Obstacle identification and laser obstacle removal method for power transmission line

Abstract

The invention provides an obstacle recognition and laser obstacle clearance method for a power transmission line, and the method comprises the steps: collecting the spectral data of an obstacle through multispectral imaging, carrying out the material classification through a convolutional neural network, analyzing the form of the obstacle in combination with a depth image, and planning an obstacle clearance track through a dynamic planning algorithm. Laser obstacle clearing parameters are dynamically adjusted according to environmental factors, accurate recognition and classification of different types of obstacles such as metal, branches and bird nests are achieved, the optimal obstacle clearing track and parameters can be generated in a self-adaptive mode, real-time monitoring and adjustment are conducted in the obstacle clearing process, the intelligence and accuracy of obstacle clearing of the power transmission line are effectively improved, and the working efficiency is improved. And safe and stable operation of a power grid is ensured. According to the technical scheme of the invention, advanced technologies such as multispectral analysis, deep learning and three-dimensional imaging are fused, the whole-process intelligentization of obstacle removal of the power transmission line is realized, and the method has important practical value.

Description

Technical Field

[0001] The invention relates to an obstacle identification and laser obstacle removal method for a power transmission line. Background Art

[0002] The safe operation of transmission lines is crucial to the stability of the power system. Any obstruction could cause line failures or even widespread power outages, threatening power supply reliability and public safety. Therefore, developing efficient obstacle identification and clearance technologies has become a key issue in power maintenance. Currently, traditional obstacle removal methods rely primarily on manual inspections or simple mechanical clearance, which are inefficient and pose safety risks. While drone inspections have made some progress, their limited recognition accuracy makes them incapable of handling the diverse obstacles found in complex environments. Furthermore, their single obstacle removal method is difficult to adapt to the characteristics of different obstacle types. These limitations result in poor obstacle removal results, making it difficult to meet the demands of modern power grids for efficient and safe maintenance.

[0003] In the handling of obstacles on power transmission lines, the core challenge lies in how to accurately distinguish the material and shape of the obstacles and optimize the obstacle clearance strategy accordingly. The diversity of obstacle materials, such as metal, branches or bird nests, requires recognition technology to capture their unique spectral characteristics for accurate classification. However, a single recognition technology is difficult to accurately distinguish material differences, which directly affects the adaptability of subsequent obstacle clearance parameters; further, the position and shape of obstacles on the line, such as crossing, hanging or entangled, increase the complexity of obstacle clearance. Different position shapes require adjustment of the angle and trajectory of the obstacle clearance tool, and the existing technology lacks the ability to dynamically analyze the position shape, resulting in low obstacle clearance efficiency and even possible damage to the line. These challenges are progressive. The lack of material recognition limits the optimization of obstacle clearance parameters, and the complexity of the position shape puts higher requirements on parameter adjustment. Summary of the Invention

[0004] The present invention proposes a method for obstacle identification and laser obstacle removal for power transmission lines. It uses multispectral imaging technology to accurately identify the material characteristics of obstacles, and dynamically optimizes laser obstacle removal parameters based on their position and morphology, thereby improving the efficiency and safety of transmission line obstacle removal.

[0005] The technical solution of the present invention is achieved as follows: A method for obstacle identification and laser obstacle removal for a power transmission line, characterized in that the method comprises: Spectral data of transmission line obstacles is collected using multispectral imaging equipment to obtain image data covering the visible light band, near-infrared band, and ultraviolet band, generating a first spectral feature dataset of the multispectral image. A convolutional neural network model is used to extract the first spectral feature vector and output the obstacle material classification result, determining whether the obstacle is metal, tree branch, or bird's nest. If the material classification result is metal, the preset metal material spectrum template library is used to calculate the spectral signal-to-noise ratio of the first spectral feature vector and the template library, obtain the spectral similarity, and determine the metal material confirmation result; if the material classification result is a branch or a bird's nest, the texture features of the image are extracted to generate a first texture feature set to determine the non-metal material confirmation result; based on the metal / non-metal material confirmation result, the preset obstacle clearance parameter database is queried to obtain the laser power range and pulse frequency range corresponding to the metal, branch, or bird's nest, and generate the initial obstacle clearance parameter configuration including laser power and pulse frequency; Using a depth camera mounted on a drone, depth image data of obstacles on the transmission line is collected. The first depth data with a predetermined resolution and three-dimensional coordinates is analyzed to extract the intersection angle, hanging height, and winding curvature to generate a first obstacle morphological description. Based on the first obstacle shape description, a dynamic programming algorithm is used to input the intersection shape angle, suspension shape height, and winding shape curvature. The motion trajectory of the laser obstacle removal tool after the intersection angle is optimized is calculated, and obstacle removal trajectory data including obstacle removal angle and curvature path planning is generated; Environmental sensors are used to collect wind speed and humidity data around the transmission lines. The effects of wind speed on laser beam deviation and humidity on laser power attenuation are analyzed. Laser power matching and pulse frequency adjustment in the obstacle removal parameters are adjusted and optimized to generate the final obstacle removal parameters that include environmental adaptation corrections. The final obstacle removal parameters are used to control the laser obstacle removal equipment, execute the obstacle removal operation, collect spectral feedback data in real time during the obstacle removal process, analyze the dimensional changes of the second spectral feature vector at a predetermined frequency of data collection, and judge the obstacle removal completion status; if the obstacle removal completion status does not reach the preset threshold, the first obstacle morphology description is updated according to the dimensional changes of the second spectral feature vector, and the dynamic programming algorithm is re-input to generate new obstacle removal trajectory data including real-time trajectory updates, and determine the new obstacle removal execution parameters.

[0006] The beneficial effects of the present invention include: collecting obstacle spectral data through multispectral imaging, using convolutional neural networks for material classification, combining deep image analysis to analyze obstacle morphology, using a dynamic programming algorithm to plan obstacle removal trajectories, and dynamically adjusting laser obstacle removal parameters based on environmental factors. This allows for accurate identification and classification of different types of obstacles, such as metal, tree branches, and bird nests. It can adaptively generate optimal obstacle removal trajectories and parameters, and monitor and adjust them in real time during the clearance process, effectively improving the intelligence and accuracy of transmission line obstacle removal and ensuring the safe and stable operation of the power grid. The technical solution of the present invention integrates advanced technologies such as multispectral analysis, deep learning, and three-dimensional imaging, achieving full-process intelligence for transmission line obstacle removal and possessing significant practical value. DETAILED DESCRIPTION

[0007] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0008] A method for identifying obstacles and clearing obstacles with laser for power transmission lines, comprising: S101. Collect spectral data of transmission line obstacles using a multispectral imaging device to obtain image data covering a visible light band range of 380-780 nanometers, a near-infrared band range of 780-2500 nanometers, and an ultraviolet band range of 200-400 nanometers, and generate a first spectral feature data set with a multispectral image resolution of 1024x1024 pixels and a spectral signal-to-noise ratio greater than 30 decibels.

[0009] Multispectral imaging equipment is used to collect spectral data of transmission line obstacles, covering visible, near-infrared, and ultraviolet bands, to generate a high-resolution multispectral image dataset. If the signal-to-noise ratio of the collected spectral data is below a preset threshold, the image is denoised using a median filter algorithm to obtain a denoised multispectral image dataset. Based on the denoised multispectral image dataset, a principal component analysis algorithm is used to extract the main spectral features of each band, generating a feature-enhanced multispectral feature dataset. Within this feature-enhanced multispectral feature dataset, if the degree of match between the obstacle spectral features and a preset obstacle spectral template exceeds a preset threshold, a spectral classification algorithm is used to determine the obstacle type, generating a classified obstacle dataset. Spatial distribution information of the obstacles is obtained from the classified obstacle dataset, and an image segmentation algorithm is used to separate the obstacle regions, generating a boundary dataset of the obstacle regions. Based on the boundary dataset of the obstacle regions, the distance between the obstacle and the transmission line is calculated. If the distance is below a preset threshold, a geometric transformation algorithm is used to determine the spatial position of the obstacle, generating an obstacle location dataset. Using this obstacle location dataset, a spatial analysis algorithm is used to calculate the impact range of the obstacle on the transmission line, generating a distribution dataset of the obstacle impact area.

[0010] S102: For the first spectral feature data set, a convolutional neural network model is used to extract a first spectral feature vector with a dimension of 256. The first spectral feature data set is input, and an obstacle material classification result is output to determine whether the obstacle is metal, branch, or bird's nest.

[0011] A 256-dimensional first spectral feature vector is extracted from the first spectral feature dataset using a preset convolutional neural network model to obtain a first feature vector set. A fully connected layer is used to map the first feature vector set to generate a classification probability distribution, thereby obtaining a first classification probability set.

[0012]

[0013] represents the classification probability of the i-th category, x represents the input feature vector, W represents the weight matrix, b represents the bias vector, K represents the total number of categories, represents the i-th row of the weight matrix, Represents the i-th component of the bias vector. If the maximum probability value in the first classification probability set is greater than the preset threshold, then the obstacle material is judged to be metal, branch, or bird's nest based on the category corresponding to the maximum probability, and the first material classification result is obtained. By comparing the first material classification result with the preset material spectral feature library, the material spectrum matching degree is obtained to obtain the first matching degree set. If the highest matching degree in the first matching degree set is greater than the preset matching threshold, the first material classification result is confirmed to be the final material category, and the final material category is obtained. Based on the final material category, the preset material attribute table is used to obtain the physical property parameters of the corresponding material to obtain the first physical property set. Using the first physical property set, a classification label for the obstacle material is generated to obtain the final classification label.

[0014] S103. If the material classification result is metal, a preset metal material spectrum template library is used to calculate the spectral signal-to-noise ratio of the first spectral feature vector and the template library, obtain spectral similarity, and determine the metal material confirmation result. If the material classification result is a tree branch or a bird's nest, texture features with a grayscale depth of 16 bits are extracted to generate a first texture feature set to determine the non-metal material confirmation result.

[0015] If the input data is determined to be metal by the material classification model, a metal spectral template is retrieved from a preset spectral template library. The spectral signal-to-noise ratio (SNR) between the first spectral feature vector and each template in the template library is calculated to obtain spectral similarity, and the metal material confirmation result is determined. Based on the spectral similarity, a preset threshold is applied to determine whether the material is metal. If the spectral similarity exceeds the threshold, a metal confirmation label is generated; if it is below the threshold, the material is marked as pending, and image data of the pending material is obtained. A 16-bit grayscale image is extracted from the image data of the pending material, and texture features are calculated using a gray-level co-occurrence matrix algorithm to generate a first texture feature set. Based on the first texture feature set, a support vector machine algorithm is used to classify the texture features and determine whether the material is a tree branch or a bird's nest, resulting in a non-metallic material classification result. From the non-metallic material classification result, statistical parameters of the texture feature set are obtained, and the mean and variance of the texture features are calculated to generate a texture feature description vector. Based on the texture feature description vector, a match is performed using a preset texture template library. The Euclidean distance between the texture feature description vector and the template library is calculated to obtain material texture similarity and determine the non-metallic material confirmation result. Extract the final material label from the non-metallic material confirmation results and combine it with the metal material confirmation results to generate a complete material classification report.

[0016] S104. Based on the metal material confirmation result or the first texture feature set, query a preset obstacle clearance parameter database to obtain a laser power range and a pulse frequency range corresponding to the metal, tree branch, or bird's nest, and generate an initial obstacle clearance parameter configuration including the laser power and pulse frequency.

[0017] The material identification result or the first texture feature set is extracted from the input data. The material category is determined using a preset feature extraction algorithm to obtain a material identification result. Based on the material identification result, a preset obstacle clearance parameter database is queried and an exact matching method is used to obtain the laser power and pulse frequency ranges corresponding to metal, branches, or bird's nests. This determines the initial parameter set. If there are multiple matches between the laser power and pulse frequency ranges in the initial parameter set, a weighted average algorithm is used to calculate the optimal parameter values, resulting in an optimized parameter set. A parameter verification model is used to perform constraint checks on the laser power and pulse frequency within the optimized parameter set to determine whether the parameters meet the operating range of the obstacle clearance device. This results in a validation parameter set. Based on the validation parameter set, an initial obstacle clearance parameter configuration containing the laser power and pulse frequency is generated. This is then converted into a configuration format recognized by the device using a data formatting tool to determine the final configuration. The laser power and pulse frequency are extracted from the final configuration. A real-time monitoring algorithm is used to monitor the operating status of the obstacle clearance device to determine whether dynamic parameter adjustments are required and generate adjustment recommendations. Based on the adjustment recommendations, a parameter update algorithm is used to fine-tune the laser power and pulse frequency in the final configuration to generate a dynamic obstacle clearance parameter configuration.

[0018] S105. Use the depth camera carried by the drone to collect depth image data of the obstacle on the transmission line, analyze the first depth data with three-dimensional coordinates and a resolution of 1280x720 pixels, extract the intersection shape angle, suspension shape height and winding shape curvature, and generate a first obstacle shape description.

[0019] Based on the above business content and the extracted related attributes, the following business is generated: the depth image of the transmission line obstacle is collected by a depth camera mounted on a drone to generate the first depth data. If the resolution of the first depth data reaches the preset threshold, the point cloud generation algorithm is used to extract the three-dimensional coordinates of the obstacle to obtain the first three-dimensional coordinate set. Based on the first three-dimensional coordinate set, the geometric analysis algorithm is used to calculate the intersection angle of the obstacle to obtain the first angle data. By comparing the first three-dimensional coordinate set with the transmission line reference height, the hanging shape height of the obstacle is calculated to obtain the first height data. For the first three-dimensional coordinate set, the curve fitting algorithm is used to analyze the winding shape curvature of the obstacle to obtain the first curvature data. Based on the first angle data, the first height data and the first curvature data, a first obstacle shape description is generated.

[0020] P represents the obstacle morphology description function, Represents the weight of the morphological feature, R represents the angle response function, H represents the height response function, and k represents the number of features. The first obstacle morphological description is classified using a preset morphological classification model to determine the obstacle type and obtain a first classification result.

[0021] S106. Based on the first obstacle shape description, a dynamic programming algorithm is used to input the intersection shape angle, the hanging shape height, and the winding shape curvature, calculate the motion trajectory of the laser obstacle removal tool after the intersection angle is optimized, and generate obstacle removal trajectory data including the obstacle removal angle and curvature path planning.

[0022] A dynamic programming algorithm is used as input for the intersection angle, suspension height, and winding curvature to calculate the optimized motion trajectory of the laser obstacle removal tool, generating obstacle removal trajectory data. Based on the obstacle removal trajectory data, obstacle removal angle data and curvature path planning are extracted. A geometric analysis algorithm is used to calculate the angle adjustment of the obstacle removal tool at each trajectory point, generating an angle adjustment dataset. Path control instructions for the obstacle removal tool are generated based on the obstacle removal trajectory data and angle adjustment dataset. A path smoothing algorithm is used to optimize the instruction sequence, generating a smoothed path control instruction set. If the smoothed path control instruction set meets a preset trajectory continuity threshold, real-time motion parameters for the obstacle removal tool are generated based on the instruction set, generating a real-time motion parameter set. If the parameters in the real-time motion parameter set exceed a preset safety range, the tool's motion speed is adjusted using a feedback control algorithm, generating an adjusted motion speed dataset. Based on the adjusted motion speed dataset, the tool's final execution trajectory is generated. A trajectory verification algorithm is used to determine the trajectory's feasibility, generating the final obstacle removal execution trajectory. Based on the final obstacle removal execution trajectory, a control signal sequence for the obstacle removal task is generated, determining the execution sequence for the task.

[0023] S107. Collect wind speed data and humidity data around the transmission line through environmental sensors, analyze the effect of wind speed on laser beam deviation and the effect of humidity on laser power attenuation, adjust and optimize the laser power matching and pulse frequency adjustment in the obstacle removal parameters, and generate final obstacle removal parameters including environmental adaptation correction.

[0024] Environmental sensors collect wind speed and humidity data around the transmission lines and store them as time series data sets to obtain raw environmental data. A preprocessing algorithm is used to denoise and standardize the raw environmental data, generating standardized wind speed and humidity time series and determining the processed environmental data. Based on the standardized wind speed time series, the effect of wind speed on laser beam offset is calculated. A linear regression algorithm is used to fit the relationship between wind speed and offset, resulting in a laser beam offset model.

[0025]

[0026] ΔS represents the final laser beam offset, β represents the wind speed influence coefficient, represents the average wind speed, γ represents the height correction coefficient, H represents the laser transmission height, and ε represents the random error term. If the fitting error of the laser beam offset model is less than the preset threshold, the emission angle of the laser beam is adjusted according to the model output to generate the corrected emission parameters. The correlation characteristics of humidity on laser power attenuation are analyzed through standardized humidity time series, and the relationship between humidity and power attenuation is fitted by polynomial regression algorithm to obtain the power attenuation model.

[0027] ΔP represents the power attenuation, represents the polynomial coefficients, Denotes the normalized humidity value, and n represents the polynomial order. Based on the power attenuation model, the laser power compensation value is calculated. Combined with the corrected emission parameters, the laser power matching and pulse frequency adjustment in the obstacle clearance parameters are adjusted to generate optimized obstacle clearance parameters. If the optimized obstacle clearance parameters meet the preset performance threshold, they are integrated with the environmental adaptation correction data to generate the final obstacle clearance parameters.

[0028] S108. Use the final obstacle removal parameters to control the laser obstacle removal equipment, perform the obstacle removal operation, collect spectral feedback data in real time during the obstacle removal process, analyze the dimensional changes of the second spectral feature vector at a data collection frequency of 10 Hz, and determine the obstacle removal completion status.

[0029] The final obstacle removal parameters are loaded into the laser obstacle removal device, and the obstacle removal operation is performed to generate initial spectral feedback data. A second spectral feature is extracted from the initial spectral feedback data using a 10 Hz frequency to obtain a feature vector sequence. The feature vector sequence is processed using a principal component analysis algorithm to extract the feature vector dimensions and generate dimensional change data. The dimensional change trend is calculated based on the dimensional change data, and a sliding window method is used for smoothing to obtain a smoothed dimensional trend. If the smoothed dimensional trend is lower than the preset threshold, the degree of obstacle removal is determined through spectral data analysis to obtain the obstacle removal status parameters. Based on the comparison of the obstacle removal status parameters with the preset completion conditions, if the conditions are met, the obstacle removal completion status is determined. The final obstacle removal parameters are adjusted based on the obstacle removal completion status, and an updated parameter sequence is generated and loaded into the laser obstacle removal device.

[0030] S109: If the obstacle clearance completion status does not reach a preset threshold, the first obstacle morphology description is updated according to the change in the dimension of the second spectral feature vector, the dynamic programming algorithm is re-input, new obstacle clearance trajectory data including real-time trajectory updates is generated, and new obstacle clearance execution parameters are determined.

[0031] If the obstacle clearance completion status does not reach the preset threshold, the vector dimension change is extracted from the second spectral feature, the characteristic vector change rate is calculated, and the obstacle morphology change trend is obtained. According to the obstacle morphology change trend, the first obstacle morphology description is updated and a second obstacle morphology description is generated. The second obstacle morphology description is processed by a dynamic programming algorithm to generate updated obstacle clearance trajectory data. The trajectory deviation feature is extracted from the updated obstacle clearance trajectory data, the degree of match between the deviation and the preset trajectory is calculated, and the trajectory adjustment coefficient is determined. According to the trajectory adjustment coefficient, the obstacle clearance execution parameters are updated to generate a new obstacle clearance execution parameter set. If the new obstacle clearance execution parameter set does not match the preset parameter range, the deviation value is extracted from the obstacle clearance execution parameter set, and the parameters are optimized by a linear regression algorithm to obtain the adjusted execution parameters. The final obstacle clearance control instruction is generated based on the adjusted execution parameters and output to the obstacle clearance device.

[0032] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying obstacles and clearing obstacles with laser for power transmission lines, characterized in that: The method comprises: Spectral data of transmission line obstacles is collected using multispectral imaging equipment to obtain image data covering the visible light band, near-infrared band, and ultraviolet band, generating a first spectral feature dataset of the multispectral image. A convolutional neural network model is used to extract the first spectral feature vector and output the obstacle material classification result, determining whether the obstacle is metal, tree branch, or bird's nest. If the material classification result is metal, the preset metal material spectrum template library is used to calculate the spectral signal-to-noise ratio of the first spectral feature vector and the template library, obtain the spectral similarity, and determine the metal material confirmation result; if the material classification result is a branch or a bird's nest, the texture features of the image are extracted to generate a first texture feature set to determine the non-metal material confirmation result; based on the metal / non-metal material confirmation result, the preset obstacle clearance parameter database is queried to obtain the laser power range and pulse frequency range corresponding to the metal, branch, or bird's nest, and generate the initial obstacle clearance parameter configuration including laser power and pulse frequency; Using a depth camera mounted on a drone, depth image data of obstacles on the transmission line is collected. The first depth data with a predetermined resolution and three-dimensional coordinates is analyzed to extract the intersection angle, hanging height, and winding curvature to generate a first obstacle morphological description. Based on the first obstacle shape description, a dynamic programming algorithm is used to input the intersection shape angle, suspension shape height, and winding shape curvature. The motion trajectory of the laser obstacle removal tool after the intersection angle is optimized is calculated, and obstacle removal trajectory data including obstacle removal angle and curvature path planning is generated; Environmental sensors are used to collect wind speed and humidity data around the transmission lines. The effects of wind speed on laser beam deviation and humidity on laser power attenuation are analyzed. Laser power matching and pulse frequency adjustment in the obstacle removal parameters are adjusted and optimized to generate the final obstacle removal parameters that include environmental adaptation corrections. The final obstacle removal parameters are used to control the laser obstacle removal equipment, execute the obstacle removal operation, collect spectral feedback data in real time during the obstacle removal process, analyze the dimensional changes of the second spectral feature vector at a predetermined frequency of data collection, and judge the obstacle removal completion status; if the obstacle removal completion status does not reach the preset threshold, the first obstacle morphology description is updated according to the dimensional changes of the second spectral feature vector, and the dynamic programming algorithm is re-input to generate new obstacle removal trajectory data including real-time trajectory updates, and determine the new obstacle removal execution parameters.

2. The method for obstacle identification and laser obstacle removal for power transmission lines according to claim 1, characterized in that: The method of extracting the first spectral feature vector using a convolutional neural network model and outputting the obstacle material classification result to determine whether the obstacle is metal, branch, or bird's nest includes: By presetting the convolutional neural network model, a 256-dimensional first spectral feature vector is extracted from the first spectral feature data set to obtain a first feature vector set; a fully connected layer is used to map the first feature vector set to generate a classification probability distribution to obtain a first classification probability set. represents the classification probability of the i-th category, x represents the input feature vector, W represents the weight matrix, b represents the bias vector, K represents the total number of categories, represents the i-th row of the weight matrix, represents the i-th component of the bias vector; If the maximum probability value in the first classification probability set is greater than a preset threshold, the obstacle material is judged to be metal, branch, or bird's nest based on the category corresponding to the maximum probability, and a first material classification result is obtained; by comparing the first material classification result with a preset material spectral feature library, the material spectrum matching degree is obtained to obtain a first matching degree set; if the highest matching degree in the first matching degree set is greater than a preset matching threshold, the first material classification result is confirmed as the final material category, and the final material category is obtained; According to the final material category, a preset material attribute table is used to obtain the physical property parameters of the corresponding material, and a classification label of the obstacle material is generated according to the obtained first physical property set to obtain a final classification label.

3. The method for obstacle identification and laser obstacle removal for power transmission lines according to claim 2, characterized in that: If the material classification result is metal, a preset metal material spectrum template library is used to calculate the spectrum signal-to-noise ratio of the first spectrum feature vector and the template library, obtain the spectrum similarity, and determine the metal material confirmation result; If the material classification result is a branch or a bird's nest, the texture features of the image are extracted to generate a first texture feature set to determine the non-metallic material confirmation result, including: If the material classification result is metal, a preset metal material spectrum template library is used to calculate the spectrum signal-to-noise ratio of the first spectrum feature vector and the template library, obtain spectrum similarity, and determine the metal material confirmation result; if the material classification result is a branch or a bird's nest, the texture features of the image are extracted to generate a first texture feature set to determine the non-metal material confirmation result, including: If the input data is judged to be metal by the material classification model, a metal spectral template is obtained from a preset spectral template library. The spectral signal-to-noise ratio of the first spectral feature vector and each template in the template library is calculated to obtain the spectral similarity and determine the metal material confirmation result. Based on the spectral similarity, a preset threshold is used to determine whether the material is metal. If the spectral similarity is higher than the threshold, a metal material confirmation label is generated. If it is lower than the threshold, it is marked as a pending material, and image data of the pending material is obtained. An image with a grayscale depth of 16 bits is extracted from the image data of the pending material. The texture features are calculated using a grayscale co-occurrence matrix algorithm to generate a first texture feature set. Based on the first texture feature set, the texture features are classified using a support vector machine algorithm to determine whether the material is a tree branch or a bird's nest, and a non-metallic material classification result is obtained. The statistical parameters of the texture feature set are obtained from the non-metallic material classification result, the mean and variance of the texture features are calculated, and a texture feature description vector is generated. A preset texture template library is used for matching, and the Euclidean distance between the texture feature description vector and the template library is calculated to obtain the material texture similarity, and the non-metallic material confirmation result is determined. Extract the final material label from the non-metallic material confirmation results and combine it with the metal material confirmation results to generate a complete material classification report.

4. The method for obstacle identification and laser obstacle removal for power transmission lines according to claim 1, characterized in that: The method collects depth image data of obstacles on the transmission line through a depth camera carried by the drone, analyzes first depth data with three-dimensional coordinates and a predetermined resolution, extracts the intersection shape angle, the hanging shape height, and the winding shape curvature, and generates a first obstacle shape description, including: A depth image of the power transmission line obstacle is collected by a depth camera mounted on a drone to generate first depth data. If the resolution of the first depth data reaches a preset threshold, a point cloud generation algorithm is used to extract the three-dimensional coordinates of the obstacle to obtain a first three-dimensional coordinate set. According to the first three-dimensional coordinate set, a geometric analysis algorithm is used to calculate the intersection angle of the obstacle to obtain first angle data; By comparing the first three-dimensional coordinate set with the transmission line reference height, the height of the obstacle's suspension form is calculated to obtain first height data; For the first three-dimensional coordinate set, a curve fitting algorithm is used to analyze the winding curvature of the obstacle to obtain first curvature data; Generate a first obstacle shape description based on the first angle data, the first height data, and the first curvature data: P represents the obstacle morphology description function, represents the weight of morphological features, R represents the angle response function, H represents the height response function, and k represents the number of features; The first obstacle morphology description is classified by using a preset morphology classification model to determine the obstacle type and obtain a first classification result.

5. The method for obstacle identification and laser obstacle removal for power transmission lines according to claim 1, characterized in that: The first obstacle shape description is described using a dynamic programming algorithm, which inputs the intersection shape angle, the suspension shape height, and the winding shape curvature, calculates the motion trajectory of the laser obstacle removal tool after the intersection angle is optimized, and generates obstacle removal trajectory data including the obstacle removal angle and curvature path planning, including: A dynamic programming algorithm is used to input the intersection angle, suspension height, and winding curvature to calculate the optimized motion trajectory of the laser obstacle removal tool. This generates obstacle removal trajectory data. The obstacle removal angle data and curvature path planning are then extracted. A geometric analysis algorithm is used to calculate the angle adjustment amount of the obstacle removal tool at each trajectory point to generate an angle adjustment data set. The obstacle removal tool's path control instructions are generated based on the obstacle removal trajectory data and angle adjustment dataset. A path smoothing algorithm is used to optimize the instruction sequence to obtain a smoothed path control instruction set. If the smoothed path control instruction set meets a preset trajectory continuity threshold, the obstacle removal tool's real-time motion parameters are generated based on the instruction set to obtain a real-time motion parameter set. If the parameters in the real-time motion parameter set exceed a preset safety range, the obstacle removal tool's motion speed is adjusted using a feedback control algorithm to obtain an adjusted motion speed dataset. Based on the adjusted motion speed dataset, the final execution trajectory of the obstacle removal tool is generated. The trajectory verification algorithm is used to judge the executability of the trajectory, obtain the final obstacle removal execution trajectory, generate the control signal sequence of the obstacle removal task, and determine the execution sequence of the obstacle removal task.

6. The method for obstacle identification and laser obstacle removal for power transmission lines according to claim 1, characterized in that: The wind speed data and humidity data around the transmission line are collected by environmental sensors, the influence of wind speed on laser beam deviation and the influence of humidity on laser power attenuation are analyzed, the laser power matching and pulse frequency adjustment in the obstacle removal parameters are adjusted and optimized, and the final obstacle removal parameters including environmental adaptation correction are generated, including: The wind speed and humidity data around the transmission line are collected by environmental sensors and stored as a time series data set to obtain the original environmental data. The original environmental data is denoised and standardized using a preprocessing algorithm to generate a standardized wind speed and humidity time series, and the processed environmental data is determined. According to the standardized wind speed time series, the disturbance law of wind speed on laser beam offset is calculated, and the relationship between wind speed and offset is fitted by linear regression algorithm to obtain the laser beam offset model. ΔS represents the final laser beam offset, β represents the wind speed influence coefficient, represents the average wind speed, γ represents the height correction coefficient, H represents the laser transmission height, and ε represents the random error term. If the fitting error of the laser beam offset model is less than the preset threshold, the emission angle of the laser beam is adjusted according to the model output to generate the corrected emission parameters. The correlation characteristics of humidity on laser power attenuation are analyzed through standardized humidity time series, and the relationship between humidity and power attenuation is fitted by polynomial regression algorithm to obtain the power attenuation model. ΔP represents the power attenuation, represents the polynomial coefficients, represents the standardized humidity value, and n represents the polynomial order; According to the power attenuation model, the laser power compensation value is calculated. Combined with the corrected emission parameters, the laser power matching and pulse frequency adjustment in the obstacle clearance parameters are adjusted to generate optimized obstacle clearance parameters. If the optimized obstacle clearance parameters meet the preset performance threshold, the parameters are integrated with the environmental adaptation correction data to generate the final obstacle clearance parameters.

7. The method for obstacle identification and laser obstacle removal for power transmission lines according to claim 1, characterized in that: The method uses the final obstacle removal parameters to control the laser obstacle removal equipment, performs the obstacle removal operation, collects spectral feedback data in real time during the obstacle removal process, analyzes the dimensional changes of the second spectral feature vector at a predetermined frequency of the data collection, and determines the obstacle removal completion status, including: The final obstacle removal parameters are loaded into the laser obstacle removal device, and the obstacle removal operation is performed to generate initial spectral feedback data. The second spectral feature is extracted from the initial spectral feedback data using a 10 Hz frequency to obtain a feature vector sequence. The feature vector sequence is processed using the principal component analysis algorithm to extract the feature vector dimension and generate dimension change data. The dimensional change trend is calculated based on the dimensional change data, and the sliding window method is used for smoothing to obtain the smoothed dimensional trend. If the smoothed dimensional trend is lower than the preset threshold, the degree of obstacle removal is determined through spectral data analysis to obtain the obstacle removal status parameter. Based on the comparison between the obstacle clearance status parameters and the preset completion conditions, if the conditions are met, the obstacle clearance completion status is determined, the final obstacle clearance parameters are adjusted according to the obstacle clearance completion status, and an updated parameter sequence is generated and loaded into the laser obstacle clearance device.

8. The method for obstacle identification and laser obstacle removal for power transmission lines according to claim 7, characterized in that: If the obstacle clearance completion status does not reach the preset threshold, the first obstacle morphology description is updated according to the change in the dimension of the second spectral feature vector, the dynamic programming algorithm is re-input, new obstacle clearance trajectory data including real-time trajectory updates is generated, and new obstacle clearance execution parameters are determined, including: If the obstacle clearance completion status does not reach the preset threshold, the vector dimension change is extracted from the second spectral feature, the feature vector change rate is calculated, the obstacle morphology change trend is obtained, the first obstacle morphology description is updated, and the second obstacle morphology description is generated; The second obstacle morphology description is processed through a dynamic programming algorithm to generate updated obstacle clearance trajectory data. The trajectory deviation characteristics are extracted from the updated obstacle clearance trajectory data, the matching degree between the deviation and the preset trajectory is calculated, and the trajectory adjustment coefficient is determined. According to the trajectory adjustment coefficient, the obstacle removal execution parameters are updated to generate a new obstacle removal execution parameter set; if the new obstacle removal execution parameter set does not match the preset parameter range, the deviation value is extracted from the obstacle removal execution parameter set, and the parameters are optimized through the linear regression algorithm to obtain the adjusted execution parameters; The final obstacle clearance control instruction is generated through the adjusted execution parameters and output to the obstacle clearance equipment.

Images