A power inspection method and system based on drone

Abstract

The present invention relates to the field of electric power inspection, and specifically to an electric power inspection method and system based on an unmanned aerial vehicle. The method includes two aspects: data initialization and electric power inspection. The data initialization completes data collection, inspection parameter calculation, and training and compression processing of an inspection model. The electric power inspection can realize offline, real-time automatic inspection in signal-free areas such as mountains by receiving inspection parameters and inspection models calculated and completed by a server. The inspection has the functions of automatic route planning, obstacle avoidance reminder, and inspection center maintenance. The inspection collects detection results of six circumferential angles of the pole tower and seven central angles of the top of the pole tower for comprehensive judgment, which can solve the problem of component occlusion in fault detection. At the same time, the problem of component missing can be detected by comparing historical angle pictures. Finally, the present invention provides exposure abnormality repair based on historical angle pictures, optimizes the quality of inspection pictures, and ensures inspection accuracy.

Description

Technical Field

[0001] The present invention relates to the field of electric power inspection, and in particular to an electric power inspection method and system based on an unmanned aerial vehicle. Background Art

[0002] Drone-based power inspection is a technical means of using modern information technologies such as drone technology, remote sensing technology, artificial intelligence, big data, etc. to conduct remote, fast, and accurate inspections of power facilities such as transmission lines and substations. This technology aims to improve the safety and reliability of power facilities, reduce operation and maintenance costs, and improve inspection efficiency.

[0003] In actual inspections, the accuracy of drone flight obstacle avoidance and inspection fault detection is the core problem to be solved in inspections. Patent publication number CN110119161B discloses a method and system for drone transmission line inspection visual obstacle avoidance, which uses binocular vision to detect transmission lines and thus avoid obstacles. However, in actual operation, due to the thin wires, the visual camera often cannot effectively capture effective images of the transmission lines.

[0004] Patent publication number CN108761237A discloses a method for data annotation and fault diagnosis of key power components based on SSD neural network. The method can reduce the workload of manual inspections, but the method does not address issues such as obstruction of power components and missing components. At the same time, the method is a post-inspection detection method. During inspections, especially in mountainous areas with poor signals, the inspection results cannot be obtained in a timely manner, and the location of the faulty component is not convenient for three-dimensional spatial description, which is not conducive to on-the-spot repairs by maintenance personnel. Summary of the invention

[0005] In order to solve the problem that the above-mentioned drone cannot perform real-time detection to obtain accurate detection results and effectively avoid obstacles such as power lines in areas with poor signals such as mountainous areas, the present invention provides a power inspection method and system based on drones.

[0006] The present invention discloses a power inspection method based on a drone, the method comprising: data initialization, power inspection;

[0007] Data initialization, including: data collection, data upload, data processing to generate inspection parameter files, and data storage;

[0008] Data collection,Data collection is divided into two stages. In the first stage, the drone is manually controlled to take off from a safe position and fly vertically along the tower. The drone camera is horizontally facing the tower (the pitch angle is 0°). After flying to a safe distance beyond the tower, the drone turns the camera downward (the pitch angle is -90°) and flies horizontally along the longest cross arm of the tower until it reaches the other end of the longest cross arm, completing the first stage of data collection. In the second stage, the route is manually planned, and the drone flies around the tower. During this period, the camera always faces the tower. After flying around once, the second stage of flight is completed. The data of the entire flight process is recorded in real time to complete data collection.

[0009] Data upload: the data collected by the drone is packaged, compressed and uploaded by tower;

[0010] Data processing, analyzes flight data by tower. In the first stage of vertical flight data, the target detection network is used to identify the tower components. The identification types include at least: tower, cross arm, insulator, wire. In the vertical flight stage, the drone first records the take-off point height H. 0 When the cross arm is identified to appear at the center height of the image, the corresponding UAV elevation information L is recorded. j , j starts from 0, indicating the crossarms arranged in order from bottom to top, until the top of the tower is at the center height of the image, and the elevation information of the drone is recorded, which is recorded as H 1 ; In the first stage of horizontal flight, when the tower is detected in the center of the image, the GPS position information of the drone is recorded (V x0 ,V y0 ), save the corresponding image to the data storage end, and record the GPS location information of the drone when the longest cross arm is detected at the center of the image (V x1 ,V y1 ), when the other end of the longest cross arm is detected at the center of the image, the GPS position information of the drone is recorded (V x2 ,V y2 ), based on this, establish the tower coordinate system O with the current intersection of the tower and the ground as the center, the concrete pole upward as the Z axis, the eastward as the X axis, and the northward as the Y axis. i， At the same time, record the coordinate system O i The origin corresponds to the longitude and latitude O i0 (V x0 ,V y0 ,H 0 ), the average center latitude and longitude height of all cross arms on the tower is O i1 (V x0 ,V y0 ,(H 1 -L 0 ) / 2+L 0), where i represents different towers; in the second flight phase, the UAV flies around the tower and uses the target detection network to identify the tower components of the flight image data. The identification types include at least: towers, crossarms, and insulators. According to the recognition results, based on the current GPS position of the UAV and the tower coordinate system O i Calculate the quadrant to which the drone belongs and select six angles of uniform distribution starting from the north direction with the tower as the center. Based on the vertical reference of the tower concrete pole and the horizontal reference of the cross arm, the coordinates of the tower components are expressed by calculating the ratio of the pixel distance from the center of the marked target to the reference object and the pixel length of the reference object itself, and the relative coordinates of the tower components (X rela_k ,Y rela_k ) and marked with number T i_k , where T represents the component type number, different components have different type numbers, i_k represents the kth T-type component on the i-th tower, and accordingly, the image data of the six angles are synchronously saved to the data storage end;

[0011] The data processing also calculates the tower obstacle avoidance area, which is L 0 -H 1 As the horizontal section, (V x1 ,V y1 )—(V x2 ,V y2 ) is the vertical section, the section direction is the vertical plane direction between the cross arm and the concrete pole, and the longest obstacle avoidance length is the length L corresponding to the longest cross arm max , expressed as |(V x1 ,V y1 )-(V x2 ,V y2 )|, that is (V x1 ,V y1 ) point to (V x2 ,V y2 ) distance, the obstacle avoidance depth is infinite. When there are multiple cross arms with different directions, the obstacle avoidance zones in different directions are perpendicular to the corresponding cross arms. In terms of obstacle avoidance length, detection (V x0 ,V y0 ) position, the crossarm in the drone image is calculated by comparing the ratio of the image pixel length of different crossarms to the longest crossarm and the actual length L of the longest crossarm. max The final obstacle avoidance area BA is calculated proportionally. i It is obtained by adding the intersection of the horizontal and vertical obstacle avoidance zones to the safety distance D.

[0012] In addition, data processing also includes training, optimizing, pruning and quantizing the target detection and fault detection network models through the collected data.

[0013] All results of data processing are saved after manual verification.

[0014] Data storage, data storage is persistently saved, and the saved data at least includes: flight data collected and uploaded, inspection parameter file data generated by data processing, and model parameter data for target detection, target tracking, and exposure correction, among which the flight data at least includes: the altitude H of the drone take-off point 0 , when the drone is flush with the top of the tower, the elevation H 1 , Cross arm corresponding height L j .

[0015] When power inspection is required, the inspection client requests the server to download the inspection parameter offline file. The offline parameters include at least: the tower coordinate system O i The origin corresponds to the longitude and latitude O i0 (V x0 ,V y0 ,H 0 ), the average center latitude and longitude height of all cross arms on the tower O i1 (V x0 ,V y0 ,(H 1 -L 0 ) / 2+L 0 ), the final obstacle avoidance area BA i , seven pictures in total from six angles around the tower and the angle of the top facing the tower, the relative coordinates of the tower components, and the quantized and compressed parameters of the target detection and fault detection network model;

[0016] During the inspection, the drone first flies to the center of the tower top and records the current GPS position and elevation (V x ,V y ,H), through i0 (V x0 ,V y0 ,H 0 ) to compare and calibrate GPS and altitude. The altitude calibration is directly based on the altitude of the take-off point of the UAV and the altitude of the GPS. i0 The elevation coordinates are compared and corrected, and the UAV is subsequently used for circumnavigation inspection. The inspection client service calculates the position of the UAV in real time, performs target identification and fault detection. Among them, the relative coordinates (X rela ,Y rela ) and the current angle of the drone tower component coordinates (X rela_k ,Y rela_k ) to calculate the number T corresponding to the faulty component i_k .

[0017] Among them, the data records in the data collection can be collected by taking photos at equal time intervals to prevent the image quality from being degraded due to video data compression when the data is saved by video recording.

[0018] The data collected and uploaded include at least: UAV flight image data, corresponding GPS location information and elevation information.

[0019] In the flight image data, when the images of the six side angles and the top center angle are not clear, the data of similar angles can be manually reselected in the database.

[0020] Furthermore, the safety distance D in the method is related to the wind speed V. During inspection, the drone automatically collects the current wind speed V, automatically adjusts the safety distance D and updates the obstacle avoidance area BA i .

[0021] When the power inspection is running, the inspection client uses the drone's GPS and elevation information to calculate the drone's position relative to the tower in real time. When the drone's position intersects the tower's obstacle avoidance zone, the client issues an obstacle avoidance warning.

[0022] Furthermore, the drone inspection also receives the drone flight route when the service end collects tower data. Before the inspection flight, the inspection end uses the obstacle avoidance area BA i To smooth the collection route for the restriction conditions, complete the automatic route planning, and during the inspection, the drone collects the current wind speed data. When the wind speed V changes, the inspection client updates the safety distance D and adjusts the obstacle avoidance area BA i , update the inspection flight route.

[0023] Furthermore, during the drone inspection, the current position is solved in real time and the current camera attitude is dynamically calculated. By dynamically adjusting the camera attitude angle, the drone camera is always facing the mean center of all crossarms of the tower O i1 (V x0 ,V y0 ,(H 1 -L 0 ) / 2+L 0 ), obtain the image data facing the tower components.

[0024] Furthermore, the drone inspection adopts a simple method, which includes: when the drone flies in a circle to detect power faults, the drone calculates the current position in real time and selects the same six angles as those used in data collection to take pictures. Finally, the drone flies to the top of the tower and collects pictures of the tower components facing the top. The collected data is used for real-time target detection and fault detection, and the detection results of each angle are compared with the corresponding tower component category T. i_k The final detection result is obtained by summarizing the detection results of the seven angles and taking the union.

[0025] In addition, the power inspection also detects missing components. For each angle, the algorithm compares the current component position with the historical component position at the corresponding angle. When the comparison results of all angles indicate that a component is missing, the inspection result marks a component missing fault.

[0026] In order to improve the quality of collected images, power inspection will also request the exposure anomaly correction network from the server. When the exposure of the image taken is abnormal due to the switching of the drone angle, the inspection service will retrieve the tower comparison map of the corresponding angle of the current drone and use the exposure anomaly correction network to correct the exposure of the abnormal exposure image.

[0027] The flight data and test results of each inspection are sent to the data storage end after the inspection is completed and returned, and data processing is performed to update the corresponding models and data. Among them, the flight data of the inspection includes: image data, drone location and elevation information.

[0028] Finally, based on the above method, the present invention provides a UAV-based power inspection system, including: an inspection client, a server and a data storage terminal; the inspection client is installed on the UAV end, separated from the server end, to complete data collection, data upload and download, and power inspection functions, wherein the power inspection function receives processed offline data from the server end to complete the corresponding function, and at the same time, the client end has a local data storage module; the server end runs a special server to perform big data and large model calculations to complete the data processing function; the data storage terminal runs independently to complete the data storage function.

[0029] Beneficial effects:

[0030] (1) The inspection client of the present invention is separated from the server. The server receives the data collected by the client, performs data processing, model training, quantization compression, and downloads it to the inspection client. The client receives the processed data without the need for a large amount of calculation. It can realize local offline, real-time drone obstacle avoidance reminders and fault detection processing when it cannot be connected to the server. Therefore, for some simple problems, the inspection personnel can deal with them immediately without having to go deep into the mountains again.

[0031] (2) The present invention integrates multiple angle images, drone location information and specially designed collection routes to establish a tower spatial coordinate system with the tower as the unit. When the drone is inspecting, the inspection client can solve the drone position in real time and calculate the spatial position of the tower components and match the corresponding component numbers. Therefore, for the same tower component, multiple angles can be used to jointly perform fault detection, solve the problem of component occlusion and provide the accurate spatial position of the faulty component (corresponding to the corresponding number) to facilitate maintenance personnel to quickly locate the problem location. At the same time, by comparing the corresponding angle images requested by the client and the server, the problem of missing components can be detected. Furthermore, since multiple sets of images are used to detect the same component, the detection accuracy of fault detection can be improved.

[0032] (3) The present invention uses target detection to establish obstacle avoidance zones for the vertical direction area and the height area of ​​the crossarm. At the same time, the obstacle avoidance zone can also be manually delineated. In addition, the drone can update the inspection route and the size of the obstacle avoidance zone in real time according to the current wind speed V and other data. The automatically planned route can be used during the inspection. In the application, manual inspection has a higher priority and can directly obtain control to conduct inspection. During the inspection, even if the wires are not visible in the image, the client can still provide obstacle avoidance reminders. In addition, in order to improve the inspection accuracy, the present invention adjusts the camera posture through coordinate conversion so that the camera is always facing the mean center O of all the crossarms of the tower. i1 (V x0 ,V y0 ,(H 1 -L 0 ) / 2+L 0 ), obtain the picture data facing the tower component and reduce background interference. On the other hand, for exposure abnormalities caused by shooting angle problems, in addition to the camera's own automatic exposure algorithm, the present invention can correct the exposure abnormality picture by comparing the historical clear pictures of the angle and the exposure abnormality repair neural network. Moreover, since the historical corresponding angle pictures contain the clear image contrast, brightness distribution, channel color gradient and other information of the tower component, it has a better effect than directly repairing the exposure abnormality. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0034] Figure 1 It is a schematic diagram of the module structure of the present invention;

[0035] Figure 2 It is a schematic diagram of the logic structure of the power inspection of the present invention;

[0036] Figure 3 It is a schematic diagram of marking the tower components of the present invention;

[0037] Figure 4 It is a schematic diagram of the obstacle avoidance area of ​​the present invention;

[0038] Figure 5 It is a three-dimensional space schematic diagram of the obstacle avoidance area of ​​the present invention;

[0039] Figure 6 It is a schematic diagram of six typical angles and the top center angle of the present invention;

[0040] Figure 7 A schematic diagram of summarizing and intersecting the test results of the present invention;

[0041] Figure 8 It is a schematic diagram of repairing abnormal exposure according to the present invention.

[0042] In the figure, A 1_1 Indicates the first insulator of the first tower; A 1_2 Indicates the second insulator of the first tower; A 1_3 Indicates the third insulator of the first tower; L max Indicates the longest cross arm distance; D indicates the safety distance; L H0 Indicates the height of the first cross arm from the starting point of the tower; L H1 Indicates the height of the second cross arm from the starting point of the tower; H H1 Indicates the height of the tower top from the tower starting point. DETAILED DESCRIPTION

[0043] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0044] Embodiment 1:

[0045] like Figure 1-Figure 8 As shown, the present invention provides a power inspection method based on a drone, the method comprising: data initialization, power inspection;

[0046] Data initialization, including: data collection, data upload, data processing to generate inspection parameter files, and data storage;

[0047] Data collection,Data collection is divided into two stages. In the first stage, the drone is manually controlled to take off from a safe position and fly vertically along the tower. The drone camera is horizontally facing the tower (the pitch angle is 0°). After flying to a safe distance beyond the tower, the drone turns the camera downward (the pitch angle is -90°) and flies horizontally along the longest cross arm of the tower until it reaches the other end of the longest cross arm, completing the first stage of data collection. In the second stage, the route is manually planned, and the drone flies around the tower. During this period, the camera always faces the tower. After flying around once, the second stage of flight is completed. The data of the entire flight process is recorded in real time to complete data collection.

[0048] Data upload: the data collected by the drone is packaged, compressed and uploaded by tower;

[0049] Data processing, analyzes flight data by tower. In the first stage of vertical flight data, the target detection network is used to identify the tower components. The identification types include at least: tower, cross arm, insulator, wire. In the vertical flight stage, the drone first records the flight starting point altitude H 0 When the cross arm is identified to appear at the center height of the image, the corresponding UAV elevation information L is recorded. j , until the top of the tower is at the center of the image, record the altitude information of the drone, and record it as H 1 ; In the first stage of horizontal flight, when the tower is detected in the center of the image, the GPS position information of the drone is recorded (V x0 ,V y0 ), save the corresponding image to the data storage end, and record the GPS location information of the drone when the longest cross arm is detected at the center of the image (V x1 ,V y1 ), when the other end of the longest cross arm is detected at the center of the image, the GPS position information of the drone is recorded (V x2 ,V y2 ), based on this, establish the tower coordinate system O with the current intersection of the tower and the ground as the center, the concrete pole upward as the Z axis, the eastward as the X axis, and the northward as the Y axis. i， At the same time, record the coordinate system O i The origin corresponds to the longitude and latitude O i0 (V x0 ,V y0 ,H 0 ), the average center latitude and longitude height of all cross arms on the tower is O i1 (V x0 ,V y0 ,(H 1 -L 0 ) / 2+L 0 ), where i represents different towers; in the second flight phase, the UAV flies around the tower and uses the target detection network to identify the tower components of the flight image data. The identification types include at least: towers, crossarms, and insulators. According to the recognition results, based on the current GPS position of the UAV and the tower coordinate system O i Calculate the quadrant to which the drone belongs and select six angles of uniform distribution starting from the north direction with the tower as the center. Based on the vertical reference of the tower concrete pole and the horizontal reference of the cross arm, the coordinates of the tower components are expressed by calculating the ratio of the pixel distance from the center of the marked target to the reference object and the pixel length of the reference object itself, and the relative coordinates of the tower components (X rela_k ,Y rela_k ) and marked with number T i_k,Correspondingly, the image data of the six angles are synchronously saved to the data storage end;

[0050] The data processing also calculates the tower obstacle avoidance area, which is L 0 -H 1 As the horizontal section, (V x1 ,V y1 )—(V x2 ,V y2 ) is the vertical section, the section direction is the vertical plane direction between the cross arm and the concrete pole, and the longest obstacle avoidance length is the longest cross arm corresponding to L max , expressed as |(V x1 ,V y1 )-(V x2 ,V y2 )|, that is (V x1 ,V y1 ) point to (V x2 ,V y2 ), the calculation formula is , the obstacle avoidance depth is infinite. When there are multiple cross arms with different directions, the obstacle avoidance zones in different directions are perpendicular to the corresponding cross arms. In terms of obstacle avoidance length, detection (V x0 ,V y0 ) position, the crossarm in the drone image is calculated by comparing the ratio of the image pixel length of different crossarms to the longest crossarm and the actual length L of the longest crossarm. max The final obstacle avoidance area BA is calculated proportionally. i It is obtained by adding the intersection of the horizontal and vertical obstacle avoidance zones to the safety distance D.

[0051] In addition, data processing also includes training, optimizing, pruning and quantizing the target detection and fault detection network models through the collected data.

[0052] All results of data processing are saved after manual verification.

[0053] Data storage, data storage is persistently saved, and the saved data at least includes: flight data collected and uploaded, inspection parameter file data generated by data processing, and model parameter data for target detection, target tracking, and exposure correction, among which the flight data at least includes: the altitude H of the drone take-off point 0 , when the drone is flush with the top of the tower, the elevation H 1 , Cross arm corresponding height L j .

[0054] When power inspection is required, the inspection client requests the server to download the inspection parameter offline file. The offline parameters include at least: the tower coordinate system O i The origin corresponds to the longitude and latitude O i0 (V x0,V y0 ,H 0 ), the average center latitude and longitude height of all cross arms on the tower O i1 (V x0 ,V y0 ,(H 1 -L 0 ) / 2+L 0 ), the final obstacle avoidance area BA i , seven pictures in total from six angles around the tower and the angle of the top facing the tower, the relative coordinates of the tower components, and the quantized and compressed parameters of the target detection and fault detection network model;

[0055] During the inspection, the drone first flies to the center of the tower top and records the current GPS position and elevation (V x ,V y ,H), through i0 (V x0 ,V y0 ,H 0 ) to compare and calibrate GPS and altitude. The altitude calibration is directly based on the altitude of the take-off point of the UAV and the altitude of the GPS. i0 The elevation coordinates are compared and corrected, and the UAV is subsequently used for circumnavigation inspection. The inspection client service calculates the position of the UAV in real time, performs target identification and fault detection. Among them, the relative coordinates (X rela ,Y rela ) and the current angle of the drone tower component coordinates (X rela_k ,Y rela_k ) to calculate the number T corresponding to the faulty component i_k .

[0056] Among them, the data records in the data collection can be collected by taking photos at equal time intervals to prevent the image quality from being degraded due to video data compression when the data is saved by video recording.

[0057] The data collected and uploaded include at least: UAV flight image data, corresponding GPS location information and elevation information.

[0058] In the flight image data, when the images of the six side angles and the top center angle are not clear, they can be manually reselected in the database.

[0059] In practical applications, for the first time, data collection needs to be carried out specifically. The drone first flies vertically and then horizontally along the tower. By extracting the position and elevation of the drone, the tower can be measured and the tower coordinate system can be established. Based on this, all subsequent calculations can convert the position information into the tower coordinate system to realize the conversion of data at different angles. During inspection, only one correction is required based on the latitude and longitude of the coordinate origin of the coordinate. After collecting the data, the drone sends the collected image data and the corresponding GPS and elevation information of the image to the server. The server runs a special server to parse the information and calculate the spatial coordinate system position, obstacle avoidance zone position, and pictures of various angles corresponding to each tower. The data required for subsequent calculations, at the same time, the server has a large computing power, and can use the collected image data to train the target detection and fault detection and exposure anomaly correction networks. The trained model needs to be compressed by model pruning, quantization and other operations to reduce the amount of calculation. Therefore, when the model is sent to the client, the client can also perform real-time calculations without worrying about computing power issues. At the same time, in practical applications, the target detection and fault detection networks can be merged into one network to further reduce the amount of model calculation. After the first data collection is completed and processed by the server and manually checked, offline and real-time inspections can be carried out during field inspections based on the server data updated by the first and subsequent inspections, solving the problem when the client and server cannot communicate, and providing real-time obstacle avoidance reminders and fault detection. For simple problems, inspection personnel can carry out repairs immediately after detection without having to go deep into special environments such as mountains again.

[0060] Embodiment 2:

[0061] like Figure 1-Figure 8 As shown, the safety distance D in the method of the present invention is related to the wind speed V. During inspection, the drone automatically collects the current wind speed V, automatically adjusts the safety distance D and updates the obstacle avoidance area BA i .

[0062] When the power inspection is running, the inspection client uses the drone's GPS and elevation information to calculate the drone's position relative to the tower in real time. When the drone's position intersects the tower's obstacle avoidance zone, the client issues an obstacle avoidance warning.

[0063] Furthermore, the drone inspection also receives the drone flight route collected by the service end tower data. Before the inspection flight, the inspection end uses the obstacle avoidance area BA i To smooth the collection route for the restriction conditions, complete the automatic route planning, and during the inspection, the drone collects the current wind speed data. When the wind speed V changes, the inspection client updates the safety distance D and adjusts the obstacle avoidance area BA i , update the inspection flight route.

[0064] Furthermore, during the drone inspection, the current position is solved in real time and the current camera attitude is dynamically calculated. By dynamically adjusting the camera attitude angle, the drone camera is always facing the mean center of all crossarms of the tower O i1 (V x0 ,V y0 ,(H 1 -L 0 ) / 2+L 0 ), obtain the image data facing the tower components.

[0065] Furthermore, the drone inspection adopts a simple method, which includes: when the drone flies in a circle to detect power faults, the drone calculates the current position in real time and selects the same six angles as those used in data collection to take pictures. Finally, the drone flies to the top of the tower and collects pictures of the tower components facing the top. The collected data is used for real-time target detection and fault detection, and the detection results of each angle are compared with the corresponding tower component category T. i_k The final detection result is obtained by summarizing the detection results of the seven angles and taking the union.

[0066] In addition, the power inspection also detects missing components. For each angle, the algorithm compares the current component position with the historical component position at the corresponding angle. When the comparison results of all angles indicate that a component is missing, the inspection result marks a component missing fault.

[0067] In order to improve the quality of collected images, power inspection will also request the exposure anomaly correction network from the server. When the exposure of the image taken is abnormal due to the switching of the drone angle, the inspection service will retrieve the tower comparison map of the corresponding angle of the current drone and use the exposure anomaly correction network to correct the exposure of the abnormal exposure image.

[0068] The flight data and test results of each inspection are sent to the data storage end after the inspection is completed and returned, and data processing is performed to update the corresponding models and data. Among them, the flight data of the inspection includes: image data, drone location and elevation information.

[0069] Finally, based on the above method, the present invention provides a UAV-based power inspection system, including: an inspection client, a server and a data storage terminal; the inspection client is installed on the UAV end, separated from the server end, to complete data collection, data upload and download, and power inspection functions, wherein the power inspection function receives processed offline data from the server end to complete the corresponding function, and at the same time, the client end has a local data storage module; the server end runs a special server to perform big data and large model calculations to complete the data processing function; the data storage terminal runs independently to complete the data storage function.

[0070] During the actual inspection, the inspection client requests the required data from the server and then goes to the field for inspection. The requested data mainly includes the tower coordinate system information, the relative position coordinates and numbers of the tower components, where the component numbers are as follows: Figure 3 As shown in the figure, A represents the component type of the insulator (i.e. T=A), A 1_1 Indicates the first insulator of the first tower, A 1_2 Indicates the second insulator of the first tower, A 1_3 It represents the third insulator of the first tower. The same part number at different angles is the same. The server processes the data and calculates the obstacle avoidance zone of each tower. The obstacle avoidance zone diagram is shown in the figure below. Figure 4 As shown in the figure, for easy viewing, all heights are relative to the ground, and widths are relative to the concrete pole axis of the tower. Based on this, during inspection, the current drone position is calculated in real time, and an alarm is immediately triggered when the drone position is within the obstacle avoidance zone. Furthermore, during inspection, inspectors can use automatically planned routes for automatic inspections. When inspectors deem it necessary to collect other data or manually discover problems, they can switch to manual inspections. Among them, automatic inspections, manual inspections, and obstacle avoidance reminders are all run in real time.

[0071] During the inspection process, the inspection client can perform fault detection of power components in real time and complete the inspection task. Since the present invention uses multi-angle data for detection and finally summarizes the multi-angle detection results, Figure 6 As shown, therefore, the obstructed objects can be detected. At the same time, multiple detections of a component can improve the detection accuracy. And because the client has requested the previous pictures of the tower at corresponding angles, the client only needs to compare the number and position of the components in the original picture to determine whether the current component is missing and where the missing position is, thereby solving the problem of missing component detection. At the same time, the inspection client can dynamically adjust the camera so that the image data collected by the camera always keeps the center of the component as the center of the image, ensuring that the image captures valid information. Moreover, since the pictures of the corresponding angles of the component history are saved, when the exposure of the collected image is abnormal, the exposure anomaly correction network can extract the normal exposure information of the angle, and combine with the neural network to perform exposure anomaly repair. Compared with the method without normal exposure picture comparison, the present invention can produce more effective effects, such as Figure 7 shown.

[0072] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection claimed by the present invention.

Claims

1. A power inspection method based on drone, characterized by: The power inspection method comprises: data initialization and power inspection; Data initialization is performed by manually remote-controlled drones to collect data in two stages: the first stage is vertical flight along the tower to the top of the tower, then horizontal flight along the cross arm, and the second stage is flying around the tower. The data is packaged and compressed by tower and uploaded to the server backend. The server processes the data after receiving the flight data. Data processing is done by processing the latitude, longitude and altitude information of the drone during flight to generate a tower coordinate system based on the tower. i And finally generate the inspection parameter file, where i represents the i-th tower; The processed data is manually checked and then stored persistently; The power inspection includes an inspection client. After the inspection client requests and obtains the inspection parameter file and detection model parameters from the server, it first calibrates the GPS and elevation information of the drone, then automatically designs the inspection route according to the restriction information in the inspection parameter file, and completes the automatic inspection through real-time drone position solution; Data processing extracts the uploaded flight data and processes it. The specific steps are as follows: First, the data processing analyzes the flight data by tower. In the first stage of vertical flight data, a target detection network is used to identify tower components. The identification types include at least: tower, cross arm, insulator, wire; In the first stage of vertical flight, the drone records the take-off point altitude H0 and the corresponding height of the cross arm L j , where j starts from 0, indicating the crossarms arranged in order from bottom to top, until the top of the tower is at the center of the image, the drone elevation information is recorded, denoted as H1; in the first stage of horizontal flight, when the tower is detected to appear at the center of the image, the drone GPS position information is recorded (V x0 , V y0 ), save the corresponding image to the data storage end, and record the GPS location information of the drone when the longest cross arm is detected at the center of the image (V x1 , V y1 ), when the other end of the longest cross arm is detected at the center of the image, the GPS position information of the drone is recorded (V x2 , V y2 ); Establish a tower coordinate system with the current intersection of the tower and the ground as the center, the concrete pole upward as the Z axis, the eastward as the X axis, and the northward as the Y axis. i， At the same time, record the coordinate system O i The origin corresponds to the longitude and latitude O i0 (V x0 , V y0 , H0), the average center latitude and longitude height of all cross arms on the tower is O i1 (V x0 , V y0 , (H1- L0) / 2 + L0), where i represents different towers; in the second flight phase, the UAV flies around the tower and uses the target detection network to identify the tower components of the flight image data. The identification types include at least: towers, crossarms, and insulators. According to the recognition results, based on the current GPS position of the UAV and the tower coordinate system O i Calculate the quadrant to which the drone belongs and select six angles of uniform distribution starting from the north direction with the tower as the center. Based on the vertical reference of the tower concrete pole and the horizontal reference of the cross arm, the coordinates of the tower components are expressed by calculating the ratio of the pixel distance from the center of the marked target to the reference object and the pixel length of the reference object itself, and the relative coordinates of the tower components (X rela_k ,Y rela_k ) and marked with number T i_k , the image data of six angles are saved to the data storage end; At the same time, data processing also calculates the tower obstacle avoidance area BA i The obstacle avoidance area takes L0-H1 as the horizontal section and (V x1 ,V y1 )—(V x2 , V y2 ) is the vertical section, the section direction is the vertical plane direction between the cross arm and the concrete pole, and the longest obstacle avoidance length is the length L corresponding to the longest cross arm max , expressed as |(V x1 , V y1 )- (V x2 , V y2 )|, that is (V x1 , V y1 ) point to (V x2 , V y2 ) distance, the obstacle avoidance depth is infinite. When there are multiple cross arms with different directions, the obstacle avoidance zones in different directions are perpendicular to the corresponding cross arms. In terms of obstacle avoidance length, detection (V x0 , V y0 ) position, the crossarm in the drone image is calculated by comparing the ratio of the image pixel length of different crossarms to the longest crossarm and the actual length L of the longest crossarm. max The final obstacle avoidance area BA is calculated proportionally. i It is obtained by the intersection of the horizontal and vertical obstacle avoidance zones plus the safety distance D; The power inspection method also detects component missing. For each angle, the algorithm compares the current component position with the historical component position at the corresponding angle. When the comparison results of all angles indicate that a component is missing, the inspection result marks a component missing fault.

2. The power inspection method based on drone according to claim 1 is characterized in that: The two-stage data collection is specifically as follows: In the first stage, the drone is manually controlled to take off from a safe position and fly vertically along the tower. The drone camera is facing the tower horizontally. After flying to a safe distance beyond the tower, the drone turns the camera direction downward and flies horizontally along the longest cross arm of the tower until it reaches the other end of the longest cross arm, completing the first stage of data collection; In the second stage, the route is manually planned and the drone flies around the tower with the camera always facing the tower. After flying around once, the second stage of flight is completed.

3. The power inspection method based on drone according to claim 1 is characterized in that: The persistently saved data includes at least: the flight data collected and uploaded, the inspection parameter file data generated by data processing, and the model parameter data of target detection, target tracking, and exposure correction. The flight data includes at least: the altitude of the drone take-off point H0, the altitude of the drone when it is flush with the top of the tower H1, the corresponding height of the cross arm L j ; Inspection parameters include at least: tower coordinate system O i The origin corresponds to the longitude and latitude O i0 (V x0 , V y0 , H0), the average center latitude and longitude height of all cross arms on the tower O i1 (V x0 ,V y0 , (H1- L0) / 2 + L0), the final obstacle avoidance area BA i , six angles around the tower and seven pictures from the top facing the tower, and the relative coordinates of the tower components; the model parameter data for target detection, fault detection, target tracking, and exposure correction are quantized and compressed data.

4. The power inspection method based on drone according to claim 1 is characterized in that: For the power inspection, the drone first flies to the center of the tower top and records the current GPS position and elevation (V x , V y ,H), through i0 (V x0 , V y0 , H0) to calibrate GPS and elevation. The elevation correction directly uses the elevation of the take-off point of the drone and O i0 The height H0 is compared and corrected, and the UAV is subsequently used for circumferential flight inspection. The inspection client service performs real-time UAV position calculation, target identification and fault detection. Among them, the tower component corresponding to the fault uses the relative cross arm and concrete pole to calculate the relative coordinates (X rela , Y rela ) and the current angle of the drone tower component coordinates (X rela_k ,Y rela_k ) to calculate the number T corresponding to the faulty component i_k , where T represents the component type number. Different components have different type numbers. i_k represents the kth T-type component on the i-th tower.

5. The power inspection method based on drone according to claim 1 is characterized in that: When the power inspection is running, the inspection client uses the drone's GPS and elevation information to calculate the drone's position relative to the tower in real time. When the drone's position intersects the tower's obstacle avoidance zone, the client issues an obstacle avoidance warning.

6. The power inspection method based on drone according to claim 1 is characterized in that: The power inspection method also accepts the flight route of the drone when the service end collects tower data. Before the inspection flight, the inspection end uses the obstacle avoidance area BA i To smooth the collection route for the restriction conditions, complete the automatic route planning. During the inspection, the drone collects the current wind speed data. When the wind speed V changes, the inspection client updates the safety distance D and adjusts the obstacle avoidance area BA i , update the inspection flight route.

7. The power inspection method based on drone according to claim 1 is characterized in that: During the drone inspection, the current position is solved in real time and the current camera attitude is dynamically calculated. By dynamically adjusting the camera attitude angle, the drone camera is always facing the mean center of all crossarms of the tower O i1 (V x0 , V y0 , (H1- L0) / 2 + L0), and obtain the image data facing the tower components.

8. The power inspection method based on drone according to claim 1 is characterized in that: The drone inspection adopts a simple method, which includes: when the drone flies in a circle to detect power faults, the drone calculates the current position in real time and selects the same six angles as those used for data collection to take pictures. Finally, the drone flies to the top of the tower and collects pictures of the tower components facing the top. The collected data is used for real-time target detection and fault detection, and the detection results of each angle are compared with the corresponding tower component category T i_k The association is performed, and the final detection result is obtained by summarizing and unioning the detection results of the seven angles.

9. The power inspection method based on drone according to claim 1 is characterized in that: The power inspection method will also request an exposure anomaly correction network from the server. When the exposure of the captured image is abnormal due to the switching of the drone angle, the inspection end service retrieves the tower comparison map of the corresponding angle of the current drone and uses the exposure anomaly correction network to perform exposure correction on the exposure anomaly image.

10. The power inspection method based on drone according to claim 1, characterized in that: The flight data and test results of each inspection are sent to the data storage end after the inspection is completed and returned, and data processing is performed to update the corresponding models and data. Among them, the flight data of the inspection includes: image data, drone location and elevation information.

11. An electric power inspection system based on drones, characterized in that: The system completes the UAV-based power inspection method described in any one of claims 1-10, including: the inspection client is installed on the UAV end, separated from the server end, and completes data collection, data upload and download, and power inspection functions, wherein the power inspection function receives processed offline data from the server end to complete the corresponding function, and at the same time, the client has a local data storage module; the server end runs a special server to perform big data and large model calculations to complete the data processing function; the data storage end runs independently to complete the data storage function.

Images