Electric power corridor vegetation management method and system based on spectral image and laser point cloud

Abstract

The invention relates to an electric power corridor vegetation management method and system based on a spectral image and a laser point cloud, and the method comprises the steps: collecting laser point cloud data and multi-spectral image data in an electric power corridor through an unmanned plane, and carrying out the fusion through a dynamic nonlinear spectral band selection mechanism, and generating a fusion point cloud containing a vegetation space and spectral features; vegetation features are extracted from the fusion point cloud through a multi-scale image convolutional network, a time sequence is constructed, a growth model is trained by adopting a bidirectional LSTM and a self-attention mechanism, and the growth trend of vegetation is predicted; and comprehensively evaluating the contact risk of the vegetation and the power line by adopting fuzzy logic based on the spatial information and the growth trend of the vegetation, and generating a visual model and an operation and maintenance instruction. According to the visual model and the operation and maintenance instruction generated by the method, the automation and intelligence level of vegetation management of the electric power gallery is remarkably enhanced, the contact risk of vegetation and a power line is effectively reduced, and safe operation of an electric power system is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of spectral imaging and laser point cloud, and particularly relates to a power corridor vegetation management method and system based on spectral imaging and laser point cloud. Background Art

[0002] With the continuous expansion of the power grid, power corridor vegetation management has become an important link to ensure the safe operation of the power system. The overgrowth of vegetation may come into contact with power lines, leading to risks such as short circuits, power outages, and even fires. Therefore, it is necessary to monitor and evaluate the spatial distribution and growth trend of vegetation. In recent years, unmanned aerial vehicle technology combined with lidar and multispectral imaging technology has been widely used in vegetation management. Laser point cloud data can provide three-dimensional spatial information of vegetation, while multispectral image data can reflect the spectral characteristics and health status of vegetation, providing a basis for vegetation classification and growth analysis. However, there are still deficiencies in data fusion, feature extraction, and risk assessment in the existing technology.

[0003] In terms of data fusion, traditional vegetation monitoring methods mostly use simple superposition of point cloud and spectral data, without fully considering the dynamic selectivity of spectral bands, resulting in limited accuracy in characterizing vegetation features in the fusion result. In terms of feature extraction and growth prediction, conventional methods usually rely on single-scale spatial analysis or statistical models based on static data, and it is difficult to capture the dynamic change characteristics of vegetation in multi-scale space and time dimensions. In addition, existing technologies mostly use deterministic threshold judgment in risk assessment, ignoring the fuzziness and complex interaction relationships among multiple variables such as vegetation height, distance, and growth rate, and the evaluation results often lack comprehensiveness and robustness. These problems limit the automation and intelligence level of power corridor vegetation management and are difficult to meet the needs of efficient operation and maintenance of large-scale power grids.

[0004] In recent years, artificial intelligence technologies such as graph convolutional network, long short-term memory network (LSTM), and fuzzy logic have shown potential in related fields. For example, graph convolutional network can be used to process the spatial relationship of point cloud data, LSTM is suitable for time series modeling, and fuzzy logic can handle uncertainty problems. However, how to combine these technologies with multispectral image and laser point cloud data to construct a complete vegetation management solution is still a difficult point in current research. In view of the above problems, it is necessary to develop a method that can achieve dynamic fusion of multi-source data, multi-scale feature extraction, time series growth prediction, and comprehensive risk assessment based on fuzzy logic, so as to improve the accuracy and practicality of power corridor vegetation management and provide technical support for the safe operation of the power system. Summary of the Invention

[0005] The purpose of the present invention is to provide a power corridor vegetation management method and system based on spectral images and laser point clouds, so as to solve the problem of low automation and intelligence in the power corridor vegetation management in the prior art.

[0006] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention provides a power corridor vegetation management method based on spectral images and laser point clouds, which is characterized in that it includes:

[0007] Collecting laser point cloud data and multi-spectral image data in the power corridor by using an unmanned aerial vehicle, and fusing them through a dynamic non-linear spectral band selection mechanism to generate a fused point cloud containing vegetation spatial and spectral characteristics;

[0008] Extracting vegetation features from the fused point cloud through a multi-scale graph convolutional network, constructing a time series and training a growth model using a bidirectional LSTM and a self-attention mechanism to predict the growth trend of the vegetation;

[0009] Based on the spatial information and growth trend of the vegetation, comprehensively evaluating the contact risk between the vegetation and the power line by using fuzzy logic, and generating a visualization model and operation and maintenance instructions.

[0010] As a further improvement of an embodiment of the present invention, the method further includes that the generating a fused point cloud containing vegetation spatial and spectral characteristics through a dynamic non-linear spectral band selection mechanism includes:

[0011] Defining the laser point cloud data in the power corridor as , where is the three-dimensional coordinate of the i-th point, and n is the number of points;

[0012] Defining the multi-spectral image data as , where is the reflection value of point i in band j, and k is the number of bands;

[0013] Inferring the vegetation type c and growth stage s from the multi-spectral image data S through a pre-trained classification model;

[0014] Calculating the weight of each spectral band j, and the formula is:

[0015] .

[0016] Where is the standard deviation of band j, is a scalar representation of vegetation characteristics and time dynamics, τ is the data acquisition timestamp, is band specificity, and are pre-trained embedding vectors, is a balance parameter, is a scaling factor;

[0017] Calculate the spectral features of each point, and the formula is:

[0018] .

[0019] Among them, is the weight of band j, is the mean value of band j, is the non - linear transformation function;

[0020] Generate the fused point cloud, which is expressed as:

[0021] .

[0022] As a further improvement of an embodiment of the present invention, the method further includes that extracting vegetation features from the fused point cloud by the multi - scale graph convolutional network includes,

[0023] Construct a graph structure from the fused point cloud , where the vertex set represents all points in the fused point cloud, and E is the edge set;

[0024] Calculate the edge weight used to characterize the spatial relationship between points i and l;

[0025] Among them, is a preset distance attenuation parameter used to control the neighborhood range;

[0026] Define multiple scales , where represents the neighborhood range of the p - th scale, and the neighborhood of point i at scale is ;

[0027] Extract local features at each scale , and the formula is:

[0028] .

[0029] Among them, is neighborhood aggregation, and are the learnable weight matrix and bias at the p - th scale;

[0030] Aggregate global features through max - pooling, which is expressed as:

[0031] .

[0032] Output the final feature representation of point i as:

[0033] .

[0034] Among them, represents the result of mapping the original features to a high-dimensional space through a multi-layer perceptron.

[0035] As a further improvement of an embodiment of the present invention, the method further includes that the constructing the time series and training the growth model by using bidirectional LSTM and self-attention mechanism includes,

[0036] Collecting vegetation features at multiple time points to construct a time series , where is the feature at the t-th moment;

[0037] Processing with bidirectional LSTM , calculating the forward hidden state and the backward hidden state ; Combining the forward and backward hidden states to obtain the bidirectional hidden state ;

[0038] By introducing the self-attention mechanism and based on the hidden state matrix , calculating the query matrix , the key matrix , and the value matrix ;

[0039] Among them, are the weights of the corresponding matrices respectively;

[0040] Calculating the attention weight matrix and the context representation matrix ;

[0041] Among them, is the dimension of the key vector for scaling.

[0042] As a further improvement of an embodiment of the present invention, the method further includes that the predicting the growth trend of the vegetation includes,

[0043] Predicting the vegetation height at the next time point through the formula ;

[0044] Among them, is the t-th column of the context representation matrix C, corresponding to the weighted context information at the t-th time step, aligned with the hidden state in the time dimension; are the predicted weight matrix and bias respectively;

[0045] Calculating the growth rate , where is the change in the predicted height at consecutive time points, is the time interval;

[0046] When training the model, historical height data is used as the supervision signal, and the model parameters are adjusted by optimizing the loss function, which is expressed as:

[0047] .

[0048] Wherein, is the true vegetation height at time step t, is the predicted vegetation height at time step t, is the total number of time steps of the time series.

[0049] As a further improvement of an embodiment of the present invention, the method further includes that the comprehensive evaluation of the contact risk between the vegetation and the power line by using fuzzy logic includes,

[0050] defining input variables, including the height difference , the closest distance d between the vegetation canopy points and the power line in the horizontal plane, and the growth rate v;

[0051] Wherein, is the power line height, and d is extracted from the laser point cloud P;

[0052] Establish the fuzzy set membership relationship of the input variables through the fuzzy membership function, which is used to quantify the fuzzy characteristics of the height difference , the horizontal distance d, and the growth rate v. Specifically, it includes,

[0053] For each input variable, define the membership function , which maps the input variable to the fuzzy set, including the fuzzy semantic levels of safety level, transition level, and danger level; the form of the membership function is determined based on the physical meaning of the input variable and the actual requirements of power corridor management;

[0054] Wherein, X represents , d or v;

[0055] Based on the fuzzy set membership relationship, configure the fuzzy inference rules between the input variables and the risk levels; the fuzzy inference rules express the influence of the input variables , d or v on the contact risk;

[0056] Combining the fuzzy set membership relationship of the input variables with the fuzzy inference rules, determine the activation degree of the fuzzy inference rules, and generate the corresponding fuzzy output subset;

[0057] The activation degree of each rule is expressed as:

[0058] .

[0059] Among them, is the activation degree of the q-th rule; is the membership degree at the input value , indicating the degree of belonging to a certain fuzzy level; is the membership degree of d at the input value , indicating the degree of d belonging to a certain fuzzy level; is the membership degree of v at the input value , indicating the degree of v belonging to a certain fuzzy level;

[0060] The corresponding fuzzy output subset is expressed as:

[0061] .

[0062] Among them, is the clipped output subset of the q-th rule, is the output membership function of the q-th rule, is the risk value range variable;

[0063] Aggregate the output subsets of all rules and generate a comprehensive membership function through the maximum operation, which is expressed as:

[0064] .

[0065] Among them, N is the total number of rules;

[0066] Defuzzify the comprehensive membership function and generate a scalar risk value using the weighted average method, which is expressed as:

[0067] .

[0068] Among them, is the risk value;

[0069] Classify the contact risk according to the risk value r and the predetermined standard to determine the potential contact threat degree between the vegetation and the power line.

[0070] As a further improvement of an embodiment of the present invention, the method further includes that the generation of the visualization model and the operation and maintenance instructions includes,

[0071] Generate a visual three-dimensional model based on the vegetation growth trend and the contact risk, and mark the risk area in the form of a heat map in the visual three-dimensional model to intuitively display the analysis results;

[0072] Automatically generate operation and maintenance instructions according to the risk level; when the risk level is high risk, trigger the instruction to trim the vegetation and specify the coordinate range to be trimmed;

[0073] Transfer the visualization model and operation and maintenance instructions to the operation and maintenance terminal to ensure that the analysis results are applied to the management of the power corridor in real time.

[0074] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention further provides a power corridor vegetation management system based on spectral images and laser point clouds, which is characterized in that it includes a data fusion module, a growth prediction module, and a risk assessment module;

[0075] The data fusion module is used to collect laser point cloud data and multi-spectral image data in the power corridor by using an unmanned aerial vehicle, and fuse and generate a fused point cloud containing vegetation spatial and spectral features through a dynamic non-linear spectral band selection mechanism;

[0076] The growth prediction module is used to extract vegetation features from the fused point cloud through a multi-scale graph convolutional network, construct a time series, and train a growth model by using a bidirectional LSTM and a self-attention mechanism to predict the growth trend of vegetation;

[0077] The risk assessment module is used to comprehensively evaluate the contact risk between vegetation and power lines based on the spatial information and growth trend of vegetation by using fuzzy logic, and generate a visualization model and operation and maintenance instructions.

[0078] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention further provides an electronic device, including a memory and a processor, characterized in that a computer program that can run on the processor is stored in the memory, and when the program is executed on the processor, the steps in the above-mentioned power corridor vegetation management method based on spectral images and laser point clouds are implemented.

[0079] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention further provides a storage medium, the storage medium stores a computer program, characterized in that when the computer program is executed by a processor, the steps in the above-mentioned power corridor vegetation management method based on spectral images and laser point clouds are implemented.

[0080] Compared with the prior art, a power corridor vegetation management method and system based on spectral images and laser point clouds provided by the present invention fuse multi-spectral image and laser point cloud data through a dynamic non-linear spectral band selection mechanism, improving the accuracy of vegetation feature extraction; using a multi-scale graph convolutional network and a bidirectional LSTM combined with a self-attention mechanism to achieve accurate prediction of the vegetation growth trend; a comprehensive risk assessment method based on fuzzy logic fully considers the uncertainty of multi-variable interaction, improving the robustness of contact risk assessment. The finally generated visualization model and operation and maintenance instructions significantly enhance the automation and intelligence level of power corridor vegetation management, effectively reduce the contact risk between vegetation and power lines, and ensure the safe operation of the power system. Description of the Drawings

[0081] Figure 1 It is the overall flowchart of the power corridor vegetation management method based on spectral images and laser point clouds according to the present invention.

[0082] Figure 2 It is the schematic architecture diagram of the power corridor vegetation management system based on spectral images and laser point clouds according to the present invention. Specific Embodiments

[0083] The present invention will be described in detail below in conjunction with the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, method, or functional transformation made by those of ordinary skill in the art based on these embodiments is included within the protection scope of the present invention.

[0084] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0085] In the first embodiment of the present invention, the present invention provides a power corridor vegetation management method based on spectral images and laser point clouds. As Figure 1 shown, the method includes

[0086] S1: Collect laser point cloud data and multispectral image data in the power corridor by an unmanned aerial vehicle, and fuse and generate a fused point cloud containing vegetation spatial and spectral features through a dynamic non-linear spectral band selection mechanism;

[0087] S2: Extract vegetation features from the fused point cloud through a multi-scale graph convolutional network, construct a time series, and train a growth model using bidirectional LSTM and self-attention mechanism to predict the growth trend of vegetation;

[0088] S3: Based on the spatial information and growth trend of vegetation, comprehensively evaluate the contact risk between vegetation and power lines using fuzzy logic, and generate a visualization model and operation and maintenance instructions.

[0089] In a specific embodiment of the present invention, the fused point cloud containing vegetation spatial and spectral features is generated by fusing through a dynamic non-linear spectral band selection mechanism, specifically as

[0090] Define the laser point cloud data in the power corridor as , where is the three-dimensional coordinate of the i-th point, and n is the number of points;

[0091] Define the multispectral image data as , where is the reflection value of point i in band j, and k is the number of bands;

[0092] Infer the vegetation type c and growth stage s from the multispectral image data S through a pre-trained classification model;

[0093] Calculate the weight of each spectral band j, and the formula is:

[0094] .

[0095] Where, is the standard deviation of band j, is the scalar representation of vegetation characteristics and temporal dynamics, τ is the data acquisition timestamp, is the band specificity, and are the pre-trained embedding vectors, is the balance parameter, is the scaling factor;

[0096] Calculate the spectral features of each point, and the formula is:

[0097] .

[0098] Where, is the weight of band j, is the mean value of band j, is the non-linear transformation function;

[0099] Generate a fused point cloud representation as:

[0100] .

[0101] It should be noted that the present invention generates a fused point cloud containing vegetation spatial information and spectral features by a dynamic non-linear band selection mechanism through the formula , fusing the laser point cloud data P collected by the drone and the multispectral image data S, , for subsequent feature extraction, growth prediction and risk assessment. The formula generates adaptive weights by integrating vegetation type, growth stage and temporal dynamics. Compared with traditional fixed band selection, it can optimize spectral information extraction according to environmental changes, improve the fusion accuracy, and is especially suitable for mixed vegetation scenarios. The formula generates spectral features through weighted sum and non-linear transformation, enhances the spectral expression of vegetation, and provides high-quality input for multi-scale feature extraction. The fused point cloud combines spatial coordinates and spectral features as the input of the subsequent graph convolutional network, connecting data fusion and feature extraction.

[0102] Furthermore, the system collects laser point cloud data P to describe the three-dimensional spatial structure of vegetation and its surrounding environment, and the number of points is determined by the acquisition range and device resolution. At the same time, it collects multi-spectral image data S to obtain the spectral reflection characteristics of vegetation, covering bands such as visible light and near-infrared, and the number of points is consistent with that of the laser point cloud to ensure spatial-spectral correspondence. The system infers the vegetation type c and growth stage s through a pre-trained classification model, providing input for weight calculation.

[0103] Furthermore, the system calculates the band weights , comprehensive standard deviation , vegetation characteristics , band specificity , and optimizes through balance parameters and scaling factors. The standard deviation , where reflects the information content of the band. Vegetation characteristics adopts a lightweight MLP (input layer dimension is the number of types + 2, hidden layer has 64 neurons, ReLU activation, output layer has 1 neuron, Sigmoid activation), and is trained through labeled data to capture growth period or seasonal changes. Band specificity extracts embedding vectors (dimension 32) through a spectral feature library and classification model, normalizes them to [0,1], and quantifies the correlation between the band and the vegetation type.

[0104] Furthermore, based on the weights , the system calculates spectral features , where enhances the dynamic range and suppresses noise. The fused point cloud combines spatial and spectral information to support subsequent multi-scale extraction.

[0105] In a specific embodiment of the present invention, vegetation features are extracted from the fused point cloud through a multi-scale graph convolutional network. Specifically,

[0106] construct a graph structure from the fused point cloud , where the vertex set represents all points in the fused point cloud, and E is the edge set;

[0107] calculate the edge weights for characterizing the spatial relationship between points i and l;

[0108] where is a preset distance attenuation parameter for controlling the neighborhood range;

[0109] define multiple scales , where represents the neighborhood range of the p-th scale, and the neighborhood of point i at scale is ;

[0110] Extract local features at each scale, and the formula is:

[0111] .

[0112] Wherein, is neighborhood aggregation, and are the learnable weight matrix and bias at the p-th scale;

[0113] Aggregate the global feature representation through max pooling as:

[0114] .

[0115] The final feature representation of output point i is:

[0116] .

[0117] Wherein, represents the result of mapping the original feature to a high-dimensional space through a multi-layer perceptron.

[0118] It should be noted that the present invention uses a multi-scale graph convolutional network to extract vegetation features from the fused point cloud, providing a high-quality representation for time series construction and growth prediction. Formula calculates the edge weight through an exponential decay function, dynamically quantifies the proximity based on the spatial distance, and is more adaptable to the uneven distribution of point cloud density than the fixed neighborhood method, suitable for the complex vegetation scenario of the power corridor. Formula Extracts local features through weighted aggregation at multiple scales, captures the spatial characteristics from branches and leaves to the canopy, and significantly improves the robustness and classification accuracy compared with the single-scale method. Formula g generates the global feature through max pooling, providing the overall context. Formula Integrates the global feature, the original feature mapping, and the multi-scale local features to form a comprehensive representation, supporting time series analysis.

[0119] Furthermore, the system converts the fused point cloud into a graph structure G, characterizing the spatial and feature relationships between points. The vertex set V contains all points, and each vertex carries three-dimensional coordinates and spectral features; the edge set E is constructed based on the spatial distance, adapting to the irregular distribution of the point cloud. The edge weight , calculated through the Euclidean distance and the distance decay parameter, highlights the association of neighboring points, and optimizes the feature extraction in the complex vegetation scenario.

[0120] Furthermore, the system defines multi-scale neighborhoods and extracts local features. Set an increasing sequence of neighborhood radii , and each point contains points with a distance less than at scale ​Neighboring points. Calculate local features through graph convolution , weighted aggregation of neighborhood features, independent weight matrix and bias Enhance scale specificity and capture local texture and overall geometry.

[0121] Furthermore, the system generates global feature g through max pooling, extracts the most significant features of the point cloud, and reflects the canopy height or spectral peak. The final feature Integrates global features, non-linear mapping of original features (MLP with multi-layer neural network), and multi-scale local features to form a comprehensive representation to support time series prediction.

[0122] In a specific embodiment of the present invention, a time series is constructed and a growth model is trained using bidirectional LSTM and self-attention mechanism. Specifically,

[0123] Vegetation features are collected at multiple time points to construct a time series , where is the feature at the t-th moment;

[0124] Processed using bidirectional LSTM , calculate the forward hidden state and the backward hidden state ; Combine the forward and backward hidden states to obtain the bidirectional hidden state ;

[0125] By introducing the self-attention mechanism and based on the hidden state matrix , calculate the query matrix , the key matrix , the value matrix ;

[0126] where are the weights of the corresponding matrices respectively;

[0127] Calculate the attention weight matrix and the context representation matrix ;

[0128] where is the dimension of the key vector for scaling.

[0129] It should be noted that the present invention models the time series of vegetation features through bidirectional LSTM and self-attention mechanism , capturing the growth dynamics. Formula Merge the forward and backward hidden states of the bidirectional LSTM to comprehensively capture the temporal context, and more accurately model the non-linear trend compared to the unidirectional LSTM. Formulas Q, K, V, A, and C calculate the time-step correlation through the self-attention mechanism, dynamically highlighting key time steps such as the rapid growth period, and enhancing the prediction stability. In implementation, Q, K, and V are linearly transformed from the hidden state matrix H, A is calculated using scaled dot-product attention, and C is weighted and summed to highlight important features.

[0130] Furthermore, the system constructs a time series based on multi-scale graph convolution features . Feature data, including canopy height, spectral characteristics, etc., are collected at fixed intervals (such as weekly), arranged in chronological order as a high-dimensional vector sequence, with a unified format to ensure consistent dimensions, covering multiple growth cycles. The length of the time series is determined by the number of collected points, supporting long-term trend modeling.

[0131] Furthermore, the bidirectional LSTM processes the time series to generate hidden states . The forward LSTM processes the features from the start of the sequence to capture the trend from the past to the present; the backward LSTM processes from the end to reflect future constraints. Each time step updates the memory unit through a gating mechanism to generate the forward and backward hidden states, which are concatenated into , retaining the complete context and supporting non-linear dynamic modeling.

[0132] Furthermore, the self-attention mechanism enhances the expression of key time steps. All constitute the matrix H, and Q, K, and V are generated through the weight matrix. The attention weight A and the context representation C highlight the features of the rapid growth period, supporting high-precision prediction.

[0133] Furthermore, the system optimizes the model parameters to ensure performance. The bidirectional LSTM sets a medium-dimensional hidden state, the forget gate bias is initialized to a positive value, and a single-layer or double-layer structure adapts to the sequence length. The attention mechanism adds dropout to prevent overfitting, and the scaling factor is based on the dimension of the key vector. Training uses batch processing and gradient clipping, and distributed computing accelerates the processing of long sequences.

[0134] In a specific embodiment of the present invention, to predict the growth trend of vegetation, specifically,

[0135] Through the formula predict the vegetation height at the next time point;

[0136] Where, is the t-th column of the context representation matrix C, corresponding to the weighted context information at the t-th time step, aligned with the hidden state in the time dimension; are the predicted weight matrix and bias respectively;

[0137] Calculate the growth rate , where is the change in predicted height at consecutive time points, is the time interval;

[0138] When training the model, historical height data is used as the supervision signal, and the model parameters are adjusted by optimizing the loss function, which is expressed as:

[0139] .

[0140] where is the true vegetation height at time step t, is the predicted vegetation height at time step t, is the total number of time steps in the time series.

[0141] It should be noted that the formula predicts the vegetation height by fusing the bidirectional hidden state and the context representation , combines the temporal dynamics and key time step information, and captures complex growth patterns more accurately than linear regression. The formula v calculates the growth rate based on the continuous height difference, quantifies the dynamic threat, and provides a basis for risk assessment. The formula L optimizes the model through the mean squared error to ensure the prediction accuracy and generalization ability. In implementation, is calculated through linear transformation, v reflects the growth trend, and L is optimized using gradient descent.

[0142] Furthermore, the system predicts the height and calculates the growth rate based on and . The input features integrate (forward and backward temporal information) and (weighted key time steps), and are linearly mapped to and the bias by the weight matrix , retaining the dynamic characteristics. The growth rate v is calculated by the difference between adjacent time points, and smoothing processing (such as weighted average) reduces noise to ensure stability. The time interval is based on the acquisition period, and the consistent unit facilitates risk comparison.

[0143] Furthermore, the system optimizes the model through supervised learning. Historical height data (from laser point cloud or field measurement) is used as the supervision signal, covering the growth cycle. The loss function L calculates the mean squared error of the prediction and the true height, and updates , and the aforementioned model parameters using the Adam optimizer, and regularization prevents overfitting. The training is divided into rounds, the performance is evaluated on the validation set, and the learning rate is adjusted to converge.

[0144] Furthermore, the system adopts optimization strategies to improve efficiency. Batch processing inputs sequences in batches and dynamically adjusts the batch size. Gradient clipping limits the norm to prevent numerical instability. Incremental learning gradually updates the model to adapt to new data. Feature selection prioritizes key dimensions such as canopy height to reduce complexity and accelerate prediction.

[0145] In a specific embodiment of the present invention, fuzzy logic is used to comprehensively evaluate the contact risk between vegetation and power lines. Specifically,

[0146] Define input variables, including the height difference , the closest distance d between the vegetation canopy points and the power line in the horizontal plane, and the growth rate v;

[0147] Among them, is the power line height, and d is extracted from the laser point cloud P;

[0148] It should be noted that by analyzing spatial and dynamic features, the basic data required for risk assessment is generated. These data describe the relative position and potential threat trend between vegetation and power lines, providing quantitative input for subsequent analysis.

[0149] Furthermore, the fuzzy set membership relationship of the input variables is established through fuzzy membership functions to quantify the fuzzy characteristics of the height difference , the horizontal distance d, and the growth rate v. Specifically,

[0150] For each input variable, a membership function is defined to map the input variable to fuzzy sets, including fuzzy semantic levels such as the safe level, the transitional level, and the dangerous level; the form of the membership function is determined based on the physical meaning of the input variable and the actual requirements of power corridor management;

[0151] Among them, X represents , d, or v;

[0152] Preferably, the membership function adopted in an embodiment of the present invention

[0153] .

[0154] Among them, is the center of the fuzzy set, is the width of the fuzzy set, corresponding to the three levels of safe, excessive, and dangerous respectively. Safe level: indicating low risk, the center is based on safety standards (such as the minimum height requirement of the power line). Transitional level: indicating medium risk, the center is located at the intermediate value between safe and dangerous. Dangerous level: indicating high risk, the center Approaching the threat threshold.

[0155] It should be noted that by converting precise data into fuzzy semantics, it is convenient to handle uncertainties. The data is mapped to different levels of risk descriptions, providing input for subsequent decision-making rules.

[0156] Furthermore, based on the membership relationship of the fuzzy set, configure the fuzzy inference rules between the input variables and the risk levels; the fuzzy inference rules express the input variables through logical relationships , d or v's influence on the contact risk;

[0157] It should be noted that by associating the fuzzified data with the risk levels, a decision-making basis is formed. The rules map combinations of data to risk levels, simulating expert judgment to ensure that the comprehensive impact is quantified.

[0158] Furthermore, combining the membership relationship of the fuzzy set of the input variables with the fuzzy inference rules, determine the activation degree of the fuzzy inference rules and generate the corresponding fuzzy output subset;

[0159] The activation degree of each rule is expressed as:

[0160] .

[0161] Among them, is the activation degree of the q-th rule; is at the input value the membership degree, indicating the degree of belonging to a certain fuzzy level; is the membership degree of d at the input value indicating the degree of d belonging to a certain fuzzy level; is the membership degree of v at the input value indicating the degree of v belonging to a certain fuzzy level;

[0162] The corresponding fuzzy output subset is expressed as:

[0163] .

[0164] Among them, is the clipped output subset of the q-th rule, is the output membership function of the q-th rule, is the risk value range variable;

[0165] It should be noted that the activation degree quantifies the triggering intensity of each rule, depending on the inference rules and the output of the membership function. The fuzzy output subset combines the activation degree of the rule with the output risk level, providing a basis for subsequent aggregation. The formula The minimum operation ensures the logical consistency of the rules, and the formula The clipping operation of ensures the accuracy of the rule contribution.

[0166] Furthermore, aggregate the output subsets of all rules, and generate a comprehensive membership function through the maximum operation, which is expressed as:

[0167] .

[0168] where N is the total number of rules;

[0169] It should be noted that by integrating the outputs of all rules, a unified fuzzy risk distribution is formed. The maximum operation retains the strongest contribution of each rule, ensuring that high-risk scenarios are not diluted and providing accurate inputs for defuzzification.

[0170] Furthermore, defuzzify the comprehensive membership function, and generate a scalar risk value using the weighted average method, which is expressed as:

[0171] .

[0172] where is the risk value;

[0173] Classify the contact risk according to the risk value r and the predetermined standard to determine the potential contact threat degree between the vegetation and the power line.

[0174] It should be noted that converting the fuzzy risk distribution into a single value facilitates decision-making. The numerical risk indicator provides inputs for subsequent classification.

[0175] In a specific embodiment of the present invention, a visualization model and an operation and maintenance instruction are generated. Specifically,

[0176] Generate a visualization three-dimensional model based on the vegetation growth trend and the contact risk, and mark the risk area in the form of a heat map in the visualization three-dimensional model to intuitively display the analysis results;

[0177] Automatically generate operation and maintenance instructions according to the risk level; when the risk level is high risk, trigger the instruction to trim the vegetation and specify the coordinate range to be trimmed;

[0178] Transmit the visualization model and the operation and maintenance instructions to the operation and maintenance terminal to ensure that the analysis results are applied to the power corridor management in real time.

[0179] It should be noted that the heat map maps the risk value r to the vegetation point through linear interpolation, uses a gradient of cold and warm tones (blue for low risk, red for high risk) to indicate the risk level, supports transparency adjustment to retain the details of the three-dimensional model, and is more intuitive than a two-dimensional report. Operation and maintenance instructions are generated based on the comparison of risk values ​​and safety standards. High-risk areas trigger pruning instructions, extract coordinates and specify ranges in combination with safety distances, and dynamically prioritize and optimize resource allocation. Compared with traditional methods, heat maps and automated instructions significantly improve risk visualization and management efficiency.

[0180] Furthermore, the system builds a 3D visualization model based on the fused point cloud, predicted height and risk value. The spatial coordinates of the point cloud reconstruct the vegetation canopy, align the power line positions, and dynamically integrate the predicted height and growth rate to reflect the future form. The thermal map marks the risk area, the color allocation is aligned with the risk level, and the transparency is dynamically adjusted to highlight the high-risk points without obscuring the geometric details.

[0181] Furthermore, the system generates operation and maintenance instructions based on the risk level. High-risk areas are judged by thresholds, and pruning instructions are generated to specify the coordinate range (point cloud index extraction) and the boundary meets the safety distance. Monitoring or observation instructions are generated for medium and low-risk areas. The instructions contain the operation type, priority and time range, and the heat map distribution guides precise positioning.

[0182] Furthermore, the system transmits the model and instructions to the operation and maintenance terminal. The model is stored in a compressed 3D format, and the instructions are structured data. The encrypted transmission supports real-time or periodic updates. The terminal decodes the model, displays the risk distribution, integrates the instructions to generate a task list, supports perspective adjustment and area query, and guides on-site management.

[0183] Optimally, the system optimizes visualization and command generation efficiency. Point cloud downsampling simplifies the model and retains key information; long corridors are processed in sections to reduce computational complexity; high-risk areas are sorted by risk value and resources are allocated first; transmission status monitoring is added to ensure reliable terminal reception.

[0184] In the second embodiment of the present invention, the present invention provides a power corridor vegetation management system based on spectral imaging and laser point cloud, such as Figure 2 As shown, the system includes a data fusion module 1, a growth prediction module 2 and a risk assessment module 3;

[0185] The data fusion module 1 is used to collect laser point cloud data and multispectral image data in the power corridor through a drone, and fuse them to generate a fused point cloud containing vegetation space and spectral characteristics through a dynamic nonlinear spectral band selection mechanism;

[0186] The growth prediction module 2 is used to extract vegetation features from the fused point cloud through a multi-scale graph convolutional network, construct a time series and use a bidirectional LSTM and self-attention mechanism to train a growth model to predict the growth trend of vegetation;

[0187] The risk assessment module 3 is used to comprehensively assess the contact risk between vegetation and power lines based on the spatial information and growth trend of vegetation using fuzzy logic, and to generate a visual model and operation and maintenance instructions.

[0188] In a third embodiment of the present invention, the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps of the power corridor vegetation management method based on spectral imaging and laser point cloud as described above are implemented.

[0189] In a fourth embodiment of the present invention, the present invention provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the power corridor vegetation management method based on spectral imaging and laser point cloud as described above.

[0190] In summary, the present invention provides a method and system for power corridor vegetation management based on spectral imaging and laser point cloud, which integrates multi-spectral imaging and laser point cloud data through a dynamic nonlinear spectral band selection mechanism to improve the accuracy of vegetation feature extraction; adopts a multi-scale graph convolutional network and a bidirectional LSTM combined with a self-attention mechanism to achieve accurate prediction of vegetation growth trends; and a comprehensive risk assessment method based on fuzzy logic fully considers the uncertainty of multivariate interactions and improves the robustness of contact risk assessment. The final generated visualization model and operation and maintenance instructions significantly enhance the automation and intelligence level of power corridor vegetation management, effectively reduce the risk of contact between vegetation and power lines, and ensure the safe operation of the power system.

[0191] It should be understood that although this specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each implementation mode may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0192] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the modules described above can refer to the corresponding process in the aforementioned method implementation, and will not be repeated here.

[0193] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present implementation scheme.

[0194] In addition, each functional module in each embodiment of the present application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0195] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for a computer system (which can be a personal computer, a server, or a network system, etc.) or a processor to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0196] Finally, it should be noted that the above implementation modes are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned implementation modes, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned implementation modes, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various implementation modes of the present application.

Claims

1. A vegetation management method for power corridors based on spectral imaging and laser point cloud, characterized by: include, The laser point cloud data and multispectral image data in the power corridor are collected by UAV, and then fused to generate a fused point cloud containing the spatial and spectral characteristics of vegetation through a dynamic nonlinear spectral band selection mechanism. The laser point cloud data within the power corridor is defined as ,in, is the three-dimensional coordinate of the i-th point, and n is the number of points; Define multispectral image data as ,in, is the reflection value of point i in band j, k is the number of bands; Infer vegetation type c and growth stage s from multispectral image data S through pre-trained classification models; Calculate the weight of each spectral band j, the formula is: ， in, is the standard deviation of band j, is a scalar representation of vegetation characteristics and temporal dynamics, τ is the data acquisition timestamp, is band specific, and is the pre-trained embedding vector, is the balance parameter, is the scaling factor; Calculate the spectral characteristics of each point, the formula is: ， in, is the weight of band j, is the mean of band j, is a nonlinear transformation function; The generated fused point cloud is represented as: ； The vegetation features are extracted from the fused point cloud through a multi-scale graph convolutional network, and the time series is constructed. The bidirectional LSTM and self-attention mechanism are used to train the growth model to predict the growth trend of vegetation. Based on the spatial information and growth trend of vegetation, fuzzy logic is used to comprehensively evaluate the contact risk between vegetation and power lines, and generate visualization models and operation and maintenance instructions.

2. The power corridor vegetation management method based on spectral imaging and laser point cloud according to claim 1 is characterized by: The method of extracting vegetation features from the fused point cloud by using a multi-scale graph convolutional network includes: Building a graph structure from fused point clouds , where the vertex set represents all points in the fused point cloud, and E is the edge set; Calculating edge weights Used to characterize the spatial relationship between points i and l; in, is the preset distance decay parameter used to control the neighborhood range; Defining multiple scales ,in Represents the neighborhood range of the pth scale, point i is in scale The neighborhood below is ; At every scale Extract local features, the formula is: ， in, is neighborhood aggregation, and is the learnable weight matrix and bias of the p-th scale; The global feature aggregation through maximum pooling is expressed as: ， The final feature representation of the output point i is: ， in, Represents the result of mapping the original features to a high-dimensional space through a multi-layer perceptron.

3. The power corridor vegetation management method based on spectral imaging and laser point cloud according to claim 2 is characterized by: The construction of the time series and the use of the bidirectional LSTM and self-attention mechanism to train the growth model include: Collect vegetation features at multiple time points and construct time series ,in, is the characteristic at the tth moment; Using bidirectional LSTM processing , calculate the forward hidden state and the reverse hidden state ; Merge the forward and backward hidden states to get a bidirectional hidden state ; By introducing the self-attention mechanism and based on the hidden state matrix , calculate the query matrix , key matrix , value matrix ; in, are the weights of the corresponding matrices respectively; Calculate the attention weight matrix and the context representation matrix ; in, The dimension of the key vector, used for scaling.

4. The power corridor vegetation management method based on spectral imaging and laser point cloud according to claim 3 is characterized by: The predicted vegetation growth trend includes: By formula Predict vegetation height at the next time point; in, is the tth column of the context representation matrix C, corresponding to the weighted context information of the tth time step, and the hidden state Aligned in the time dimension; are the predicted weight matrix and bias respectively; Calculate growth rate ,in, Predict the change in altitude for consecutive time points, is the time interval; When training the model, historical height data is used as a supervisory signal, and the model parameters are adjusted by optimizing the loss function. The loss function is expressed as: ， in, is the actual vegetation height at time step t, is the predicted vegetation height at time step t, is the total number of time steps in the time series.

5. The power corridor vegetation management method based on spectral imaging and laser point cloud according to claim 4 is characterized in that: The comprehensive assessment of the risk of contact between vegetation and power lines using fuzzy logic includes: Define input variables, including height difference , the shortest distance d between the vegetation canopy point and the power line on the horizontal plane, and the growth velocity v; in, is the height of the electric line, d extracted from the laser point cloud P; The fuzzy set membership relationship of the input variables is established through the fuzzy membership function to quantify the height difference , horizontal distance d and growth velocity v, including: For each input variable, define the membership function , mapping the input variables to fuzzy sets, including fuzzy semantic levels of safety level, transition level, and danger level; the membership function The form is determined based on the physical meaning of the input variables and the actual needs of power corridor management; Among them, X represents , d or v; Based on the fuzzy set membership, fuzzy reasoning rules between input variables and risk levels are configured; the fuzzy reasoning rules express the input variables through logical relationships. The impact of , d or v on exposure risk; Combining the fuzzy set membership of the input variables with the fuzzy inference rules, determining the activation degree of the fuzzy inference rules, and generating corresponding fuzzy output subsets; The activation degree of each rule is expressed as: ， in, is the activation degree of the qth rule; for In the input value The degree of membership at The degree to which it belongs to a certain level of ambiguity; Enter the value for d The degree of membership at , which indicates the degree to which d belongs to a certain fuzzy level; Enter the value for v The degree of membership at indicates the degree to which v belongs to a certain fuzzy level; The corresponding fuzzy output subset is expressed as: ， in, is the pruned output subset of the qth rule, is the output membership function of the qth rule, is the risk value range variable; Aggregate the output subsets of all rules and generate a comprehensive membership function through maximum operation, which is expressed as: ， Where N is the total number of rules; The comprehensive membership function is defuzzified and the scalar risk value is generated by weighted average method, which is expressed as: ， in, is the risk value; The contact risk is classified according to the risk value r and predetermined standards to determine the potential contact threat level between vegetation and power lines.

6. The power corridor vegetation management method based on spectral imaging and laser point cloud according to claim 1 is characterized by: The generation of visualization models and operation and maintenance instructions includes: Generate a visual three-dimensional model based on vegetation growth trends and contact risks, and mark risk areas in the visual three-dimensional model in the form of a heat map to intuitively display analysis results; Automatically generate operation and maintenance instructions based on risk levels; when the risk level is high, trigger the instruction to trim vegetation and specify the coordinate range that needs to be trimmed; Transmit visualization models and operation and maintenance instructions to the operation and maintenance terminal to ensure that the analysis results are applied to power corridor management in real time.

7. A power corridor vegetation management system based on spectral imaging and laser point cloud, applied to the power corridor vegetation management method based on spectral imaging and laser point cloud as claimed in claim 1, characterized in that: It includes data fusion module, growth prediction module and risk assessment module; The data fusion module is used to collect laser point cloud data and multispectral image data in the power corridor through the UAV, and fuse them to generate a fused point cloud containing vegetation space and spectral characteristics through a dynamic nonlinear spectral band selection mechanism; The growth prediction module is used to extract vegetation features from the fused point cloud through a multi-scale graph convolutional network, construct a time series and use a bidirectional LSTM and self-attention mechanism to train a growth model to predict the growth trend of vegetation; The risk assessment module is used to comprehensively assess the contact risk between vegetation and power lines based on the spatial information and growth trend of vegetation using fuzzy logic, and to generate a visual model and operation and maintenance instructions.

8. An electronic device, comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the power corridor vegetation management method based on spectral imaging and laser point cloud as described in any one of claims 1-6 are implemented.

9. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the power corridor vegetation management method based on spectral imaging and laser point cloud as described in any one of claims 1 to 6 are implemented.

Images