of the transmission line topology map, and the transmission lines between the several tower positions are used as the edge structure of the transmission line topology map. A meteorological feature prediction model is constructed based on deep learning. The meteorological features corresponding to each transmission line in the transmission line topology map are obtained through the meteorological feature prediction model. The meteorological features include wind speed, wind direction, precipitation, etc., and corresponding meteorological feature labels are added to the corresponding edge structures of each transmission line in the transmission line topology map.

[0040] It should be further explained that, in the specific implementation process, the process of the simulation module constructing the dynamic resolution comparison table of the transmission line includes:

[0041] Based on the idea of ​​imitation learning, a simulator is used to generate a simulated transmission line with several defect points under different meteorological characteristics, and the video data of the simulated transmission line is obtained by the UAV monitoring terminal using different shooting resolutions. The defect points of the simulated transmission line are obtained according to the video data, and the defect points are compared with several defect points of the simulated transmission line to obtain the defect detection accuracy corresponding to different shooting resolutions used by the UAV monitoring terminal under different meteorological characteristics. According to the defect detection accuracy corresponding to different shooting resolutions used by the UAV monitoring terminal under different meteorological characteristics, a dynamic resolution comparison table of the transmission line is constructed.

[0042] It should be further explained that, in the specific implementation process, the defective points of the transmission line refer to various problems or abnormal positions that may exist on the transmission line, which may affect the safety and reliability of the transmission line. Some common defective points of the transmission line include: cracks on the surface of the conductor, which may lead to insulation degradation or failure; conductor breakage or excessive amplitude or too fast vibration frequency, which may affect the power transmission efficiency or cause a short circuit in the line; tree or foreign object intrusion: tree growth or other foreign object intrusion leads to poor contact with the transmission line, which may cause a short circuit or failure.

[0043] It should be further explained that, in the specific implementation process, the process of the data acquisition module collecting video data of each line segment of the transmission line to be detected through the UAV monitoring terminal includes:

[0044] The UAV monitoring terminal determines the shooting resolution of several line segments contained in each transmission line to be detected according to the transmission line dynamic resolution comparison table and the meteorological feature labels, and collects the video data of each line segment of the transmission line to be detected along the extension direction of the transmission line to be detected according to the shooting resolution.

[0045] It should be further explained that, in the specific implementation process, the image analysis module extracts several characteristic areas in each line segment image to perform displacement vector analysis, and the process of obtaining the background displacement vector and the local displacement vector includes:

[0046] The video data of the route segment is converted into multiple grayscale images, and the gradient features of each local area in each grayscale image are obtained by sliding windows. The gradient feature set includes the grayscale gradient direction and grayscale gradient size of the pixels around the central pixel of the local area. The gradient size refers to the gradient amplitude or intensity calculated around the central pixel, which reflects the intensity or change rate of the grayscale change of the image near the point. A larger gradient size usually indicates that the image has a significant grayscale change near the point, while a smaller gradient size may indicate a relatively gentle grayscale change. The grayscale gradient direction represents eight directions of the central pixel, up, down, left, right and diagonal. A two-dimensional coordinate system is constructed, and the gradient features in each local area in each grayscale image are mapped in the two-dimensional coordinate system. The gradient features in each local area in each grayscale image are matched consistently, and the local areas with common gradient features between each grayscale image are marked as feature areas. The displacement vectors of several feature areas between adjacent grayscale images are obtained, and the occurrence probabilities of different displacement vectors in multiple grayscale images are obtained. The displacement vector corresponding to the highest occurrence probability is extracted and marked as the background displacement vector, and the other displacement vectors are marked as local displacement vectors.

[0047] It should be further explained that, since the UAV needs to perform the task of power transmission line monitoring in various environments, foreground moving objects often appear in the background of the collected video images, such as vehicles, pedestrians, plants, birds, animals, etc. on the road. Therefore, the motion in the video sequence collected by the UAV is a combination of global motion caused by background changes, local motion caused by foreground changes in the scene, and local motion of the transmission line. Since the speeds of global motion and local motion are usually different, the background area corresponding to the global motion is eliminated when calculating the local motion trajectory of the transmission line to avoid the influence of background information on the analysis of the local motion trajectory of the transmission line. At the same time, since the motion law of the foreground moving object is difficult to predict, the local motion caused by the foreground moving object cannot be eliminated according to the analysis speed, which will lead to a decrease in the analysis accuracy of the local motion of the transmission line, thereby affecting the accuracy and reliability of the local motion detection of the transmission line. Therefore, the present invention constructs a line segment feature recognition model based on deep learning, identifies the local motion trajectory through the line segment feature recognition model to distinguish the transmission line, and eliminates the influence of the foreground moving object on the local motion detection of the transmission line.

[0048] It should be further explained that, in the specific implementation process, the defect point detection module constructs a line segment feature recognition model, removes the feature area corresponding to the background displacement vector, and determines whether there are defect points in the transmission line according to the local displacement vectors corresponding to other feature areas and the line segment feature recognition model. The process includes:

[0049] Acquire a feature area corresponding to a background displacement vector in multiple frames of grayscale images, remove the feature area from each frame of grayscale images, acquire local displacement vectors corresponding to other feature areas in the multiple frames of grayscale images, and construct displacement trajectories of other feature areas in the multiple frames of grayscale images according to the local displacement vectors corresponding to other feature areas, construct a line segment feature recognition model based on deep learning, input the displacement trajectories corresponding to other feature areas into the line segment feature recognition model, output the feature area included in the transmission line in each frame of grayscale images and the vibration frequency and amplitude of the transmission line according to the line segment feature recognition model, preset a vibration frequency threshold and an amplitude threshold of the transmission line, compare the vibration frequency and amplitude of the transmission line with the vibration frequency threshold and the amplitude threshold respectively, if the vibration frequency is greater than the vibration frequency threshold, generate vibration frequency alarm information of the line segment, if the amplitude is greater than the amplitude threshold, generate amplitude alarm information of the line segment, if the vibration frequency is less than or equal to the vibration frequency threshold and the amplitude is less than or equal to the amplitude threshold, perform surface defect detection of the feature area included in the transmission line;

[0050] If the feature area included in the power transmission line does not exist in each frame of the grayscale image output by the line segment feature recognition model, the fracture alarm information of the line segment is generated.

[0051] It should be further explained that, in a specific implementation process, the process of the defect point detection module performing surface defect detection on the characteristic area of ​​the transmission line includes:

[0052] The grayscale co-occurrence matrix technology is used to obtain the grayscale co-occurrence matrix of the feature area included in the transmission line in each frame of grayscale image, and the texture eigenvalue is extracted according to the grayscale co-occurrence matrix of the feature area. The standard texture eigenvalue of the line segment and the similarity coefficient threshold are preset, and the texture eigenvalue of the feature area is compared with the standard texture eigenvalue to obtain the similarity coefficient of the feature area. If the similarity coefficient of the feature area is less than the similarity coefficient threshold, the surface defect alarm information of the line segment is generated.

[0053] It should be further explained that, in a specific implementation process, the process of comparing the texture feature value of the feature area with the standard texture feature value for similarity includes:

[0054] The gray-level co-occurrence matrix around each central pixel in the local area is calculated by sliding the window. The specific details include quantizing the gray-level of the surface image data into discrete gray-levels, dividing the gray-level value range into several levels, for example, an 8-bit image can be divided into 16, 32, and 64 levels, and defining the parameters required for the gray-level co-occurrence matrix, including distance (d) and direction (θ), which are determined according to the actual application requirements. The specific steps will not be repeated here. Subsequently, based on the calculated gray-level co-occurrence matrix, a series of texture feature values ​​are extracted to construct a texture feature set. The texture features in the texture feature set include: contrast, correlation, energy, homogeneity, entropy, and inverse variance.

[0055] The calculation formula for obtaining the similarity coefficient between the texture feature value of the feature area and the standard texture feature value is:

[0056]

[0057] Among them, z represents the similarity coefficient between the texture eigenvalue of the feature area and the standard texture eigenvalue, d represents the contrast included in the texture eigenvalue, dq represents the contrast included in the standard texture eigenvalue, w represents the energy included in the texture eigenvalue, wd represents the energy included in the standard texture eigenvalue, e represents the entropy included in the texture eigenvalue, eq represents the entropy included in the standard texture eigenvalue, r represents the homogeneity included in the texture eigenvalue, rq represents the homogeneity included in the standard texture eigenvalue, u represents the inverse variance included in the texture eigenvalue, uq represents the inverse variance included in the standard texture eigenvalue, and a1, a2, a3, a4 and a5 represent weight factors.

[0058] It should be further explained that, in the specific implementation process, the process of building a meteorological feature prediction model and a line segment feature recognition model based on deep learning includes:

[0059] The historical meteorological characteristic data corresponding to each transmission line in the target area is obtained by means of GIS, and the historical meteorological characteristic data is used as the test set and training set of the meteorological characteristic prediction model, and the historical line segment video data collected by the unmanned aerial vehicle monitoring terminal is used as the test set and training set of the line segment feature recognition model;

[0060] The training set of the meteorological feature prediction model and the training set of the line segment feature recognition model are respectively input into the meteorological feature prediction model and the line segment feature recognition model for training until the loss function training is stable, and the model parameters are saved. The meteorological feature prediction model and the line segment feature recognition model are respectively tested by the test set of the meteorological feature prediction model and the test set of the line segment feature recognition model until they meet the preset requirements, and the meteorological feature prediction model and the line segment feature recognition model are output.

[0061] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A power transmission line defect point monitoring system based on deep learning, comprising a monitoring center, characterized in that: The monitoring center is communicatively connected to a transmission line topology module, a simulation module, a data acquisition module, an image analysis module and a defect point detection module; The transmission line topology map module is used to construct a transmission line topology map of the target area and a meteorological feature prediction model, and add meteorological feature labels to the transmission line topology map; The simulation module is used to construct a dynamic resolution comparison table for transmission lines; The data acquisition module is used to collect video data of each line segment of the power transmission line to be detected through the unmanned aerial vehicle monitoring terminal; The image analysis module is used to extract several characteristic areas in each line segment image to perform displacement vector analysis and obtain background displacement vectors and local displacement vectors; The defect point detection module is used to construct a line segment feature recognition model, remove the feature area corresponding to the background displacement vector, and determine whether there are defect points in the transmission line based on the local displacement vectors corresponding to other feature areas and the line segment feature recognition model.

2. A power transmission line defect point monitoring system based on deep learning according to claim 1, characterized in that: The transmission line topology module constructs a transmission line topology map of the target area and a meteorological feature prediction model. The process of adding meteorological feature labels to the transmission line topology map includes: The positions of power grid lines in the target area are obtained by GIS means, and a transmission line topology map in the target area is constructed according to the positions of the power grid lines. The positions of several pole towers in the power grid lines are used as the point structure of the transmission line topology map, and the transmission lines between the several pole tower positions are used as the edge structure of the transmission line topology map. A meteorological feature prediction model is constructed based on deep learning, and the meteorological features corresponding to each transmission line in the transmission line topology map are obtained through the meteorological feature prediction model, and corresponding meteorological feature labels are added to the corresponding edge structures of each transmission line in the transmission line topology map.

3. A power transmission line defect point monitoring system based on deep learning according to claim 2, characterized in that: The process of constructing the dynamic resolution comparison table of the power transmission line by the simulation module includes: Based on the idea of ​​imitation learning, a simulator is used to generate a simulated transmission line with several defect points under different meteorological characteristics, and the video data of the simulated transmission line is obtained by the UAV monitoring terminal using different shooting resolutions. The defect points of the simulated transmission line are obtained according to the video data, and the defect points are compared with several defect points of the simulated transmission line to obtain the defect detection accuracy corresponding to different shooting resolutions used by the UAV monitoring terminal under different meteorological characteristics. According to the defect detection accuracy corresponding to different shooting resolutions used by the UAV monitoring terminal under different meteorological characteristics, a dynamic resolution comparison table of the transmission line is constructed.

4. A power transmission line defect point monitoring system based on deep learning according to claim 3, characterized in that: The process of the data acquisition module collecting video data of each line segment of the power transmission line to be detected through the UAV monitoring terminal includes: The UAV monitoring terminal determines the shooting resolution of several line segments contained in each transmission line to be detected according to the transmission line dynamic resolution comparison table and the meteorological feature labels, and collects the video data of each line segment of the transmission line to be detected along the extension direction of the transmission line to be detected according to the shooting resolution.

5. A power transmission line defect point monitoring system based on deep learning according to claim 4, characterized in that: The image analysis module extracts several characteristic areas in each line segment image to perform displacement vector analysis, and the process of obtaining background displacement vectors and local displacement vectors includes: The video data of the route segment is converted into multiple grayscale images. The gradient features of each local area in each grayscale image are obtained by sliding windows. A two-dimensional coordinate system is constructed. The gradient features in each local area in each grayscale image are mapped in the two-dimensional coordinate system. The gradient features in each local area in each grayscale image are matched consistently. Local areas with common gradient features between each grayscale image are marked as feature areas. The displacement vectors of several feature areas between adjacent grayscale images are obtained. The occurrence probabilities of different displacement vectors in multiple grayscale images are obtained. The displacement vector corresponding to the highest occurrence probability is extracted and marked as the background displacement vector, and the other displacement vectors are marked as local displacement vectors.

6. A power transmission line defect point monitoring system based on deep learning according to claim 5, characterized in that: The defect point detection module constructs a line segment feature recognition model, removes the feature area corresponding to the background displacement vector, and determines whether there are defect points in the transmission line according to the local displacement vectors corresponding to other feature areas and the line segment feature recognition model. The process includes: Acquire a feature area corresponding to a background displacement vector in multiple frames of grayscale images, remove the feature area from each frame of grayscale images, acquire local displacement vectors corresponding to other feature areas in the multiple frames of grayscale images, and construct displacement trajectories of other feature areas in the multiple frames of grayscale images according to the local displacement vectors corresponding to other feature areas, construct a line segment feature recognition model based on deep learning, input the displacement trajectories corresponding to other feature areas into the line segment feature recognition model, output the feature area included in the transmission line in each frame of grayscale images and the vibration frequency and amplitude of the transmission line according to the line segment feature recognition model, preset a vibration frequency threshold and an amplitude threshold of the transmission line, compare the vibration frequency and amplitude of the transmission line with the vibration frequency threshold and the amplitude threshold respectively, if the vibration frequency is greater than the vibration frequency threshold, generate vibration frequency alarm information of the line segment, if the amplitude is greater than the amplitude threshold, generate amplitude alarm information of the line segment, if the vibration frequency is less than or equal to the vibration frequency threshold and the amplitude is less than or equal to the amplitude threshold, perform surface defect detection of the feature area included in the transmission line; If the feature area included in the power transmission line does not exist in each frame of the grayscale image output by the line segment feature recognition model, the fracture alarm information of the line segment is generated.

7. A power transmission line defect point monitoring system based on deep learning according to claim 6, characterized in that: The process of the defect point detection module performing surface defect detection on the characteristic area of ​​the power transmission line includes: The grayscale co-occurrence matrix technology is used to obtain the grayscale co-occurrence matrix of the feature area included in the transmission line in each frame of grayscale image, and the texture eigenvalue is extracted according to the grayscale co-occurrence matrix of the feature area. The standard texture eigenvalue of the line segment and the similarity coefficient threshold are preset, and the texture eigenvalue of the feature area is compared with the standard texture eigenvalue to obtain the similarity coefficient of the feature area. If the similarity coefficient of the feature area is less than the similarity coefficient threshold, the surface defect alarm information of the line segment is generated.

8. A power transmission line defect point monitoring system based on deep learning according to claim 7, characterized in that: The process of building a meteorological feature prediction model and a line segment feature recognition model based on deep learning includes: The historical meteorological characteristic data corresponding to each transmission line in the target area is obtained by means of GIS, and the historical meteorological characteristic data is used as the test set and training set of the meteorological characteristic prediction model, and the historical line segment video data collected by the unmanned aerial vehicle monitoring terminal is used as the test set and training set of the line segment feature recognition model; The training set of the meteorological feature prediction model and the training set of the line segment feature recognition model are respectively input into the meteorological feature prediction model and the line segment feature recognition model for training until the loss function training is stable, and the model parameters are saved. The meteorological feature prediction model and the line segment feature recognition model are respectively tested by the test set of the meteorological feature prediction model and the test set of the line segment feature recognition model until they meet the preset requirements, and the meteorological feature prediction model and the line segment feature recognition model are output.

Images