Detection method of power transmission line video image monitoring device

Abstract

The invention provides a detection method of a power transmission line video image monitoring device, and belongs to the field of power transmission line detection. The problem of low power transmission line detection efficiency is solved. The method comprises the following steps: acquiring an original video, and training a recognition model; acquiring a monitoring video, and detecting the component by using the identification model; judging whether the power transmission line has a safety accident or not according to the monitoring video, and if so, feeding back; if not, judging whether the swing amplitude of the power transmission line is safe or not, and if not, feeding back; if so, acquiring power data of the upstream end and the downstream end of the power transmission line, and judging whether the power transmission line has a safety accident; if yes, feeding back; if not, continuing monitoring; monitoring an area where a safety accident occurs in the power transmission line until part replacement and maintenance are completed in the area; according to the invention, the monitoring video of the power transmission line is acquired, analyzed and processed, the area where the safety accident occurs on the power transmission line is determined, and the detection efficiency is improved.

Description

Technical Field

[0001] This invention relates to a detection method for a video image monitoring device for power transmission lines, belonging to the field of power transmission line detection. Background Technology

[0002] Existing methods for testing transmission lines have the following shortcomings: The lag in problem detection: In the traditional transmission line inspection mode, problems are mainly detected by regular manual inspections. However, there are time intervals in manual inspections. During this period, if there is damage, missing parts or safety accidents on the line, they cannot be detected in time. It is impossible to arrive at the scene as soon as the fault occurs, which delays the repair opportunity and expands the scope of power outage and losses.

[0003] Insufficient accuracy of manual inspection: During manual inspection, due to factors such as the inspector's experience, mental state, and visual fatigue, oversights and misjudgments may occur. Some minor component damage or early safety hazards may not be accurately identified, leading to the gradual deterioration of the problem.

[0004] High manpower input and low efficiency: Traditional manual inspection requires a large amount of manpower, and inspectors need to spend a lot of time traveling and conducting on-site inspections; at the same time, frequent on-site inspections also expose inspectors to higher safety risks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a detection method for a video image monitoring device for power transmission lines, thereby solving the problem of low detection efficiency for power transmission lines.

[0006] To achieve the above objectives, the present invention provides a detection method for a power transmission line video image monitoring device, comprising: Adjust the tilt angle of the solar panel of the monitoring device; acquire the original video and perform two-stage training on the basic model to obtain the recognition model; acquire the monitoring video of the transmission line and use the recognition model to determine whether there is any damage or missing parts in the monitoring video; if so, provide feedback. The system determines whether there is a safety accident on the transmission line based on the monitoring video. If there is, feedback is provided. If not, the system analyzes the monitoring video to determine whether the swing amplitude of the transmission line is safe. If not, feedback is provided. If safe, the system acquires the power data of the upstream and downstream ends of the transmission line in real time to make a second determination of whether there is a safety accident on the transmission line. If there is, feedback is provided and the monitoring position of the monitoring device is adjusted. If not, monitoring continues. The monitoring device for power transmission lines is divided into two parts: one part monitors areas in the power transmission line where components are damaged, missing, or there are safety accidents, until the components in that area are replaced and repaired; the other part detects normal areas on the power transmission line and reports any abnormal conditions.

[0007] Furthermore, the steps for constructing the recognition model are as follows: The original video is processed by frame extraction. The visible components of the transmission line in each frame are defined by bounding boxes, and the category and status of the visible components in each frame, as well as the structural relationship between the components, are labeled. Select a base model. In the first stage, train the model's component detection capability; in the second stage, train the model's component state judgment capability; after training, the recognition model is obtained. Real-time acquisition of monitoring videos of power transmission lines, and use of recognition models to determine whether there are damaged or missing components in the monitoring videos; If such damage or loss occurs, the location of the transmission line with the damaged or missing component is recorded and reported to the relevant authorities; if not, monitoring continues.

[0008] Furthermore, the training steps for the first phase are as follows: During the forward propagation process, the backbone network extracts features from the input image, performs upsampling and lateral connections through the feature pyramid, and generates multi-level feature maps that fuse semantic information. On each feature map, dense candidate bounding boxes are generated through the anchor box mechanism. For each candidate box, its features are extracted using ROI Align; the features of each candidate box are classified into multiple categories using fully connected layers and convolutional layers, and the probability distribution of each category is output; the offset of the bounding box relative to the candidate box is predicted using fully connected layers and convolutional layers, and the center coordinates of the predicted box are calculated using the center coordinate decoding formula based on the offset of the bounding box coordinates. Using nonmaximum suppression, within a given prediction box, the bounding box with the highest confidence and the lowest overlap with other boxes is selected as the final output prediction bounding box. During backpropagation, localization loss, classification loss, and confidence loss are designed to update the model parameters. If the model training result is that the model can accurately identify the location of each component in the transmission line and identify the category of each component under different environments, different lighting conditions, and certain background interference, then the second stage of training is performed; otherwise, the first stage of training continues.

[0009] Furthermore, the steps for updating the model parameters are as follows: In the first stage of training, the parameters of the convolutional layers near the input end in the backbone network are updated first, and then the relevant parameters of the backbone network in extracting high-level features and the output of the detection head are updated to ensure that there is no localization deviation between the predicted bounding boxes output by the model and the real bounding boxes. Gradually reduce the weight coefficient of the model in the localization loss, increase the weight coefficient of the model in the classification and confidence loss, and adjust the relevant parameters for extracting high-level semantic features in the backbone network so that the model can capture the subtle differences in the morphology and structure of different components and distinguish the transmission line from the background. As the predicted bounding box positions output by the model gradually converge to the true bounding box positions, a data augmentation strategy is introduced to train the model and maintain its generalization ability.

[0010] Furthermore, the training steps for the second phase are as follows: Without changing the model parameters, the model is trained by comparing normal and damaged images of various components that make up the transmission line. Before training, the damage type of the damaged images of the components is labeled so that the model can determine whether the components are normal or damaged. Based on the structural relationships between components and the design specifications and operation and maintenance rules of transmission lines, the connection relationships of transmission lines are defined, and then a knowledge graph is constructed in the form of structured data; based on the existing entities and relationships in the knowledge graph, the structural rules of transmission lines are obtained. During training, the original video is used as video clips as model input. Each video clip is processed frame by frame to obtain the position of each type of component in the current frame. In the time dimension, the positions of each type of component in the video clip are correlated, and the occurrence position of the same component in consecutive frames is tracked to obtain the time trajectory and confidence change of each component. A structural diagram of the transmission line is generated based on the current video clip. The nodes in the structural diagram represent components that have been detected or should theoretically exist.

[0011] Furthermore, the subsequent training steps for the second phase are as follows: First, using original videos of normal transmission lines as training samples, the thresholds for judging structural consistency rules in the model are gradually adjusted. Then, using original videos with missing components as training samples, the differences between the output structural diagrams of transmission lines with missing components and the structural rules of the transmission lines are compared. If a component is marked as present in the structural rules, but is not detected after exceeding a set number of consecutive frames, the probability of its presence continues to decrease, and the component is marked as having a high probability of being missing. During this process, the threshold used in the model to judge the structural consistency rules gradually converges, and the weights of the consecutive undetected frame threshold, the component presence probability, and the structural matching degree threshold are updated, so that the model can judge the normal, damaged, or missing state of the component and obtain the recognition model.

[0012] Furthermore, the steps to determine whether the swing amplitude of a transmission line is safe are as follows: The surveillance video is time-synchronized and frame-sampling is performed to unify the frame rate, and electronic image stabilization and enhancement are performed on videos with jitter. The recognition model is used to retrieve and locate power transmission lines in the video, determining the spatial position of the power transmission lines in each frame of the surveillance video and extracting the movement trajectory of the lines. Several discrete feature points are uniformly selected along the direction of the conductor. As the monitoring video changes frame by frame over time, these feature points are tracked frame by frame to obtain the trajectory of the two-dimensional pixel coordinates of each feature point in the monitoring video over time. Time-domain analysis of the change trajectory of feature points: Calculate the average position of each feature point within the time window and use it as the reference position of the point in a windless or balanced state; compare the actual position of each feature point at each moment with the reference position to obtain the pixel displacement sequence of the feature point over time. Analysis of the swing characteristics of the conductor: Extract the maximum swing amplitude x from the displacement sequence of feature points. （max） And calculate the root mean square value of the swing amplitude corresponding to each time window.

[0013] Further steps to determine whether the sway amplitude of the transmission line is safe are as follows: Determine x （max） With maximum permissible swing x （safe） Size relationship: If x （max） ≥x （safe） If so, the location of the power transmission line should be immediately reported to the relevant departments; If x （max） <x （safe） If so, in the monitoring video, determine whether the root mean square value of the swing amplitude corresponding to each time window shows an increasing trend; if so, immediately report the location of the power transmission line to the relevant departments. If not, frequency domain analysis is performed on the displacement sequence of the feature points, and the dominant frequency characteristics of the conductor's oscillation are extracted through Fourier transform to determine the potential resonance risk of the conductor's vibration. Displacement sequence D of feature points （fea） Performing a discrete Fourier transform yields a frequency domain complex sequence X containing the imaginary part. （fea） ; Take sequence X （fea） Square the modulus to obtain the power spectrum; find the frequency index k corresponding to the maximum power value in the power spectrum. （max） Based on the sampling frequency of the surveillance video, the dominant frequency f is calculated.（dom） ; Determine the dominant frequency f of the oscillation of the transmission line. （dom） Whether it is close to a certain natural frequency of the transmission line; If so, it indicates that the power transmission line in the area is at risk of resonance, and the actual area of ​​the power transmission line should be the key area for monitoring by the monitoring device. If not, then do not process it.

[0014] Furthermore, the steps for a secondary assessment of whether a safety incident has occurred on a transmission line are as follows: The system acquires real-time power data from the upstream and downstream ends of the transmission line to determine if there is a safety accident. If so, it provides feedback and adjusts the monitoring position of the monitoring device; if not, it continues monitoring. Introducing voltage phase angle change weights λ for active and reactive power. （P） and λ （Q） Voltage difference change weight μ （P） and μ （Q） Establish a mapping relationship between voltage, current, and power at both ends of the transmission line, and iteratively correct the power balance at both ends of the transmission line. Obtain the actual voltage, current and power received at the downstream end of the transmission line, and calculate the ideal transmission power when transmitting power from the upstream end to the downstream end based on the equivalent resistance, reactance, conductance and susceptance of the transmission line. Obtain the actual transmission power at the upstream end of the transmission line and compare it with the ideal transmission power: If the actual transmission power at the upstream end is equal to or approximately equal to the ideal transmission power, then the transmission line is considered to be normal. If there is a deviation between the actual transmission power and the ideal transmission power at the upstream end, the transmission power at the upstream end of the transmission line will be adjusted to correct the phase angle difference of the terminal voltage at the upstream and downstream ends.

[0015] Furthermore, the steps for correcting the voltage phase angle difference are as follows: Step a1: Calculate the power difference ΔP between the ideal transmission power and the actual transmission power at the upstream end. （tra） ; Step a2: Calculate the change in phase angle difference δ, Δδ; Step a3: Adjust the phase angle difference δ （1） According to δ （1） And the mapping of voltage, current and power at both ends of the transmission line, and recalculate the active power when transmitting power from the upstream end to the downstream end; Step a4: Repeat steps a3 to a4 until the phase angle difference δ in the i-th iteration is reached. （i） The phase angle difference δ with the (i-1)th iteration （i-1） Same, and δ （i）The corresponding active power is the same as the ideal transmission power; According to δ （i） Adjust the voltage phase angle difference between the upstream and downstream ends of the transmission line; and determine whether the actual active power received at the downstream end remains unchanged. If it remains unchanged, it indicates that the transmission line is normal; If the actual active power received at the downstream end decreases, it indicates that an overheating accident has occurred in the transmission line. If the actual active power received downstream increases, it indicates that a short circuit has occurred in the transmission line.

[0016] Compared with the prior art, the beneficial effects of the present invention are: High real-time performance: This invention can realize real-time monitoring of transmission lines. By acquiring monitoring video image information on the transmission lines and using recognition models, the image content can be quickly analyzed. Once abnormal situations such as component damage, missing parts or safety accidents are detected, feedback can be given immediately, enabling maintenance personnel to grasp the line status at the first time.

[0017] Wide coverage: This invention can monitor a large area of ​​power transmission lines, especially in mountainous and jungle areas with complex terrain and inconvenient transportation. Video surveillance can overcome geographical barriers, cover places that are difficult for manual personnel to reach, and reduce blind spots in inspections.

[0018] Improved detection accuracy: This invention can directly and accurately determine whether components are damaged or missing through monitoring video. For line faults that cannot be detected by video, this invention acquires the power data of the upstream and downstream ends of each transmission line. By analyzing the power data, it can determine whether some safety accidents that cannot be detected by video have occurred on the transmission line, which has high monitoring comprehensiveness and accuracy. Attached Figure Description

[0019] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram of the monitoring device of the present invention; Figure 3 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Please see Figure 1 and Figure 2A detection method for a power transmission line video image monitoring device includes: Step S1: Adjust the tilt angle of the solar panel of the monitoring device (so that the surface of the solar panel is perpendicular to the sunlight and the orientation of the solar panel is consistent with the direction of the sun); acquire the original video and perform two-stage training on the basic model to obtain the recognition model; acquire the monitoring video of the transmission line and use the recognition model to determine whether there is any damage or missing parts in the monitoring video; if so, provide feedback. The specific steps of step S1 are as follows: Acquire monitoring videos of power transmission lines under different seasons, lighting conditions, and weather conditions, as well as monitoring videos of various components on the power transmission lines, as the original videos for model training. Before training, the original video is processed by frame extraction (usually generating an image sequence at a frequency of 1 to 5 frames per second). The visible components of the transmission line in each frame are defined by bounding boxes (such as transmission conductors, insulator strings, vibration dampers, shielding rings, etc.), and the category and status of the visible components in each frame are labeled. The status is divided into: normal, damaged, and missing. Between components with structural relationships, the structural relationships between the components are labeled (e.g., the connection relationship between "insulator string" and "tower crossarm", the subordinate relationship between "vibration damper" and "transmission line", the subordinate relationship between "shielding ring" and "transmission line", etc.), so that the model can extract the high-level semantic relationships between the components in the subsequent training process. Choose an object detection network consisting of a feature extraction backbone network, a feature pyramid structure, and a detection head as the base model (e.g., Faster R-CNN, Mask R-CNN, YOLOv3, etc.). The first stage involves training the model's component detection capabilities. In the early stages of training, during the forward propagation process, the backbone network extracts features from the input image, performs upsampling and lateral connections through the feature pyramid, and generates multi-level feature maps that fuse semantic information. On each scale of the feature map, dense candidate bounding boxes are generated through the anchor box mechanism. For each candidate box, extract its corresponding features through ROI Align (such as Faster R-CNN) or direct feature pooling (such as YOLO): bounding box coordinate offset, class probability (obtained through Softmax / Sigmoid algorithm) and target confidence (i.e. whether the candidate box contains a certain part). Internal calculations of the detection head: Classification branch: Fully connected layers / convolutional layers perform multi-class classification on the features of each candidate box (such as transmission lines, insulator strings, vibration dampers), and output the probability distribution of each class; Regression branch: The fully connected / convolutional layer predicts the offset of the bounding box relative to the candidate box. Based on the bounding box coordinate offset, the center coordinates of the predicted box are calculated using the center coordinate decoding formula. Central coordinate decoding formula: ; Where, x （for） and y （for） , where and represent the horizontal and vertical coordinates of the predicted bounding box center, respectively; x （bor） and y （bor） , where and represent the horizontal and vertical coordinates of the center of the candidate box, respectively; ΔX （bor） and ΔY （bor） , representing the horizontal offset of the predicted bounding box relative to the candidate box and the vertical offset of the predicted bounding box relative to the candidate box, respectively; w （bor） and h （bor） , representing the width and height of the candidate box, respectively; By using non-maximum suppression (NMS), the bounding box with the highest confidence and the lowest overlap with other boxes is selected as the final output predicted bounding box within the given prediction bounding box. During backpropagation, design constraints for updating model parameters: Localization loss: used to measure the position and scale error between the predicted bounding box and the ground truth bounding box (i.e., the bounding box coordinate offset of the predicted bounding box relative to the ground truth bounding box). Classification and confidence loss: used to determine whether the predicted class probability is consistent with the true class label, and whether the model's judgment of the target confidence is accurate; During the first stage of training, the parameters of the convolutional layers near the input end in the backbone network are updated slightly, while the parameters of the backbone network in extracting high-level features and the output of the detection head are updated significantly, so that there is no localization deviation between the predicted bounding box output by the model and the real bounding box (i.e., the bounding box coordinate offset of the predicted bounding box relative to the real bounding box is approximately zero). Gradually reduce the weight coefficient of the model in the localization loss, increase the weight coefficient of the model in the classification and confidence loss, and adjust the relevant parameters for extracting high-level semantic features in the backbone network so that the model can capture the subtle differences in the shape and structure of different components (e.g., distinguishing insulator strings from tower components in the background), and enable the model to distinguish transmission lines from the background in the image. As the predicted bounding box positions output by the model gradually converge to the true bounding box positions, data augmentation strategies (such as scale changes, brightness perturbations, and random cropping) are introduced to train the model, so that the model can maintain a certain generalization ability. After the first phase of training is completed, if the training result is that the model can accurately identify the position of each component in the transmission line and identify the category of each component under different environments, different lighting conditions and with certain background interference, then the second phase of training will be carried out; otherwise, the first phase of training will continue. The second stage involves training the model's ability to determine the state of its components. After the first phase of training is completed, without changing the model parameters, the model is trained by comparing normal and damaged images of various components that make up the transmission line. Before training, the damage type of the damaged component images is labeled so that the model can determine whether the component is normal or damaged. Based on the structural relationships between components and the design specifications and operation and maintenance rules of transmission lines, the connection relationships of transmission lines are defined (i.e., the types, quantities, and spatial distribution of components that should be configured on the transmission lines), and then a knowledge graph is constructed using structured data; based on the existing entities and relationships in the knowledge graph, the structural rules of the transmission lines are obtained. During training, the original video is used as video clips as model input. Each video clip is processed frame by frame to obtain the position of each type of component in the current frame. In the time dimension, the positions of each type of component in the video clip are correlated, and the appearance position of the same component in consecutive frames is tracked to obtain the "time trajectory" of each component (i.e., the position change, state change (i.e. whether it is normal or damaged), and confidence change (i.e. whether it is occluded) of the same physical component in consecutive video frames over time). After constructing the temporal trajectory of the components, a structural diagram of the transmission line is generated based on the current video segment. The nodes in the structural diagram represent components that have been detected or should theoretically exist. The attributes of the nodes include: component category, spatial location, and temporal existence probability (i.e., the probability that the component exists within a certain time or the probability that the component is occluded within a certain time). The connection relationship between nodes is determined by the structural rules of the transmission line. In the process of the model outputting different structure diagrams for different video segments, the original video of a normal transmission line is first used as the training sample. The thresholds used to judge the structural consistency rules in the model are gradually adjusted (for example, the "consecutive undetected frame threshold" is used to determine how many consecutive frames a component is not detected before considering it as possibly missing; the "minimum existence probability threshold" is used to determine how low the probability of a detected component is before considering it as not actually existing; and the "structural matching degree threshold" is used to determine how much the actual position of a detected component is offset from the "ideal position in the structural rules of the transmission line" before considering it as incorrectly installed). Then, the original video with missing components is used as the training sample to compare the differences between the "structure diagram output by the transmission line with missing components" and the "structural rules of the transmission line". If a component is marked as "must exist" in the structural rules, and is not detected for more than a set number of consecutive frames under conditions such as unobstructed field of view and camera stability, the probability of the component's existence continues to decrease, and the component is marked as "high probability missing state". During this process, the "threshold used to judge the structural consistency rules" in the model gradually converges, and the weights of the consecutive undetected frame threshold, component existence probability, and structural matching degree threshold are updated, so that the model can judge the normal, damaged, or missing state of the component and obtain the recognition model. Real-time acquisition of monitoring videos of power transmission lines, and use of recognition models to determine whether there are damaged or missing components in the monitoring videos; If such damage or loss occurs, the location of the transmission line with the damaged or missing component is recorded and reported to the relevant authorities; if not, monitoring continues.

[0022] Step S2: Determine whether there is a safety accident on the transmission line (e.g., line fire, breakage, etc.) based on the monitoring video. If so, provide feedback. If not, analyze the monitoring video and determine whether the swing amplitude of the transmission line is safe. If not, provide feedback. If safe, acquire the power data (i.e., the terminal voltage and terminal current at the upstream and downstream ends of the transmission line in real time) and make a second judgment on whether there is a (undetected) safety accident on the transmission line (e.g., line short circuit, line overheating, etc., which cannot be determined by video). If so, provide feedback (to the relevant departments) and adjust the monitoring position of the monitoring device. If not, continue monitoring. Please see Figure 3 The specific steps of step S2 are as follows: The surveillance video undergoes time synchronization and frame extraction to unify the frame rate, and electronic image stabilization is applied to videos exhibiting jitter. Using relatively static structures in the surveillance video (such as towers, crossarms, insulators, etc.) as background feature points; in video segments with shaking, affine transformation or homography transformation is used to correct the image of each frame to eliminate the displacement effect caused by slight camera shaking or platform vibration. After image stabilization is completed, image enhancement processing is performed on the video frames in the monitoring video, including brightness correction, contrast enhancement and local adaptive histogram equalization, to improve the distinguishability between the power transmission line and the background. The recognition model is used to retrieve and locate the conductors in the video, determining the spatial position of the transmission conductors in each frame of the surveillance video; (when determining the position of the transmission conductors in the surveillance video, the pixel-level position of the insulator string or shielding ring can also be obtained at the same time; because the insulator or shielding ring is a relatively rigid fixed structure in the transmission line, when the transmission conductor is obstructed in the surveillance video, the position of the transmission conductor can be determined through these fixed structures). Extracting the trajectory of the power transmission line: Within the detected transmission line area, several discrete feature points are uniformly selected along the direction of the transmission line. These feature points, once determined in the initial frame, will serve as representative points of the transmission line's motion state. As the monitoring video changes frame by frame over time, these feature points are tracked frame by frame to obtain the two-dimensional pixel coordinate change trajectory of each feature point in the monitoring video over time. Time-domain analysis of the change trajectory of feature points: Calculate the average position of each feature point within the time window and use it as the reference position of the point in a windless or balanced state; (the time window can be determined based on the time difference between adjacent peaks or troughs of the feature point in the changing trajectory or by the user and relevant technical personnel) Compare the actual position of each feature point with the reference position at each moment to obtain the pixel displacement sequence of the feature point over time. This displacement sequence represents the lateral and longitudinal sway amplitude of the transmission line under the action of wind. Introducing a spatial scale mapping mechanism: Based on the insulator string (i.e. the aforementioned fixed structure) area in the monitoring video, calculate the pixel length of the insulator string in the image, and combine it with the actual physical length of the insulator of this type to calculate the proportional relationship between the pixel scale and the actual length. Based on this proportional relationship, convert the pixel displacement of the feature points on the transmission line into displacement in actual length units to obtain the motion curve of the swaying amplitude of the transmission line in actual space as a function of time. Analysis of the swing characteristics of the conductor: Extract the maximum swing amplitude x from the displacement sequence of feature points. （max） And calculate the root mean square value of the swing amplitude for each time window; Determine x （max） (Transmission conductor) maximum permissible swing x （safe） Size relationship: If x （max） ≥x （safe） If so, the location of the power transmission line should be immediately reported to the relevant departments; If x （max） <x （safe） If so, in the monitoring video, determine whether the root mean square value of the swing amplitude corresponding to each time window shows an increasing trend; if so, immediately report the location of the power transmission line to the relevant departments. If not, frequency domain analysis is performed on the displacement sequence of the feature points, and the dominant frequency characteristics of the conductor's oscillation are extracted through Fourier transform to determine the potential resonance risk of the conductor's vibration. Displacement sequence D of feature points （fea） Performing a Discrete Fourier Transform (DFT) yields a frequency domain complex sequence X containing the imaginary part. （fea） ; Take sequence X（fea） Square the modulus to obtain the power spectrum (of each frequency that makes up the oscillation amplitude of the transmission line); find the frequency index k corresponding to the maximum power value in the power spectrum. （max） The dominant frequency f is calculated by combining the sampling frequency fs (unit: Hz, i.e., frames per second) of the surveillance video. （dom） ; Determine the dominant frequency f of the oscillation of the transmission line. （dom） Is it close to a certain natural frequency of the transmission line (i.e., the "dominant frequency f")? （dom） The difference between "the natural frequency of a certain order of the transmission line" and "the natural frequency of a certain order of the transmission line" is within 10% of the natural frequency of that order. If so, it indicates that the power transmission line in the area is at risk of resonance, and the actual area of ​​the power transmission line should be the key area for monitoring by the monitoring device. If not, then do not process it; The system acquires real-time power data from the upstream and downstream ends of the transmission line to make a secondary assessment of whether there are any undetected safety incidents (such as line short circuits, discharges, or other safety incidents that cannot be determined by video). If such incidents exist, the system reports back to the relevant departments and adjusts the monitoring position of the monitoring device. If no such incidents exist, the system continues monitoring. The transmission line and its upstream and downstream ends are considered as a "π-type equivalent circuit", and the terminal voltage V at the upstream and downstream ends is obtained. （1） and V （2） Terminal current I （1） and I （2） ; Establish a basic equivalent model for transmission lines: ; Where Y represents the "equivalent series admittance" of the transmission line; The "equivalent resistance" of the R transmission line; The equivalent reactance of the X transmission line; Y （c） This refers to the "equivalent parallel susceptance" of a transmission line. Introducing voltage phase angle change weights λ for active and reactive power. （P） and λ （Q） Voltage difference change weight μ （P） and μ （Q） Establish a mapping relationship between voltage, current, and power at both ends of the transmission line: ; Among them, P （1） and Q （1） , representing the active power and reactive power upstream of the transmission line, respectively; δ represents the phase angle difference between the upstream and downstream voltages; Iterative correction of the power balance at both ends of the transmission line: Obtain the actual voltage, current and power received at the downstream end of the transmission line, and calculate the ideal transmission power when transmitting power from the upstream end to the downstream end based on the equivalent resistance, reactance, conductance and susceptance of the transmission line (perform power flow calculation); Obtain the actual transmission power at the upstream end of the transmission line and compare it with the ideal transmission power: If the actual transmission power at the upstream end is equal to or approximately equal to the ideal transmission power (after retaining 2 decimal places), then the transmission line is considered to be normal. If there is a deviation between the actual transmission power and the ideal transmission power at the upstream end, the transmission power at the upstream end of the transmission line will be adjusted to correct the phase angle difference of the terminal voltage at both ends: Step a1: Calculate the power difference ΔP between the ideal transmission power and the actual transmission power at the upstream end. （tra） ; Step a2: Calculate the change in phase angle difference δ, Δδ: Where η represents the iteration factor; Step a3: Adjust the phase angle difference δ （1） : According to δ （1） And the mapping of voltage, current and power at both ends of the transmission line, and recalculate the active power when transmitting power from the upstream end to the downstream end; Step a4: Repeat steps a3 to a4 until the phase angle difference δ in the i-th iteration is reached. （i） The phase angle difference δ with the (i-1)th iteration （i-1） Same, and δ （i） The corresponding active power is the same as the ideal transmission power; According to δ （i） Adjust the voltage phase angle difference between the upstream and downstream ends of the transmission line; and determine whether the actual active power received at the downstream end remains unchanged. If it remains unchanged, it indicates that the transmission line is normal; If the actual active power received at the downstream end decreases, it indicates that an overheating accident has occurred in the transmission line. If the actual active power received at the downstream end increases, it indicates that a short circuit has occurred in the transmission line. Step S3: Divide the monitoring device of the transmission line into two parts; one part monitors the area of ​​the transmission line where there is component damage, missing parts or safety accidents until the parts in that area are replaced and repaired; the other part detects the normal area of ​​the transmission line and reports the abnormal conditions (i.e., component damage, missing parts or safety accidents on the transmission line).

[0023] The above formulas are all dimensionless calculations. The formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation. For example, there are weighting coefficients and proportional coefficients. The values ​​set are to quantify each parameter to obtain a specific value, which is convenient for subsequent comparison. The values ​​of the weighting coefficients and proportional coefficients are only required to not affect the proportional relationship between the parameters and the quantified values.

[0024] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A detection method for a video image monitoring device for power transmission lines, characterized in that, The method includes: Acquire the original video and train the basic model in two stages to obtain the recognition model; acquire the monitoring video of the transmission line and use the recognition model to determine whether there are any damaged or missing parts in the monitoring video; if so, provide feedback. The system determines whether there is a safety accident on the transmission line based on the monitoring video. If there is, feedback is provided. If not, the system analyzes the monitoring video to determine whether the swing amplitude of the transmission line is safe. If not, feedback is provided. If safe, the system acquires the power data of the upstream and downstream ends of the transmission line in real time to make a second determination of whether there is a safety accident on the transmission line. If there is, feedback is provided and the monitoring position of the monitoring device is adjusted. If not, monitoring continues. Monitor areas in power transmission lines where components are damaged, missing, or there are safety incidents, until the components in that area are replaced and repaired.

2. The detection method of the video image monitoring device for transmission lines according to claim 1, characterized in that, The steps to build a recognition model are as follows: The original video is processed by frame extraction. The visible components of the transmission line in each frame are defined by bounding boxes, and the category and status of the visible components in each frame, as well as the structural relationship between the components, are labeled. Select a base model; in the first stage, train the model's component detection capabilities. The second stage involves training the model's ability to determine the state of its components. After training, a recognition model is obtained; Real-time acquisition of monitoring videos of power transmission lines, and use of recognition models to determine whether there are damaged or missing components in the monitoring videos; If this occurs, record the location of the power transmission line where the component is damaged or missing, and report it to the relevant authorities; If it does not appear, continue monitoring.

3. The detection method of the video image monitoring device for transmission lines according to claim 2, characterized in that, The training steps for the first phase are as follows: During the forward propagation process, the backbone network extracts features from the input image, performs upsampling and lateral connections through the feature pyramid, and generates multi-level feature maps that fuse semantic information. On each feature map, dense candidate bounding boxes are generated through the anchor box mechanism. For each candidate box, extract its features; The features of each candidate box are classified into multiple categories through fully connected layers and convolutional layers, and the probability distribution of each category is output. Then, the offset of the bounding box relative to the candidate box is predicted through fully connected layers and convolutional layers. Based on the offset of the bounding box coordinates, the center coordinates of the predicted box are calculated using the center coordinate decoding formula. Using nonmaximum suppression, within a given prediction box, the bounding box with the highest confidence and the lowest overlap with other boxes is selected as the final output prediction bounding box. During backpropagation, localization loss, classification loss, and confidence loss are designed to update the model parameters. If the model training result is that the model identifies the location of each component in the transmission line and identifies the category of each component, then the second stage of training is performed; otherwise, the first stage of training continues.

4. The detection method of the video image monitoring device for transmission lines according to claim 3, characterized in that, The steps for updating model parameters are as follows: During the first stage of training, the parameters of the convolutional layers near the input end in the backbone network are updated, as well as the parameters related to the extraction of high-level features and the output of the detection head, so that there is no localization deviation between the predicted bounding boxes output by the model and the real bounding boxes. Gradually reduce the weight coefficient of the model in the localization loss, increase the weight coefficient of the model in the classification and confidence loss, and adjust the relevant parameters for extracting high-level semantic features in the backbone network so that the model can capture the subtle differences in the morphology and structure of different components and distinguish the transmission line from the background. As the predicted bounding box positions output by the model gradually converge to the true bounding box positions, a data augmentation strategy is introduced to train the model and maintain its generalization ability.

5. The detection method of the video image monitoring device for transmission lines according to claim 3, characterized in that, The training steps for the second phase are as follows: Without changing the model parameters, the model is trained by comparing normal and damaged images of various components that make up the transmission line. Before training, the damage type of the damaged images of the components is labeled so that the model can determine whether the components are normal or damaged. Based on the structural relationships between components and the design specifications and operation and maintenance rules of transmission lines, the connection relationships of transmission lines are defined, and then a knowledge graph is constructed in the form of structured data; based on the existing entities and relationships in the knowledge graph, the structural rules of transmission lines are obtained. During training, the original video is used as video clips as model input. Each video clip is processed frame by frame to obtain the position of each type of component in the current frame. In the time dimension, the positions of each type of component in the video clip are correlated, and the occurrence position of the same component in consecutive frames is tracked to obtain the time trajectory and confidence change of each component. A structural diagram of the power transmission line is generated based on the current video clip. The nodes in the structural diagram represent components that have been detected or should exist.

6. The detection method of the video image monitoring device for transmission lines according to claim 5, characterized in that, The subsequent training steps for the second phase are as follows: First, using original videos of normal transmission lines as training samples, the thresholds for judging structural consistency rules in the model are gradually adjusted. Then, using original videos with missing components as training samples, the differences between the output structural diagrams of transmission lines with missing components and the structural rules of the transmission lines are compared. If a component is marked as present in the structural rules, but is not detected after exceeding a set number of consecutive frames, the probability of its presence continues to decrease, and the component is marked as a high-probability missing state. This causes the threshold used in the model to judge the structural consistency rules to gradually converge. Then, the weights of the consecutive undetected frame threshold, the component presence probability, and the structural matching degree threshold are updated, so that the model can judge the normal, damaged, or missing state of the component, thus obtaining the recognition model.

7. The detection method of the video image monitoring device for transmission lines according to claim 1, characterized in that, The steps to determine whether the swing amplitude of a transmission line is safe are as follows: The surveillance video is time-synchronized and frame-sampling is performed to unify the frame rate, and electronic image stabilization and enhancement are performed on videos with jitter. The recognition model is used to retrieve and locate power transmission lines in the video, determining the spatial position of the power transmission lines in each frame of the surveillance video and extracting the movement trajectory of the lines. Several feature points are uniformly selected along the direction of the conductor, and the feature points are tracked frame by frame to obtain the two-dimensional pixel coordinate change trajectory of each feature point in the monitoring video over time. Time-domain analysis of the change trajectory of feature points: Calculate the average position of each feature point within the time window as the reference position for that point; compare the actual position of each feature point at each time step with the reference position to obtain the pixel displacement sequence of that feature point over time. Analysis of the swing characteristics of the conductor: Extract the maximum swing amplitude x from the displacement sequence of feature points. （max） And calculate the root mean square value of the swing amplitude corresponding to each time window.

8. The detection method of the video image monitoring device for transmission lines according to claim 1, characterized in that, The following steps are required to determine whether the swing amplitude of a transmission line is safe: Determine x （max） With maximum permissible swing x （safe） Size relationship: If x （max） ≥x （safe） If so, the position of the transmission line will be fed back; If x （max） <x （safe） In the monitoring video, it is determined whether the root mean square value of the swing amplitude corresponding to each time window shows an increasing trend; if so, the position of the transmission line is immediately reported. If not, frequency domain analysis is performed on the displacement sequence of the feature points, and the dominant frequency characteristics of the conductor's oscillation are extracted through Fourier transform to determine the potential resonance risk of the conductor's vibration. Displacement sequence D of feature points （fea） Performing a discrete Fourier transform yields a frequency domain complex sequence X containing the imaginary part. （fea） ; Take sequence X （fea） Square the modulus to obtain the power spectrum; find the frequency index k corresponding to the maximum power value in the power spectrum. （max） Based on the sampling frequency of the surveillance video, the dominant frequency f is calculated. （dom） ; Determine the dominant frequency f of the oscillation of the transmission line. （dom） Whether it is close to a certain natural frequency of the transmission line; If so, it indicates that the power transmission line in the area is at risk of resonance, and the actual area of ​​the power transmission line should be the key area for monitoring by the monitoring device. If not, then do not process it.

9. The detection method of the video image monitoring device for transmission lines according to claim 1, characterized in that, The steps for a secondary assessment of whether a transmission line has a safety incident are as follows: The system acquires real-time power data from the upstream and downstream ends of the transmission line to determine if there is a safety accident. If so, it provides feedback and adjusts the monitoring position of the monitoring device; if not, it continues monitoring. Introducing voltage phase angle change weights λ for active and reactive power. （P） and λ （Q） Voltage difference change weight μ （P） and μ （Q） Establish a mapping relationship between voltage, current, and power at both ends of the transmission line, and iteratively correct the power balance at both ends of the transmission line. Obtain the actual voltage, current and power received at the downstream end of the transmission line, and calculate the ideal transmission power when transmitting power from the upstream end to the downstream end based on the equivalent resistance, reactance, conductance and susceptance of the transmission line. Obtain the actual transmission power at the upstream end of the transmission line and compare it with the ideal transmission power: If the actual transmission power at the upstream end is equal to or approximately equal to the ideal transmission power, then the transmission line is considered to be normal. If there is a deviation between the actual transmission power and the ideal transmission power at the upstream end, the transmission power at the upstream end of the transmission line will be adjusted to correct the phase angle difference of the terminal voltage at the upstream and downstream ends.

10. The detection method of the video image monitoring device for transmission lines according to claim 9, characterized in that, The steps to correct the voltage phase angle difference are as follows: Step a1: Calculate the power difference ΔP between the ideal transmission power and the actual transmission power at the upstream end. （tra） ; Step a2: Calculate the change in phase angle difference δ, Δδ; Step a3: Adjust the phase angle difference δ （1） According to δ （1） And the mapping of voltage, current and power at both ends of the transmission line, and recalculate the active power when transmitting power from the upstream end to the downstream end; Step a4: Repeat steps a3 to a4 until the phase angle difference δ in the i-th iteration is reached. （i） The phase angle difference δ with the (i-1)th iteration （i-1） Same, and δ （i） The corresponding active power is the same as the ideal transmission power; According to δ （i） Adjust the voltage phase angle difference between the upstream and downstream ends of the transmission line according to the value of . And determine whether the actual active power received downstream remains unchanged; If it remains unchanged, it indicates that the transmission line is normal; If the actual active power received at the downstream end decreases, it indicates that an overheating accident has occurred in the transmission line. If the actual active power received downstream increases, it indicates that a short circuit has occurred in the transmission line.

Images