High-voltage line laser obstacle removing self-adaptive cutting system

Abstract

The invention discloses a high-voltage line laser obstacle clearance self-adaptive cutting system, and the system comprises a data processing module which is used for collecting image data and carrying out the preprocessing of the image data; the motion platform module is used for driving the laser obstacle removing equipment to move and realize rotary motion; the visual identification module is used for carrying out pixel-level segmentation on the image, identifying the type of a foreign matter and marking a bounding box; the path planning module is used for extracting a foreign matter contour point set, constructing a grid model and generating an optimal cutting path; the electronic fence module is used for setting a safe area; the power adjusting module is used for setting laser output power and focusing parameters; and the energy communication module is used for supplying power to each functional module and performing data interaction with a remote monitoring platform through a 5G network. According to the invention, intelligent identification, path planning and laser self-adaptive removal of foreign matters of the high-voltage line are realized, and the system has the beneficial effects of high efficiency, safety and full-automatic operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-voltage power facility maintenance, and particularly to a high-voltage line laser obstacle removal adaptive cutting system. BACKGROUND

[0002] In the operation process of a high-voltage transmission line, the line is often disturbed by non-metallic foreign objects from the external environment, such as a kite string, a plastic film, a dust screen, a nylon rope, and a tree branch, etc. Once these foreign objects are entangled on the conductor, they may cause a short circuit, a trip, a discharge, a spark, or even a fire, which seriously threatens the safe and stable operation of the power grid. In order to ensure the reliability of the transmission line, regular line obstacle removal operations have become a key link in the power operation and maintenance work.

[0003] At present, the main high-voltage line obstacle removal methods mainly include manual tower climbing removal and unmanned aerial vehicle mechanical arm operation. Manual operation needs to be operated with power off, which is time-consuming and has high safety risks, especially in complex terrain or high-altitude operation scenarios. Although the unmanned aerial vehicle carrying a mechanical device can realize partial non-contact operation, it is difficult to effectively handle slender and flexible foreign objects (such as a kite string) due to the limitations of the load, flight stability, and control accuracy in a strong electromagnetic interference environment.

[0004] Some systems also use handheld or fixed laser devices for obstacle removal. The basic principle is to use a high-energy laser beam to achieve the melting or vaporization of non-metallic foreign objects. However, such devices generally have problems such as path dependence on manual setting, lack of image recognition capability, and laser power adjustment depending on manual experience. Some devices do not have a safety area recognition mechanism, which may cause damage to the conductor fittings or insulators, and need to be manually monitored during operation, which has a low degree of automation.

[0005] In addition, the existing laser obstacle removal system lacks real-time tracking and path dynamic correction capability when dealing with dynamic environments (such as foreign object swinging caused by wind blowing and line vibration), resulting in low laser cutting efficiency, poor reliability, and discontinuous and unstable obstacle removal operation. At the same time, different materials and distances of foreign objects require different laser power and focusing settings, and the existing technology generally does not realize real-time adaptive adjustment based on target type and distance, which further limits the wide application of the system.

[0006] Therefore, how to provide a high-voltage line laser obstacle removal adaptive cutting system is a problem that needs to be solved by those skilled in the art. SUMMARY

[0007] One purpose of the present application is to propose a high-voltage line laser obstacle cleaning adaptive cutting system, which comprehensively applies image semantic segmentation, target detection, path planning, laser power adaptive adjustment and electronic fence control and other technologies, and describes in detail the whole process of realizing foreign matter identification, path generation, dynamic tracking and laser accurate cutting in the high-voltage line environment, which has the advantages of high automation, high obstacle cleaning efficiency and strong operation safety.

[0008] The high-voltage line laser obstacle cleaning adaptive cutting system according to the embodiment of the present application comprises:

[0009] A data processing module is configured to collect image data of the high-voltage line and pre-process the image data through an edge processing device;

[0010] A motion platform module is configured to drive the laser obstacle cleaning device to move in horizontal and vertical directions and realize 360-degree rotary motion;

[0011] A visual recognition module is configured to perform pixel-level segmentation on the pre-processed image by using an improved DeepLabv3+ semantic segmentation model, and identify the type of foreign matter and label a bounding box by combining a target detection algorithm based on YoloV5;

[0012] A path planning module is configured to extract a foreign matter contour point set according to the image segmentation result, construct a grid model by using Delaunay triangulation, and generate an optimal cutting path based on an A* algorithm, wherein the path supports straight line, curve and polygon modes, and automatically avoids a safety prohibited area set around the high-voltage device;

[0013] An electronic fence module is configured to set a safety area in a three-dimensional model of the high-voltage line device, and when a laser spot approaches a boundary of a prohibited area defined by the electronic fence module, automatically stop laser output and trigger a mechanical braking device;

[0014] A power adjustment module is configured to automatically set laser output power and focusing parameters according to a target distance fed back by laser ranging and a type of foreign matter obtained by multi-spectrum recognition;

[0015] An energy communication module is configured to power each functional module and perform data interaction with a remote monitoring platform through a 5G network.

[0016] Optionally, the modules are realized by the following method:

[0017] S1, start the system power supply, and perform self-checking operation of the laser obstacle cleaning device, the edge processing device and the motion platform;

[0018] S2, collect image data of the high-voltage line operation area through a plurality of cameras, and transmit the image data to the edge processing device for image pre-processing;

[0019] S3, the improved DeepLabv3+ semantic segmentation model is used for pixel-level segmentation of the preprocessed image, and the YoloV5-based target detection algorithm is combined to identify the foreign object type and label the boundary box;

[0020] S4, according to the identification result, the foreign object contour point set is extracted, the path planning module constructs a grid model based on Delaunay triangulation, and an A* algorithm is applied to generate an optimal cutting path that avoids the safety prohibited area;

[0021] S5, the target distance data is obtained by the laser range finder and the binocular vision system, and combined with the foreign object type, the power adjustment module is called to calculate the adaptive laser output power and focusing parameters;

[0022] S6, the motion platform drives the laser head to perform cutting operation according to the optimal cutting path, and the binocular vision system monitors the foreign object displacement in real time, estimates the target motion vector by optical flow method, and dynamically adjusts the cutting path;

[0023] S7, during the laser cutting process, the electronic fence module continuously monitors the position of the light spot, and when the distance between the light spot and the safety prohibited area is less than the set threshold, the laser emission is stopped and the brake device is triggered;

[0024] S8, after cutting is completed, the system re-scans the target area to confirm the obstacle removal effect, and if residual foreign objects are detected, the path is re-planned for secondary cutting;

[0025] S9, after the operation is completed, the system generates a task report including operation time, position, foreign object type and energy consumption data, and uploads it to the remote monitoring platform through the 5G network.

[0026] Optionally, the laser obstacle removal device comprises an obstacle removal shell, a power switch, a laser generator, a transformer, a power regulator, a driver, a laser focusing lens and a focusing assembly, which removes foreign objects by emitting laser beams, and remotely cuts and removes non-metallic foreign objects on high-voltage lines. The laser beam uses a wavelength of 1080nm near-infrared band, which can penetrate non-metallic materials while being reflected by metal conductors.

[0027] Optionally, the edge processing device is equipped with an NVIDIA Jetson Xavier NX module, which performs defogging, noise reduction, contrast enhancement processing on image data, and executes path planning and sends control instructions, supports TensorRT inference optimization, and the processing delay is not higher than 100ms.

[0028] Optionally, the motion platform includes a moving platform and a rotating platform, the moving platform is composed of a support frame, a moving motor, a moving screw and a guide rail, and the rotating platform includes a support seat, a rotating motor and a rotating disc, which realizes two-dimensional translation and 360-degree rotation of the laser obstacle removal device.

[0029] Optionally, the S3 specifically comprises:

[0030] S31, input the pre-processed image into an improved DeepLabv3+ model for pixel-level semantic segmentation, the DeepLabv3+ model takes Xception network as the backbone structure, accesses a one-dimensional hollow convolution kernel group containing different inflation rates at the end of the encoder, outputs a feature map, the convolution kernel group constitutes multiple parallel branches, each branch processes different scale feature channels, and all branch outputs are fused in the decoding stage and combined with the feature map for up-sampling to output a pixel-level mask of the foreign object area;

[0031] S32, splice and fuse the pixel-level mask with the pre-processed image, and input it into a target detection model based on YoloV5 structure, the target detection model introduces a CBAM channel-spatial attention module in the Backbone stage, and inserts an SPPF module in the Neck structure to extract context information of different receptive field ranges in the feature map in a parallel manner, and outputs a detection result containing target categories, bounding box coordinates and category confidence;

[0032] S33, joint encoding of the bounding box coordinates, category confidence and mask in the detection result and output.

[0033] Optionally, the splicing and fusion process specifically includes the following:

[0034] Align the size of the three channels of the pre-processed image with the mask image, and unify the feature dimension of the image;

[0035] Normalize the mask image to adjust the pixel value range to between 0 and 1;

[0036] In the feature channel dimension, the mask image is attached as the fourth channel after the RGB channel of the pre-processed image to form a four-channel fusion image;

[0037] Apply linear transformation to compress the fusion image to a three-channel tensor, and use the pixel-by-pixel weighted average strategy to generate a new fusion image as the input of the target detection model.

[0038] Optionally, the S4 specifically comprises:

[0039] S41, receive the foreign object bounding box information and the pixel-level mask, perform image contour extraction operation, extract the foreign object edge contour through the gradient-based edge detection operator, obtain the edge point set constituting the foreign object outer contour, and perform denoising and smoothing processing on the point set to form a sparse but continuous contour point sequence;

[0040] S42, input the contour point sequence to the path planning module, use the Delaunay triangulation algorithm to construct a non-intersecting grid structure connecting all the contour points, each edge of the non-intersecting grid structure represents a potential laser cutting path segment, and a directed graph structure is established on the basis of the grid topology, the nodes represent the spatial positions in the edge point set, the edges represent the path segments, and each edge is assigned a weight value;

[0041] S43, based on the safety prohibited area preset by the electronic fence module, mark the edges and nodes in the non-intersecting grid structure that fall within the safety prohibited area, and set them as non-passable areas, which are excluded in the path planning module;

[0042] S44, apply A* algorithm to the constructed non-intersecting grid structure, set the cutting starting point as the center point of the foreign object and the wire contact area, and the end point as the boundary point of the free end of the foreign object, perform shortest path search according to the edge weight and path cumulative cost, and dynamically generate the cutting path sequence;

[0043] S45, path smoothing and redundant segment elimination processing is performed on the cutting path sequence to form a continuous executable cutting trajectory, the trajectory supports mixed combination of straight line segments, curve segments and polygon segments, and corresponding coordinate point sets and cutting order indexes are labeled when the path is output, for motion platform to call and execute.

[0044] Optionally, the S5 specifically includes:

[0045] S51, the laser range finder emits a pulse light beam to the center point of the foreign object, records the round-trip flight time and converts it into a first distance value, the binocular vision system captures a stereo image of the foreign object, calculates a second distance value through parallax, and performs consistency check on the first distance value and the second distance value, if the difference is within the set tolerance range, take the average value as the target distance, otherwise retrigger the measurement process until the consistency condition is met;

[0046] S52, the foreign object type label output by the visual recognition module is input to the power adjustment module, the power adjustment module retrieves the preset material parameter table to obtain the energy coefficient and focal point coefficient corresponding to the foreign object type, the preset material parameter table includes the energy absorption coefficient, optimal cutting power range, corresponding focusing distance and laser spot diameter parameters of a plurality of non-metallic foreign objects, which are used to provide matching laser adjustment reference values for different types of foreign objects, and the target distance, energy coefficient and focal point coefficient after verification are used for power calculation to generate laser output power instructions and focusing lens displacement instructions;

[0047] S53, the driver in the laser obstacle clearing equipment receives the laser output power instruction, adjusts the output current of the laser generator in real time to set the output power, the focusing assembly drives the connecting rod mechanism to move the focusing lens according to the focusing lens displacement instruction, realizes the synchronous control of the spot diameter and the power density, and feedbacks the state signal to the power adjustment module for closed loop verification after the adjustment is completed, and sends an execution completion flag to the motion platform.

[0048] Optionally, the S6 specifically comprises:

[0049] S61, the motion platform receives the smoothed cutting path sequence output by the path planning module, and drives the laser obstacle clearing equipment through the control of the moving motor and the rotating motor according to the spatial coordinates and the cutting sequence index of each point in the path, so as to perform two-dimensional translation and rotation motion in the horizontal and vertical directions, so that the laser head completes each cutting action along the set trajectory in turn, and the current position and path deviation are recorded in real time during the execution process by the path controller;

[0050] S62, the binocular vision system synchronously collects the foreign matter image sequence in the cutting process, inputs the image stream into the edge processing equipment, calculates the feature point displacement of the foreign matter edge area between continuous frames based on the optical flow method, obtains the motion vector information of the actual boundary of the foreign matter, and compares with the current execution node in the path controller to judge whether there is target displacement phenomenon;

[0051] S63, when the edge processing equipment judges that the target has displacement and the current path no longer covers the complete boundary of the foreign matter, a path reconstruction operation is performed, the executed path is truncated, and the cutting trajectory of the remaining boundary area is recalculated, the path reconstruction is dynamically generated according to the new pose and motion direction of the foreign matter provided by the binocular vision system, and is spliced with the original path to form an updated execution path, and the reconstructed path is uploaded to the motion platform to complete the adaptive cutting process.

[0052] The beneficial effects of the present application are:

[0053] Firstly, the improved DeepLabv3+ semantic segmentation model and the target detection algorithm based on YoloV5 are introduced, pixel-level recognition and category judgment of the foreign matter are realized, Delaunay triangulation and A* algorithm in the path planning module are combined, the optimal cutting path avoiding the high-voltage equipment can be automatically generated, human intervention is effectively reduced, and the accuracy of recognition and operation is significantly improved.

[0054] Secondly, the application constructs a distance perception mechanism based on laser ranging and binocular vision fusion, combines multi-spectral material identification and pre-set material parameter table, realizes automatic adjustment of laser power and focusing parameters, and through the motion platform accurately controls the cutting action of the laser head, and under the interference of wind or vibration, still can estimate the target displacement through the optical flow method and correct the path in real time, guarantee the cutting continuity and stability in complex environment.

[0055] Finally, the application embeds an electronic fence module in the system, through three protection means of safety forbidden area setting, dynamic target identification and emergency braking mechanism, ensures the protection of key components such as wires, fittings and insulators during laser operation, and improves the operation safety and reliability of the whole system in unattended state. BRIEF DESCRIPTION OF DRAWINGS

[0056] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, together with the embodiments of the application, to explain the application, and do not constitute a limitation on the application. In the drawings:

[0057] Figure 1 The module structure diagram of the high-voltage line laser obstacle removal adaptive cutting system proposed by the application;

[0058] Figure 2 The method flow chart of the high-voltage line laser obstacle removal adaptive cutting system proposed by the application;

[0059] Figure 3 The path planning and laser power adaptive control flow chart of the high-voltage line laser obstacle removal adaptive cutting system proposed by the application. DETAILED DESCRIPTION

[0060] The application will now be described in further detail with reference to the drawings. These drawings are simplified schematic diagrams, and only illustrate the basic structure of the application in a schematic manner, and therefore only show the components related to the application.

[0061] Reference Figure 1 The high-voltage line laser obstacle removal adaptive cutting system comprises:

[0062] The data processing module is used for collecting image data of the high-voltage line, and pre-processing the image data through an edge processing device;

[0063] The motion platform module is used for driving the laser obstacle removal device to move in horizontal and vertical directions and realize 360-degree rotary motion;

[0064] The visual recognition module is used for performing pixel-level segmentation on the pre-processed image by using an improved DeepLabv3+ semantic segmentation model, and identifying the foreign matter type and labeling the boundary box by combining a target detection algorithm based on YoloV5.

[0065] A path planning module is configured to extract foreign object contour point sets according to the image segmentation result, construct a grid model by using Delaunay triangulation, and generate an optimal cutting path based on an A* algorithm, wherein the path supports straight line, curve and polygon modes, and automatically avoids a safety forbidden area set around the high-voltage device;

[0066] An electronic fence module is configured to set a safety area in the three-dimensional model of the high-voltage line device, and automatically stop laser output and trigger a mechanical braking device when a laser spot approaches the boundary of the forbidden area defined by the electronic fence module;

[0067] A power adjustment module is configured to automatically set laser output power and focusing parameters according to target distance feedback of laser ranging and foreign object types obtained by multi-spectral recognition;

[0068] An energy communication module is configured to power each functional module and perform data interaction with a remote monitoring platform through a 5G network.

[0069] The present application integrates seven modules of data processing, motion platform, visual recognition, path planning, electronic fence, power adjustment and energy communication, and constructs a complete and efficient high-voltage line laser obstacle removal system, realizes intelligent operation from image acquisition to laser cutting, and improves the automation level and safety guarantee capability of obstacle removal.

[0070] Reference Figures 2-3 In the embodiment, the modules are realized by the following method:

[0071] S1, start the system power supply, and perform self-checking operation of the laser obstacle removal device, the edge processing device and the motion platform;

[0072] S2, collect image data of the high-voltage line operation area by the plurality of cameras, and transmit the image data to the edge processing device for image preprocessing;

[0073] S3, perform pixel-level segmentation on the preprocessed image by using an improved DeepLabv3+ semantic segmentation model, and identify the foreign object type and label the boundary box by combining a target detection algorithm based on YoloV5;

[0074] S4, extract foreign object contour point sets according to the identification result, and the path planning module constructs a grid model based on Delaunay triangulation, and generates an optimal cutting path that avoids the safety forbidden area by using an A* algorithm;

[0075] S5, obtain target distance data by a laser range finder and a binocular vision system, and combine the foreign object type to call the power adjustment module to calculate the adaptive laser output power and focusing parameters;

[0076] S6, the motion platform drives the laser head to perform the cutting operation according to the optimal cutting path, and the binocular vision system monitors the displacement of the foreign matter in real time, estimates the target motion vector through the optical flow method, and dynamically adjusts the cutting path;

[0077] S7, during the laser cutting process, the electronic fence module continuously monitors the spot position, stops laser emission and triggers the brake device when the spot distance from the safety prohibited area is less than the set threshold;

[0078] S8, after the cutting is completed, the system re-scans the target area to confirm the obstacle removal effect, and if residual foreign matter is detected, the path is re-planned to perform secondary cutting;

[0079] S9, after the operation is completed, the system generates a task report including operation time, position, foreign matter type and energy consumption data, and uploads it to the remote monitoring platform through the 5G network.

[0080] The present application provides a complete operation process covering identification, planning, adjustment, execution and feedback, and clearly defines the coordination mechanism between system modules, ensuring orderly connection of each step, accurate information transmission, and ensuring efficient execution of obstacle removal tasks and remote operation capability.

[0081] In the embodiment, the laser obstacle removal device comprises an obstacle removal shell, a power switch, a laser generator, a transformer, a power regulator, a driver, a laser focusing lens and a focusing assembly, which removes foreign matter by emitting a laser beam, and remotely cuts and removes non-metallic foreign matter on high-voltage lines.

[0082] The present application realizes non-contact remote removal of non-metallic foreign matter by using a near-infrared waveband laser beam combined with a precision focusing device, effectively solving the risk problem of traditional mechanical cutting methods in high-voltage live environments, and improving the safety and scope of operation.

[0083] In the embodiment, the edge processing device is equipped with an NVIDIA Jetson Xavier NX module, which performs defogging, noise reduction, contrast enhancement processing on image data, and executes path planning and sends control instructions, supports TensorRT inference optimization, and the processing delay is not higher than 100ms.

[0084] The present application adopts an edge processing device equipped with Jetson Xavier NX for image processing and control decision-making, and uses a high-performance computing platform to realize image defogging, enhancement and path control operations, with the advantages of fast processing speed and timely system response.

[0085] In the embodiment, the motion platform comprises a moving platform and a rotating platform, the moving platform is composed of a support frame, a moving motor, a moving screw rod and a guide rail, and the rotating platform comprises a support seat, a rotating motor and a rotating disc, so that two-dimensional translation and 360-degree rotation of the laser obstacle clearing device are realized.

[0086] The application realizes flexible positioning and accurate cutting track execution of the laser device in a three-dimensional space by constructing a composite motion platform with translation and rotation capabilities, and meets the multi-angle operation requirements in a complex foreign matter distribution scene.

[0087] In the embodiment, the S3 specifically comprises:

[0088] S31, input the preprocessed image into an improved DeepLabv3+ model for pixel-level semantic segmentation, the DeepLabv3+ model takes an Xception network as a backbone structure, a one-dimensional hollow convolution kernel group containing different expansion rates is connected at the end of the encoder, and a feature map is output, the convolution kernel group constitutes a plurality of parallel branches, each branch processes a feature channel of different scales, and all branch outputs are fused in the decoding stage and combined with the feature map for up-sampling to output a pixel-level mask of the foreign matter area, the mask matrix is calculated as follows:

[0089]

[0090] Wherein, M i,j represents the mask probability value of the pixel point (i, j), f k (x i,j ) is the response value of the kth channel at the pixel point (i, j), w k is the channel weighting coefficient, b is the bias term, sigma is the Softmax function, and C represents the number of feature channels.

[0091] S32, the pixel-level mask is spliced and fused with the preprocessed image and input into a target detection model based on the YoloV5 structure, the CBAM channel-spatial attention module is introduced in the Backbone stage, the SPPF module is inserted in the Neck structure to extract the context information of different receptive field ranges in the feature map in a parallel manner, and the detection result containing the target category, the bounding box coordinates and the category confidence is output, and the target detection score is calculated as follows:

[0092]

[0093] Wherein, S is the detection score, IOU is the intersection over union of the predicted bounding box and the mask outline, D is the category confidence, the value range is 0 to 1, log2(·) represents the logarithmic function, and alpha and beta are adjustment coefficients.

[0094] S33, combine and encode the bounding box coordinates and class confidence in the detection result with the mask, and output.

[0095] The application combines a semantic segmentation model and a target detection algorithm, realizes high-precision identification of foreign object types and boundaries through fine image analysis, provides a reliable perception basis for subsequent path planning and power adjustment, and enhances the identification accuracy and processing consistency of the system.

[0096] In the embodiment, the specific process of splicing and fusion is as follows:

[0097] The three channels of the preprocessed image are size-aligned with the mask image, and the feature dimension of the image is unified;

[0098] The mask image is normalized to adjust the pixel value range to 0 to 1;

[0099] In the feature channel dimension, the mask image is attached as the fourth channel after the RGB channel of the preprocessed image to form a four-channel fusion image;

[0100] The linear transformation is applied to the fusion image to compress it to a three-channel tensor, and a new fusion image is generated by using the pixel-by-pixel weighted average strategy as the input of the target detection model.

[0101] The application designs a specific image and mask fusion processing procedure, so that the target detection model can receive the fused information input, fully utilize the segmentation features based on the original image, and improve the precision and structural integrity of the target detection result.

[0102] In the embodiment, the S4 specifically comprises:

[0103] S41, receiving foreign object bounding box information and pixel-level mask, applying an edge detection operator to obtain foreign object edge contour, and extracting a contour point set, denoted as a two-dimensional coordinate set, and sorting, denoising and boundary point spacing resampling processing are performed on the two-dimensional coordinate set to form a sparse and continuous contour point sequence, and the construction function of the contour point set is as follows:

[0104]

[0105] wherein P(x,y) represents the spatial coordinates of the effective contour points, is the original image gray gradient, M(x,y) is the pixel value of the mask image at the corresponding position, and arg max is the independent variable value when the function takes the maximum value;

[0106] S42, input the processed contour point sequence into a path planning module, perform a Delaunay triangulation operation based on the contour point set, construct a non-intersecting two-dimensional triangular mesh graph, and set a path cost according to the Euclidean distance of the two end points of each mesh edge, while eliminating the forbidden area defined by the electronic fence module, to construct a passable area graph G;

[0107] S43, in the constructed two-dimensional triangular mesh graph, taking the foreign object and the conductor contact point as the starting node and the foreign object free end boundary point as the target node, performing an A* algorithm path search, the cost function using the sum of the heuristic distance and the cumulative path weight, and the path search function being defined as follows:

[0108] F(n)=G(n)+H(n);

[0109] Wherein, F(n) is the total cost function of the current node, G(n) is the actual cumulative path cost from the starting point to the current node, and H(n) is the estimated cost of the current node to the target node.

[0110] S44, converting the node sequence in the path search result into a continuous cutting trajectory, performing curvature smoothing and angle correction processing, generating a trajectory set with coordinate labels and cutting sequence indexes, and sending to the motion platform module to perform a laser head cutting action.

[0111] The present application realizes path modeling of the foreign object edge by constructing a non-intersecting triangular mesh and a directed graph structure, and generates an efficient cutting path under the condition of obstacle avoidance in combination with the A* algorithm, thereby significantly reducing path redundancy and equipment operation cost.

[0112] In the embodiment, the S5 specifically comprises:

[0113] S51, the laser range finder emits a pulsed light beam to the foreign object center point, records the round-trip flight time and converts the first distance value, the binocular vision system synchronously captures a stereo image of the foreign object, calculates the second distance value through parallax, and performs consistency check on the first distance value and the second distance value, if the difference is within the set tolerance range, the average value is taken as the target distance, otherwise the measurement process is retriggered until the consistency condition is met.

[0114] S52, the foreign object type label output by the visual recognition module is input to the power adjustment module, the power adjustment module retrieves a preset material parameter table to obtain the energy coefficient and the focal point coefficient corresponding to the foreign object type, the preset material parameter table includes the energy absorption coefficient, the optimal cutting power range, the corresponding focusing distance and the laser spot diameter parameter of a plurality of non-metallic foreign objects, and is used to provide matched laser adjustment reference values for different types of foreign objects, and the target distance, the energy coefficient and the focal point coefficient after the check are used for power calculation to generate a laser output power instruction and a focusing lens displacement instruction.

[0115] S53, the driver in the laser obstacle clearing equipment receives the laser output power instruction, adjusts the output current of the laser generator in real time to set the output power, the focusing assembly drives the connecting rod mechanism to move the focusing lens according to the focusing lens displacement instruction, realizes the synchronous control of the spot diameter and the power density, and feeds back the state signal to the power adjustment module for closed loop verification after the adjustment is completed, and sends an execution completion flag to the motion platform.

[0116] The application realizes fine adjustment of power and focusing based on the laser ranging and binocular vision combined ranging mechanism, in combination with a foreign matter material parameter table, ensures matching of laser energy and foreign matter, improves cutting efficiency and reduces a heat influence range.

[0117] In the embodiment, the S6 specifically comprises:

[0118] S61, the motion platform receives the smoothed cutting path sequence output by the path planning module, drives the laser obstacle clearing equipment by controlling the moving motor and the rotating motor according to the spatial coordinates and the cutting sequence index of each point in the path, performs two-dimensional translation and rotation motion in the horizontal and vertical directions, and makes the laser head complete each cutting action along the set trajectory in turn, and the current position and path deviation in the execution process are recorded by the path controller in real time;

[0119] S62, the binocular vision system synchronously collects the foreign matter image sequence in the cutting process, inputs the image stream into the edge processing equipment, calculates the feature point displacement of the foreign matter edge region between continuous frames based on the optical flow method, obtains the motion vector information of the actual boundary of the foreign matter, and compares with the current execution node in the path controller to judge whether there is target displacement phenomenon;

[0120] S63, when the edge processing equipment judges that the target has displacement and the current path no longer covers the complete boundary of the foreign matter, a path reconstruction operation is performed, the executed path is truncated and the cutting trajectory of the remaining boundary region is recalculated, the path reconstruction is dynamically generated according to the new pose and motion direction of the foreign matter provided by the binocular vision system, and is spliced with the original path to form an updated execution path, and the reconstructed path is uploaded to the motion platform to complete the adaptive cutting process.

[0121] The application can update the cutting path in real time under the condition that the foreign matter moves or shakes through the dynamic path correction mechanism and the optical flow tracking method, guarantees the target coverage degree in the continuous cutting process, and effectively improves the cutting stability and system robustness in a complex environment.

[0122] Example 1:

[0123] To verify the feasibility of the application in implementation, the application is applied to unmanned inspection and obstacle removal operation of a high-voltage transmission line corridor in a mountainous area. The line is long-term affected by wind, and light foreign matters such as plastic film, dust cloth, nylon rope and kite line often wrap around the conductor or insulator. In some areas, the high-altitude operation conditions are poor, and manual tower climbing or insulating arm car operation conditions are not available, so an efficient and safe remote obstacle removal method is needed to ensure the stability of the line operation.

[0124] In this scenario, the application is deployed on a mobile high-altitude platform, uses solar power as an auxiliary power supply to support long-time operation, and accesses a remote control center through 5G communication. In actual operation, the system first scans the conductor area by multiple high-definition cameras, processes the images through an edge processing module for defogging, denoising and contrast enhancement, and generates clear operation images. Then, the system uses an improved DeepLabv3+ model for pixel-level foreign matter segmentation, extracts foreign matter contour masks, and further identifies foreign matter categories and spatial boundaries through a YoloV5 model.

[0125] In a typical operation, the system identifies 7 independent foreign matters, including 3 plastic films, 2 nylon ropes and 2 kite lines. The system automatically generates a cutting path for each foreign matter and sorts the cutting order, giving priority to the interference at the conductor hanging point. Taking the nylon rope as an example, after identifying its material through spectral analysis, the system automatically sets the laser power to 1320W, the laser focusing distance to 47 meters, the spot diameter to 4.3 millimeters, and the power density to 1.2×10 6 W / cm 2 . The average cutting time is not more than 24 seconds, and the removal rate is 100%. For a wider plastic film target, the system divides it into two cutting sections and uses a segmented focusing strategy to avoid heat diffusion affecting the cutting quality, and the entire obstacle removal process is completed within 54 seconds.

[0126] During this period, the binocular vision system tracks the foreign matter edge in real time at a frame rate of 15Hz, and uses the optical flow method to analyze its displacement trend. A wind disturbance causes a section of kite line to deviate by about 3.6 cm, and the system completes path re-planning and continues cutting within 62 milliseconds, without deviation or residue. The entire operation lasts about 18 minutes, and the system corrects the path 12 times during the cutting process, with an average correction distance of less than 5 cm, showing strong environmental adaptability.

[0127] The electronic fence module is continuously enabled throughout the process. When the laser spot is detected to be less than 3 cm away from the edge of an insulator, the system automatically turns off the laser output and locks the motion platform, completing the shutdown action within 72 ms to avoid potential injury risks. After the entire obstacle removal operation is completed, the system automatically generates an operation report and uploads it to the remote monitoring platform, including data such as foreign object type, obstacle removal time, laser energy consumption, path deviation, and fault warning records.

[0128] The following table is a typical data record for this operation, showing the practical application performance and operation efficiency of the application in complex natural environments:

[0129] Table 1: Actual data statistics for obstacle removal operation

[0130]

[0131] As shown in Table 1, the application not only has high foreign object recognition accuracy and real-time dynamic response capability, but also can achieve efficient, precise, and safe remote laser cutting operation under various material and distance conditions. The system is stable in overall operation, accurate in path tracking, and meets the actual needs of unattended obstacle removal scenarios.

[0132] The above description is only a preferred specific implementation of the application, but the protection scope of the application is not limited thereto. Any skilled person in the art can make equivalent substitutions or changes to the technical solutions and inventive concepts of the application within the technical scope disclosed by the application, which should be covered within the protection scope of the application.

Claims

1. High-voltage line laser obstacle removal adaptive cutting system, characterized by: include: A data processing module is used to collect image data of high-voltage lines and pre-process the image data through edge processing equipment; Motion platform module, used to drive the laser obstacle removal equipment to move horizontally and vertically and achieve 360-degree rotation; The visual recognition module uses the improved DeepLabv3+ semantic segmentation model to perform pixel-level segmentation on the preprocessed image, and combines the object detection algorithm based on YoloV5 to identify the type of foreign objects and mark the bounding box; The path planning module is used to extract the foreign body contour point set based on the image segmentation results, construct a mesh model using Delaunay triangulation, and generate an optimal cutting path based on the A* algorithm. The path supports straight line, curve and polygon modes and automatically avoids the safety restricted area set around high-voltage equipment; The electronic fence module is used to set the safety zone in the three-dimensional model of high-voltage line equipment. When the laser spot approaches the boundary of the restricted area defined by the electronic fence module, the laser output is automatically stopped and the mechanical brake device is triggered; The power adjustment module is used to automatically set the laser output power and focusing parameters according to the target distance feedback from the laser ranging and the type of foreign matter obtained by multi-spectral recognition; The energy communication module is used to supply power to each functional module and exchange data with the remote monitoring platform through the 5G network.

2. The high-voltage line laser obstacle removal adaptive cutting system according to claim 1 is characterized in that: The modules are implemented as follows: S1. Start the system power supply and perform self-test operations on the laser obstacle removal equipment, edge processing equipment and motion platform; S2. Collect image data of the high-voltage line operation area through multiple cameras and transmit it to the edge processing device for image preprocessing; S3, using the improved DeepLabv3+ semantic segmentation model to perform pixel-level segmentation on the preprocessed image, and combining it with the object detection algorithm based on YoloV5 to identify the type of foreign objects and mark the bounding box; S4. Extract the foreign body contour point set based on the recognition results. The path planning module constructs a mesh model based on Delaunay triangulation and applies the A* algorithm to generate the optimal cutting path that avoids the safety restricted area. S5. Obtain target distance data through the laser rangefinder and binocular vision system, and call the power adjustment module to calculate the appropriate laser output power and focusing parameters based on the type of foreign matter; S6. The motion platform drives the laser head to perform the cutting operation according to the optimal cutting path, and the binocular vision system monitors the displacement of foreign objects in real time, estimates the target motion vector through the optical flow method, and dynamically adjusts the cutting path; S7. During the laser cutting process, the electronic fence module continuously monitors the position of the light spot. When the distance between the light spot and the safety restricted area is less than the set threshold, the laser emission is stopped and the braking device is triggered. S8. After the cutting is completed, the system rescans the target area to confirm the clearance effect. If any residual foreign matter is detected, the system replans the path and performs a second cutting. S9. After the operation is completed, the system generates a task report including operation time, location, foreign object type and energy consumption data, and uploads it to the remote monitoring platform through the 5G network.

3. The high-voltage line laser obstacle removal adaptive cutting system according to claim 2 is characterized in that: The laser obstacle removal equipment includes an obstacle removal housing, a power switch, a laser generator, a transformer, a power regulator, a driver, a laser focusing lens and a focusing assembly. It removes foreign objects by emitting a laser beam and remotely cuts and removes non-metallic foreign objects on high-voltage lines. The laser beam uses a wavelength in the 1080nm near-infrared band, which can penetrate non-metallic materials and be reflected by metal wires.

4. The high-voltage line laser obstacle removal adaptive cutting system according to claim 2 is characterized in that: The edge processing device is equipped with an NVIDIA Jetson Xavier NX module, which performs dehazing, noise reduction, and contrast enhancement on image data, as well as performs path planning and issues control instructions. It supports TensorRT inference optimization and has a processing delay of no more than 100ms.

5. The high-voltage line laser obstacle removal adaptive cutting system according to claim 2 is characterized in that: The motion platform includes a mobile platform and a rotating platform. The mobile platform is composed of a support frame, a mobile motor, a mobile screw and a guide rail. The rotating platform includes a support seat, a rotating motor and a rotating disk to achieve two-dimensional translation and 360-degree rotation of the laser obstacle removal equipment.

6. The high-voltage line laser obstacle removal adaptive cutting system according to claim 2 is characterized in that: The S3 specifically includes: S31. Input the preprocessed image into the improved DeepLabv3+ model for pixel-level semantic segmentation. The DeepLabv3+ model uses the Xception network as the main structure, connects a one-dimensional hole convolution kernel group with different expansion rates at the end of the encoder, and outputs a feature map. The convolution kernel group constitutes multiple parallel branches, each branch processes feature channels of different scales, and fuses all branch outputs in the decoding stage and combines the feature map for upsampling to output a pixel-level mask of the foreign body area. S32, stitching and fusing the pixel-level mask with the preprocessed image, and inputting the result into a target detection model based on the YoloV5 structure. The target detection model introduces the CBAM channel-spatial attention module in the Backbone stage, inserts the SPPF module into the Neck structure to extract the context information of different receptive fields in the feature map in parallel, and outputs the detection result including the target category, bounding box coordinates and category confidence; S33: jointly encode the bounding box coordinates, category confidence, and mask in the detection result and output them.

7. The high-voltage line laser obstacle removal adaptive cutting system according to claim 6 is characterized in that: The specific process of splicing and fusion is as follows: Align the three channels of the preprocessed image with the mask image to unify the feature dimensions of the image; Normalize the mask image and adjust the pixel value range to between 0 and 1; In the feature channel dimension, the mask image is added as the fourth channel to the RGB channels of the preprocessed image to form a four-channel fused image; A linear transformation is applied to the fused image to compress it into a three-channel tensor, and a pixel-by-pixel weighted averaging strategy is used to generate a new fused image as the input of the target detection model.

8. The high-voltage line laser obstacle removal adaptive cutting system according to claim 2 is characterized in that: The S4 specifically includes: S41, receiving foreign object bounding box information and pixel-level mask, performing image contour extraction operation, extracting foreign object edge contour by using gradient-based edge detection operator, obtaining edge point set constituting the foreign object outer contour, and performing denoising and smoothing on the point set to form a sparse but continuous contour point sequence; S42. Input the contour point sequence into the path planning module, and use the Delaunay triangulation algorithm to construct a non-intersecting grid structure connecting all contour points, where each edge of the non-intersecting grid structure represents a potential laser cutting path segment. A directed graph structure is established based on the grid topology, where nodes represent spatial locations where edge points are concentrated, edges represent path segments, and weights are assigned to each edge. S43, based on the safety restricted area preset by the electronic fence module, marking the edges and nodes in the non-intersecting grid structure that fall within the safety restricted area, setting them as impassable areas, and eliminating them in the path planning module; S44. Applying the A* algorithm to the constructed non-intersecting grid structure, setting the cutting starting point as the center point of the contact area between the foreign object and the wire, and the end point as the boundary point of the free end of the foreign object, performing a shortest path search based on the edge weight and the cumulative cost of the path, and dynamically generating a cutting path sequence; S45. Perform path smoothing and redundant segment elimination processing on the cutting path sequence to form a continuous executable cutting trajectory. The trajectory supports a mixed combination of straight line segments, curve segments, and polygonal segments. When the path is output, the corresponding coordinate point set and cutting sequence index are marked for the motion platform to call and execute.

9. The high-voltage line laser obstacle removal adaptive cutting system according to claim 2, characterized in that: The S5 specifically includes: S51. The laser rangefinder emits a pulse beam toward the center of the foreign object, records the round-trip flight time and converts it into a first distance value. The binocular vision system is synchronized to capture a stereo image of the foreign object, and a second distance value is obtained through parallax calculation. The first distance value and the second distance value are checked for consistency. If the difference is within a set tolerance range, the average value is taken as the target distance. Otherwise, the measurement process is retriggered until the consistency condition is met. S52. The foreign object type label output by the visual recognition module is sent to the power adjustment module. The power adjustment module retrieves a preset material parameter table to obtain an energy coefficient and a focus coefficient corresponding to the foreign object type. The preset material parameter table includes energy absorption coefficients, optimal cutting power ranges, corresponding focus distances, and laser spot diameter parameters for various non-metallic foreign objects. The table is used to provide matching laser adjustment reference values ​​for different types of foreign objects. The calibrated target distance, energy coefficient, and focus coefficient are used to perform power calculations to generate a laser output power instruction and a focusing lens displacement instruction. S53. The driver in the laser obstacle removal equipment receives the laser output power instruction and adjusts the output current of the laser generator in real time to set the output power. The focusing component drives the connecting rod mechanism to move the focusing lens according to the focusing lens displacement instruction to achieve synchronous control of the spot diameter and power density. After the adjustment is completed, the status signal is fed back to the power regulation module for closed-loop verification, and an execution completion flag is sent to the motion platform.

10. The high-voltage line laser obstacle removal adaptive cutting system according to claim 2, characterized in that: The S6 specifically includes: S61: The motion platform receives the smoothed cutting path sequence output by the path planning module. Based on the spatial coordinates of each point in the path and the cutting sequence index, the motion platform controls the mobile motor and the rotary motor to jointly drive the laser obstacle removal device to perform two-dimensional translation and rotation in the horizontal and vertical directions, so that the laser head sequentially completes each cutting action along the set trajectory. During the execution process, the path controller records the current position and path deviation in real time. S62. The binocular vision system synchronously collects a sequence of images of foreign objects during the cutting process, inputs the image stream into the edge processing device, calculates the displacement of feature points in the edge area of ​​the foreign object between consecutive frames based on the optical flow method, obtains the motion vector information of the actual boundary of the foreign object, and compares it with the current execution node in the path controller to determine whether there is a target offset phenomenon; S63. When the edge processing device determines that the target has shifted and the current path no longer covers the complete boundary of the foreign object, it performs a path reconstruction operation, truncates the executed path, and recalculates the cutting trajectory for the remaining boundary area. The path reconstruction is dynamically generated based on the new position and movement direction of the foreign object provided by the binocular vision system, and is spliced ​​with the original path to form an updated execution path. At the same time, the reconstructed path is uploaded to the motion platform to complete the adaptive cutting process.

Images