Bird hazard prevention and early warning method, system and device for power transmission line

Abstract

The application discloses a bird damage prevention and early warning method, system and device for a power transmission line, and relates to the technical field of data processing. The method comprises the following steps: obtaining real-time bird multi-modal monitoring data; synchronously performing preliminary identification to a deep learning fusion identification model to obtain an initial identification result; performing multi-directional subdivision migration verification to update the initial identification result and generate a multi-directional bird identification data set; performing dynamic evaluation to obtain a bird damage risk level, to early warn and classify the power transmission line, and to generate a classified early warning instruction; and executing the classified early warning instruction to trigger a prevention device to perform bird damage prevention intervention on the power transmission line. The technical problems that the bird activity monitoring precision around the power transmission line is insufficient, the identification result reliability is not high, and the bird damage risk early warning is not timely in the prior art are solved, the technical effect that the bird activity precision perception and dynamic evaluation are realized through multi-modal monitoring and fusion identification is achieved, and the timeliness and effectiveness of the bird damage prevention and early warning of the power transmission line are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method, system, and equipment for bird damage prevention and early warning of power transmission lines. Background Technology

[0002] With the continuous expansion of transmission line scale and the increasing complexity of the areas traversed by the lines, birds are increasingly frequenting transmission line towers, crossarms, and insulators for resting, nesting, and defecating. This bird-related faults, such as flashovers, short circuits, and equipment contamination, have become a significant factor affecting the safe and stable operation of transmission lines. Current bird-related prevention and control methods for transmission lines mainly rely on manual inspections, video surveillance, or single-sensor monitoring to identify and issue early warnings of bird activity. However, due to environmental factors such as complex terrain, climate change, and lighting conditions, existing technologies generally suffer from insufficient monitoring accuracy and poor identification stability when monitoring bird activity around transmission lines, making it difficult to capture bird activity characteristics in a timely and accurate manner. Furthermore, existing bird-related early warning methods are mostly based on static thresholds or single monitoring results, lacking comprehensive analysis of multi-source monitoring data and correlation assessment of historical line operation information. This leads to delayed bird-related risk assessment, untimely early warning responses, and a need to improve the targeting and effectiveness of prevention and control measures. Summary of the Invention

[0003] This application provides a method, system, and equipment for bird damage prevention and early warning of transmission lines, which solves the technical problems in the prior art of insufficient accuracy in monitoring bird activities around transmission lines and low reliability of identification results, resulting in untimely early warning of bird damage risks.

[0004] The first aspect of this application provides a bird damage prevention and early warning method for power transmission lines, the method comprising: A dual-layer positioning method is used to traverse the transmission line. Based on the dual positioning information, a multi-dimensional sensor group is deployed to collect data and obtain real-time multimodal bird monitoring data. A deep learning fusion recognition model is constructed, and the real-time multimodal bird monitoring data is synchronized to the deep learning fusion recognition model for preliminary recognition to obtain initial recognition results. Based on the initial recognition results, multi-directional subdivision transfer verification is performed to generate multiple verification values. The initial recognition results are updated according to the multiple verification values ​​to generate a multi-directional bird recognition dataset. The regional operation log of the transmission line is introduced, and dynamic evaluation is performed based on the regional operation log and the multi-directional bird recognition dataset to obtain the bird damage risk level and classify the transmission line for early warning, generating a graded early warning command. The graded early warning command is executed to trigger the prevention and control device to intervene in bird damage prevention and control of the transmission line.

[0005] A second aspect of this application provides a bird damage prevention and early warning system for power transmission lines, the system comprising: Data Acquisition Component: Traverses the transmission line for dual-layer positioning, deploys a multi-dimensional sensor group based on the dual positioning information to collect data, and obtains real-time multimodal bird monitoring data; Preliminary Identification Component: Constructs a deep learning fusion recognition model, synchronizes the real-time multimodal bird monitoring data to the deep learning fusion recognition model for preliminary identification, and obtains initial identification results; Result Verification Component: Performs multi-directional subdivision transfer verification based on the initial identification results, generates multiple verification values, updates the initial identification results based on the multiple verification values, and generates a multi-directional bird identification dataset; Risk Assessment Component: Introduces the regional operation log of the transmission line, performs dynamic assessment based on the regional operation log and the multi-directional bird identification dataset, obtains the bird damage risk level, classifies the transmission line for early warning, and generates graded early warning instructions; Prevention and Intervention Component: Executes the graded early warning instructions to trigger prevention and control devices to intervene in bird damage prevention and control of the transmission line.

[0006] A third aspect of this application provides an electronic device, comprising: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the bird damage prevention and early warning method for power transmission lines provided in this application.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, a dual-layer positioning method is used to traverse the transmission line. Based on the dual positioning information, a multi-dimensional sensor group is deployed to collect data, obtaining real-time multimodal bird monitoring data. Next, a deep learning fusion recognition model is constructed. The real-time multimodal bird monitoring data is synchronized to the deep learning fusion recognition model for preliminary recognition, obtaining initial recognition results. Further, multi-directional subdivision transfer verification is performed based on the initial recognition results, generating multiple verification values. The initial recognition results are updated based on these verification values, generating a multi-directional bird recognition dataset. Then, the regional operation logs of the transmission line are introduced. Dynamic evaluation is performed based on the regional operation logs and the multi-directional bird recognition dataset to obtain bird damage risk levels and classify the transmission line for early warning, generating graded early warning commands. Finally, the graded early warning commands are executed to trigger prevention and control devices to intervene in bird damage prevention and control of the transmission line. This method solves the technical problems of insufficient monitoring accuracy and low reliability of recognition results in existing technologies, leading to untimely bird damage risk warnings. It achieves the technical effect of accurate perception and dynamic assessment of bird activity through multimodal monitoring and fusion recognition, thereby improving the timeliness and effectiveness of bird damage prevention and control early warnings for transmission lines. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic flowchart of a bird damage prevention and early warning method for power transmission lines provided in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a bird damage prevention and early warning system for power transmission lines provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application.

[0010] Explanation of reference numerals in the attached figures: Data acquisition component 11, preliminary identification component 12, result verification component 13, risk assessment component 14, prevention and intervention component 15, processor 21, memory 22, input device 23, output device 24. Detailed Implementation

[0011] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0012] Example 1, as Figure 1 As shown in the embodiment of this application, a bird damage prevention and early warning method for power transmission lines is provided, wherein the method includes: The transmission lines are traversed to perform dual-layer positioning. Based on the dual positioning information, a multi-dimensional sensor group is deployed to collect data and obtain real-time multimodal bird monitoring data.

[0013] In this embodiment, the transmission line to be monitored is traversed segment by segment, discretized according to tower sections, span sections, or preset spatial intervals. For each traversed segment, corresponding geographical environmental information is collected, including geographical elements such as topographic relief, vegetation cover, water distribution, bird migration routes, and historically high-incidence areas of bird damage, constructing a geographic information layer reflecting the natural environmental characteristics surrounding the transmission line. Simultaneously, equipment operation information of the transmission line is collected, including tower locations, conductor directions, insulator string distribution, bird-proof device installation locations, and line structural parameters, constructing an equipment topology layer reflecting the equipment structure and operating status of the transmission line. After obtaining the geographic information layer and the equipment topology layer, they are mapped to a unified spatial coordinate system for overlay analysis. Based on spatial overlap, proximity, and weighted overlay relationships, a composite risk weight corresponding to each spatial location is calculated. The composite risk weight is used to characterize the degree of risk of birds affecting the transmission line when active at the corresponding location. Based on the composite risk weight, the transmission line is located using a dual-layer method, and key monitoring locations that simultaneously meet the geographical environmental risk characteristics and equipment structural sensitivity characteristics are selected as dual location information. After determining the dual location information, sensor deployment is planned based on the dual location information and the corresponding composite risk weights. Differentiated multi-dimensional sensor deployment strategies are configured for different key monitoring locations. Specifically, multi-dimensional sensor arrays, including visible light cameras, infrared cameras, acoustic sensors, and tower vibration sensors, are deployed at locations with high composite risk weights. At locations with low composite risk weights, basic monitoring sensors are deployed or the sampling frequency is reduced. Continuous data acquisition of the transmission line and its surrounding environment is achieved through this multi-dimensional sensor array, generating real-time multimodal bird monitoring data that includes video, audio, and vibration data.

[0014] Furthermore, a dual-layer positioning method is used to traverse the transmission line, and a multi-dimensional sensor group is deployed based on the dual positioning information to collect data and obtain real-time multimodal bird monitoring data. The method includes: The process involves: traversing the transmission lines to perform regional geographic structure analysis and constructing a geographic information layer; traversing the transmission lines to perform equipment location distribution analysis and constructing an equipment topology layer; performing a dual-layer overlay analysis based on the geographic information layer and the equipment topology layer to generate overlay analysis results; calculating spatial location weights based on the overlay analysis results to determine composite risk weights; performing dual-layer positioning based on the composite risk weights to determine dual positioning information; performing spatial analysis on the composite risk weights based on the dual positioning information to generate weighted spatial distribution data; setting up a sensor deployment array based on the weighted spatial distribution data; deploying multi-dimensional sensors according to the sensor deployment array to construct a multi-dimensional sensor group; and collecting data from the transmission lines using the multi-dimensional sensor group to obtain real-time bird multimodal monitoring data.

[0015] Preferably, the transmission line is segmented according to a preset traversal rule, which can be set based on tower number, line span, or spatial distance threshold. For each traversed segment, a geographical structure analysis of the area surrounding the transmission line is performed, collecting geographical environmental elements related to bird activity, including topography, vegetation distribution, water body location, airspace openness, and historically high-incidence areas of bird damage, to construct a geographic information layer reflecting the natural environmental characteristics surrounding the transmission line. Simultaneously, a location distribution analysis of the transmission line equipment within the traversed segment is performed, collecting information such as tower location, conductor direction, insulator string distribution, bird-proof device installation location, and line structural parameters, to construct an equipment topology layer reflecting the structural and spatial relationships of the transmission line equipment. After obtaining the geographic information layer and the equipment topology layer, they are mapped to a unified spatial coordinate system for dual-layer overlay analysis. By analyzing the spatial overlap, proximity, and correlation between geographic environmental elements and equipment structural elements, overlay analysis results are generated. Based on these results, spatial location weights are calculated for each location to comprehensively assess the potential risk of birds affecting power transmission lines at those locations, determining the composite risk weight for each location. Subsequently, dual-layer positioning is performed on the geographic information layer and the equipment topology layer according to the composite risk weights. Spatial locations with both high geographic environmental risk characteristics and sensitive equipment structural characteristics are selected to determine the dual positioning information. Based on this, spatial analysis is performed on the composite risk weights corresponding to the dual positioning information to generate weighted spatial distribution data, reflecting the bird damage risk distribution at different spatial locations. Based on the weighted spatial distribution data, a sensor deployment array is set up. A dense multi-dimensional sensor deployment strategy is configured in spatial locations with high composite risk weights, while a basic or low-frequency sampling sensor deployment strategy is configured in spatial locations with low composite risk weights. Multi-dimensional sensors, including video acquisition sensors, acoustic acquisition sensors, and tower vibration sensors, are deployed according to the sensor deployment array to construct a multi-dimensional sensor group. The multi-dimensional sensor group continuously collects data on the transmission line and its surrounding area to form real-time bird multimodal monitoring data containing video data, audio data, and vibration data. The real-time bird multimodal monitoring data is then output to the subsequent identification and evaluation process.

[0016] Furthermore, the method for determining dual-layer positioning information based on the composite risk weights includes: Spatial analysis is performed based on power transmission lines to construct a three-dimensional spatial coordinate system. The geographic information layer and the equipment topology layer are mapped to the three-dimensional spatial coordinate system for correlation analysis to generate correlation identification results. The correlation identification results are verified. If the verification is successful, dual-layer density analysis is performed on the correlation identification results to generate layer density parameters. Based on the layer density parameters, the geographic information layer and the equipment topology layer are traversed to identify priorities and construct a priority list. The geographic information layer and the equipment topology layer are linked for positioning according to the priority list, and the overlapping position information is extracted as the dual positioning information.

[0017] Preferably, spatial analysis is performed on the monitoring area based on the spatial distribution characteristics of the transmission line, and a three-dimensional spatial coordinate system is established with the transmission line direction as the reference axis. The three-dimensional spatial coordinate system includes at least the length coordinate along the line direction, the height coordinate perpendicular to the ground, and the lateral spatial coordinate, which are used to uniformly describe the spatial positional relationship around the transmission line.

[0018] After constructing a three-dimensional spatial coordinate system, the geographic information layer and the equipment topology layer are mapped into the three-dimensional spatial coordinate system. Through spatial location alignment and feature association analysis, the correspondence between geographic environmental features and line equipment features in the same or adjacent spatial locations is identified, generating association identification results. These association identification results characterize the degree of spatial coupling between geographic environmental features and equipment structural features. Subsequently, the validity of the association identification results is verified, including at least spatial consistency verification and data integrity verification. When the verification passes, dual-layer density analysis is performed on the association identification results to statistically analyze the feature distribution density of the geographic information layer and the equipment topology layer in each spatial unit, generating corresponding layer density parameters. Based on the layer density parameters, the geographic information layer and the equipment topology layer are traversed separately, and each spatial location is prioritized. The priority is determined based on the magnitude of the layer density parameter of the corresponding spatial location and its correlation with the composite risk weight, thereby constructing a geographic information layer priority list and an equipment topology layer priority list. Finally, the geographic information layer and the device topology layer are linked for positioning according to the priority list. Spatial locations that are simultaneously located in the high priority range are selected, and the overlapping location information corresponding to the spatial locations is extracted as dual positioning information for subsequent sensor deployment and monitoring data collection.

[0019] A deep learning fusion recognition model is constructed, and the real-time bird multimodal monitoring data is synchronized to the deep learning fusion recognition model for preliminary recognition to obtain initial recognition results.

[0020] Furthermore, a deep learning fusion recognition model is constructed, and the real-time bird multimodal monitoring data is synchronized to the deep learning fusion recognition model for preliminary recognition to obtain initial recognition results. The method includes: A deep learning fusion recognition model is constructed, which includes a visual recognition subnetwork, an acoustic recognition subnetwork, a vibration recognition subnetwork, and a cross-modal attention fusion module set up in parallel. Real-time multimodal bird monitoring data is synchronized to the deep learning fusion recognition model for preliminary classification, generating video streams, audio streams, and tower vibration signals. The video stream is synchronized to the visual recognition subnetwork to generate video recognition results, the audio stream is synchronized to the acoustic recognition subnetwork to generate audio recognition results, and the tower vibration signals are synchronized to the vibration recognition subnetwork to generate vibration recognition results. The video recognition results, audio recognition results, and vibration recognition results are synchronized to the cross-modal attention fusion module for adaptive weight fusion to obtain the initial recognition results.

[0021] In this embodiment of the application, a deep learning fusion recognition model is constructed to achieve multimodal joint identification of bird activity around power transmission lines. The deep learning fusion recognition model uses a basic architecture of multimodal input, parallel processing of multiple sub-networks, and fusion output to convert real-time bird monitoring data from different sensors into identification results that can be used for risk assessment.

[0022] Regarding model input configuration, the deep learning fusion recognition model employs at least three parallel sub-network input structures: a visual recognition sub-network, an acoustic recognition sub-network, and a vibration recognition sub-network. The visual recognition sub-network receives video streams or image sequences from image acquisition devices as input; the acoustic recognition sub-network receives audio signals from acoustic sensors; and the vibration recognition sub-network receives temporal vibration signals from tower vibration sensors. During the input phase, each sub-network synchronizes data through a unified time window configuration and sampling alignment mechanism, enabling joint processing of data from different modalities at the same time scale.

[0023] In terms of feature extraction structure, the visual recognition subnetwork extracts spatial and temporal features from the input video data and outputs a visual feature vector; the acoustic recognition subnetwork performs time-frequency feature analysis on the input audio signal and outputs an acoustic feature vector; and the vibration recognition subnetwork performs temporal pattern recognition on the input vibration signal and outputs a vibration feature vector. The feature vectors output by each subnetwork are uniformly mapped to the same feature space through linear mapping, normalization, or dimension alignment to facilitate subsequent fusion.

[0024] Regarding the feature fusion structure, the visual feature vector, acoustic feature vector, and vibration feature vector are input into the cross-modal attention fusion module. The cross-modal attention fusion module calculates the corresponding attention weights based on the correlation between different modal features, and performs weighted fusion of each modal feature according to the attention weights to generate a fused feature vector, which is used to comprehensively characterize the overall state of bird activity at the current monitoring time.

[0025] Regarding the model output structure, an identification output layer is constructed based on the fused feature vector, and the fused feature vector is classified or regressed to output the initial identification result. The initial identification result includes at least one of the following: bird presence status, bird activity area location, activity intensity, or behavior category. It is output in the form of structured data and used as input for subsequent multi-directional subdivision migration verification and bird damage risk assessment.

[0026] Preferably, the collected real-time multimodal bird monitoring data is synchronously input into a deep learning fusion recognition model for preliminary classification processing. The multimodal monitoring data is streamed according to the data source and signal type, forming corresponding video streams, audio streams, and pole vibration signals. The video stream contains video data reflecting the bird's appearance and spatial location information; the audio stream contains audio data reflecting the acoustic characteristics of bird calls or activities; and the pole vibration signal contains vibration data reflecting the impact of bird resting, touching, or flying actions on the pole. Subsequently, the video stream is synchronously input into a visual recognition subnetwork. Feature extraction and discriminant analysis are performed on the video stream to generate corresponding video recognition results. Similarly, the audio stream is synchronously input into an acoustic recognition subnetwork. Acoustic feature analysis and discriminant processing are performed on the audio stream to generate corresponding audio recognition results. Finally, the pole vibration signal is synchronously input into a vibration recognition subnetwork. Correlation analysis and feature discrimination are performed on the vibration signal to generate corresponding vibration recognition results. Finally, the video recognition results, audio recognition results, and vibration recognition results are synchronously transmitted to the cross-modal attention fusion module. Through adaptive fusion processing of weight allocation and attention guidance on the recognition results of different modalities, a fusion feature representation is generated. Based on the fusion feature representation, an initial recognition result is output to characterize the current bird activity state, which serves as the basic input for subsequent multi-directional subdivision migration verification and bird damage risk assessment.

[0027] A cross-modal attention fusion module is positioned after the visual recognition sub-network, the acoustic recognition sub-network, and the vibration recognition sub-network. It is used to jointly model and adaptively fuse the feature information output by different modal sub-networks. The input of the cross-modal attention fusion module includes the visual feature vector output by the visual recognition sub-network, the acoustic feature vector output by the acoustic recognition sub-network, and the vibration feature vector output by the vibration recognition sub-network. Each feature vector corresponds to monitoring data within the same time window.

[0028] The cross-modal attention fusion module includes a feature alignment unit, an attention weight calculation unit, and a feature fusion unit. The feature alignment unit performs dimensionality unification and temporal alignment on feature vectors from different sub-networks, enabling comparison and fusion of features from different modalities within the same feature space. The attention weight calculation unit calculates the attention weight of each modal feature in the current monitoring scenario based on the correlation between feature vectors from different modalities, characterizing the contribution of each modal information to bird activity recognition.

[0029] Specifically, the cross-modal attention fusion module performs correlation analysis on visual feature vectors, acoustic feature vectors, and vibration feature vectors, dynamically adjusts the attention weights corresponding to each modal feature based on the analysis results, and performs weighted processing on each modal feature vector according to the attention weights; then, the weighted modal feature vectors are fused to generate a fused feature vector, which is used to comprehensively characterize the overall characteristics of bird activities in multiple dimensions of vision, acoustics, and vibration.

[0030] Furthermore, the method includes synchronizing the video stream to the visual recognition subnetwork to generate a video recognition result, synchronizing the audio stream to the acoustic recognition subnetwork to generate an audio recognition result, and synchronizing the tower vibration signal to the vibration recognition subnetwork to generate a vibration recognition result. The video stream is processed frame by frame by the visual recognition subnetwork, multi-scale feature maps are extracted and fused to obtain multi-frame detection results; target tracking is performed based on the multi-frame detection results to generate multiple continuous trajectory segments of bird targets as video recognition results; the audio stream is processed frame by frame, multi-frame features are extracted, and a time-frequency feature matrix is ​​constructed based on the multi-frame features; the time-frequency feature matrix is ​​synchronized to the acoustic recognition subnetwork for acoustic discrimination to generate audio recognition results; a short-time Fourier transform is performed based on the tower vibration signal to generate a time-spectrum map; the time-spectrum map is synchronized to the vibration recognition subnetwork for bird correlation judgment to generate vibration recognition results.

[0031] Preferably, the video stream is processed frame by frame by a visual recognition subnetwork. Target feature extraction is performed on each video frame, and multi-scale feature maps are extracted at different scales. These multi-scale feature maps are then fused to obtain multi-frame detection results representing the spatial location and appearance features of bird targets. Based on these multi-frame detection results, bird targets in consecutive video frames are associated and matched, and target tracking is performed to generate multiple continuous trajectory segments reflecting the bird's movement process, which serve as the video recognition results. Simultaneously, the audio stream is processed frame by frame, and corresponding acoustic features are extracted within each audio frame. A time-frequency feature matrix is ​​constructed based on the multi-frame acoustic features. This time-frequency feature matrix is ​​synchronously input into the acoustic recognition subnetwork. Acoustic discriminant analysis is performed on the time-frequency feature matrix to identify acoustic patterns related to bird activity, generating audio recognition results. Furthermore, the acquired tower vibration signals undergo time-domain to frequency-domain conversion, and short-time Fourier transform is used to analyze the tower vibration signals, generating a time-frequency spectrum reflecting the time-frequency distribution characteristics of the vibration signals. This time-frequency spectrum is synchronously input into the vibration recognition subnetwork. By performing feature analysis and correlation judgment on the time-frequency spectrum, the vibration characteristics generated by bird behaviors such as perching, touching, or flying on the tower are identified, generating vibration recognition results. Through the above processing, video recognition results, audio recognition results, and vibration recognition results are obtained respectively, and each recognition result is output to the subsequent cross-modal fusion and verification process.

[0032] Furthermore, the method of synchronizing the video recognition result, the audio recognition result, and the vibration recognition result to the cross-modal attention fusion module for adaptive weight fusion to obtain the initial recognition result includes: The video recognition results, audio recognition results, and vibration recognition results are synchronized to the cross-modal attention fusion module for feature analysis, generating visual features, acoustic features, and vibration features. The visual features have visual feature vectors, the acoustic features have acoustic feature vectors, and the vibration features have vibration feature vectors. Based on the visual features and acoustic features, cross-fusion analysis is performed to calculate a first cross-attention weight. Based on the visual features and vibration features, cross-fusion analysis is performed to calculate a second cross-attention weight. The acoustic feature vector and the vibration feature vector are weighted according to the first cross-attention weight and the second cross-attention weight to generate a weighted result. The visual feature vector is concatenated to the weighted result to construct a fused feature vector. Based on the fused feature vector, regression analysis is performed to construct the initial recognition result.

[0033] Preferably, unified feature analysis processing is performed on the video recognition results, audio recognition results, and vibration recognition results to extract visual features, acoustic features, and vibration features to characterize the bird's activity state. The visual features are encoded as visual feature vectors, the acoustic features as acoustic feature vectors, and the vibration features as vibration feature vectors, to achieve standardized representation of different modal features in the feature space. Cross-fusion analysis is performed based on the correlation between the visual and acoustic feature vectors. By calculating the correlation and matching degree between them, a first cross-attention weight is generated to characterize the influence of visual features on acoustic features. Simultaneously, cross-fusion analysis is performed based on the correlation between the visual and vibration feature vectors to calculate a second cross-attention weight to characterize the influence of visual features on vibration features. Based on the first and second cross-attention weights, adaptive weighting processing is performed on the acoustic and vibration feature vectors, suppressing feature components with low correlation to visual features and strengthening feature components with high correlation to visual features, generating a weighted result. On this basis, the visual feature vector and the weighted result are concatenated to construct a fused feature vector containing multimodal information. Regression analysis is performed based on fused feature vectors, and the output is an initial identification result that characterizes the bird presence status, activity intensity or behavioral risk at the current monitoring time. This serves as the basic input for subsequent multi-directional subdivision migration verification and dynamic assessment of bird damage risk.

[0034] After obtaining the fused feature vector, it is input into a preset regression analysis module for processing. This module establishes a mapping between the fused feature vector and bird activity status, performing numerical prediction and state estimation. Specifically, by weighting and nonlinearly mapping each feature dimension of the fused feature vector, a continuous prediction result characterizing the relevance of bird activity is output. This continuous prediction result reflects the probability of bird presence, activity intensity, or risk correlation. Based on this continuous prediction result, it is parsed according to preset identification rules, converting it into a structured initial identification result. This initial identification result includes at least the conclusion of bird presence, the corresponding spatial location identifier, and confidence parameters related to bird activity, serving as the basic input for subsequent multi-directional subdivision migration verification and bird damage risk assessment.

[0035] Based on the initial recognition results, multi-directional subdivision migration verification is performed to generate multiple verification values. The initial recognition results are then updated according to the multiple verification values ​​to generate a multi-directional bird recognition dataset.

[0036] Furthermore, based on the initial recognition result, multi-directional subdivision transfer validation is performed to generate multiple validation values. The initial recognition result is then updated according to the multiple validation values ​​to generate a multi-directional bird recognition dataset. The method includes: A multi-directional subdivision transfer verification process is initiated based on the initial recognition results. This process includes at least a fine-grained visual verification direction, a spatiotemporal behavior verification direction, and a cross-modal consistency verification direction, all of which are parallel directions. Based on the initial recognition results, localization and cropping are performed to determine regions of high detail interest (GPIs). These GPIs are then verified using the fine-grained visual verification direction to generate a first verification value. Based on the initial recognition results, continuous dynamic recording of bird targets is performed to construct motion trajectory sequences and behavioral state time sequences. A bird behavior knowledge graph is introduced, and the motion trajectory sequences and behavioral state time sequences are mapped to the bird behavior knowledge graph for matching, generating multiple behavior matching results. These multiple behavior matching results are then verified using the spatiotemporal behavior verification direction to generate a second verification value. The video recognition results, audio recognition results, and vibration recognition results are compared to calculate a multi-modal consistency metric. Finally, the multi-modal consistency metric is verified using the cross-modal consistency verification direction to generate a third verification value.

[0037] Preferably, the initial recognition result is used as a unified input to start the multi-directional subdivision transfer verification process; the multi-directional subdivision transfer verification process sets at least a fine-grained visual verification direction, a spatiotemporal behavior verification direction, and a cross-modal consistency verification direction in parallel, and each verification direction runs independently to review and correct the initial recognition result from different dimensions.

[0038] In the fine-grained visual verification direction, the bird target position in the video recognition result is re-localized and cropped based on the initial recognition result, a high-detail region of interest containing key appearance features is determined, and a fine-grained feature comparison and consistency verification is performed on the high-detail region of interest to generate a first verification value to characterize the reliability of visual recognition.

[0039] In the spatiotemporal behavior verification direction, bird targets are continuously and dynamically recorded based on the initial identification results to construct corresponding motion trajectory sequences and behavioral state time sequences that change over time. A bird behavior knowledge graph is introduced, and the motion trajectory sequences and behavioral state time sequences are mapped to the bird behavior knowledge graph for behavior pattern matching to generate multiple behavior matching results. Based on the spatiotemporal behavior verification direction, the rationality and consistency of the multiple behavior matching results are verified to generate a second verification value to characterize the conformity of bird behavior.

[0040] In the cross-modal consistency verification direction, the video recognition results, audio recognition results, and vibration recognition results are jointly compared and analyzed. The multimodal consistency measure of different modal recognition results in terms of time dimension, spatial dimension, and discrimination result is calculated. The multimodal consistency measure is verified based on the cross-modal consistency verification direction to generate a third verification value to characterize the degree of consistency of multimodal recognition.

[0041] After obtaining the first, second, and third verification values, the multiple verification values ​​are summarized and correlated. The initial identification results are updated and corrected according to the credibility weights corresponding to each verification value to form updated bird identification results. Furthermore, the updated bird identification results are associated and stored with the multiple verification values ​​and their source verification directions to generate a multi-directional bird identification dataset containing multi-directional verification information for subsequent bird damage risk assessment and early warning decision-making.

[0042] After obtaining multiple verification values, to reliably correct and update the initial recognition result, the multiple verification values ​​are first uniformly normalized to ensure that the verification values ​​generated from different verification directions are within a comparable numerical range. Then, verification weights are configured for each verification value based on the verification confidence level corresponding to each verification direction. These verification weights reflect the relative importance of fine-grained visual verification, spatiotemporal behavioral verification, and cross-modal consistency verification in the overall recognition update process. Based on this, the normalized multiple verification values ​​are weighted and fused with their corresponding verification weights to calculate a comprehensive verification score. The comprehensive verification score is compared with a preset update threshold. When the comprehensive verification score meets the update conditions, the initial recognition result is determined to be a high-confidence result, and the category label, confidence parameter, or behavioral judgment result of the initial recognition result is enhanced and updated. When the comprehensive verification score is within a preset transition range, the initial recognition result undergoes partial correction and update, adjusting its confidence level or retaining multiple candidate recognition results. When the comprehensive verification score is below the preset update threshold, the initial recognition result is determined to be a low-confidence result, and the initial recognition result is suppressed or marked as awaiting further observation. Through the weighted fusion and threshold determination process based on multiple verification values ​​described above, the initial identification results are dynamically updated to generate updated bird identification results. The updated bird identification results retain the initial identification information while incorporating multi-directional verification results, which serve as the input basis for constructing a multi-directional bird identification dataset and subsequent dynamic assessment of bird damage risk.

[0043] The regional operation log of the transmission line is introduced, and dynamic evaluation is performed based on the regional operation log and the multi-directional bird recognition dataset to obtain the bird damage risk level, and the transmission line is classified into early warning levels to generate classified early warning instructions.

[0044] After generating a multi-directional bird recognition dataset, in order to achieve dynamic assessment and graded early warning of bird damage risk to transmission lines, a regional operation log corresponding to the transmission lines is introduced. The regional operation log includes at least transmission line equipment status information and meteorological environment information. The equipment status information includes tower operation status, insulator status, line load level, and existing bird protection device operation status, etc. The meteorological environment information includes wind speed, wind direction, rainfall, temperature, visibility, and seasonal climate characteristics, etc.

[0045] In the dynamic assessment process, firstly, real-time threat dimension indicators reflecting the current bird activity intensity, frequency, behavior type, and recognition reliability are extracted based on a multi-directional bird identification dataset. Simultaneously, historical environmental dimension indicators reflecting the transmission line's operating environment and capacity are extracted based on regional operation logs. The real-time threat dimension indicators and the historical environmental dimension indicators are then time-aligned and weighted to construct a comprehensive bird damage risk value to comprehensively characterize the bird damage risk level. Further, a bird damage risk level mapping rule is constructed. Based on the correspondence between the comprehensive bird damage risk value and preset risk intervals, the comprehensive bird damage risk value is mapped to multiple discrete bird damage risk levels, and the transmission lines are classified into early warning levels based on these levels. For different bird damage risk levels, corresponding graded early warning instructions are generated. These instructions at least indicate the early warning level, the affected line section, and the recommended prevention and control response strategy.

[0046] Furthermore, based on the regional operation logs and the multi-directional bird recognition dataset, a dynamic assessment is performed to obtain the bird damage risk level, and a warning classification is applied to the transmission line to generate a graded warning instruction. The method includes: Real-time threat dimension indicators are extracted based on the multi-directional bird identification dataset, and historical environmental dimension indicators are extracted based on the regional operation log. The real-time threat dimension indicators and the historical environmental dimension indicators are weighted and fused to construct a comprehensive bird damage risk value. A risk level mapping rule is constructed, and the comprehensive bird damage risk value is mapped to a bird damage risk level based on the risk level. Risk is described based on the bird damage risk level, and core risk features are defined. Multi-level early warning analysis is performed on the transmission line according to the core risk features, and the graded early warning instructions are generated.

[0047] Preferably, real-time threat dimension indicators reflecting the current bird activity status are extracted based on a multi-directional bird identification dataset. These real-time threat dimension indicators include at least bird activity frequency, activity intensity, behavior type distribution, and corresponding identification confidence parameters, used to characterize the immediate threat level posed by bird activity to the transmission line. Simultaneously, historical environmental dimension indicators reflecting the transmission line's operating environment and historical operating conditions are extracted based on regional operation logs. These historical environmental dimension indicators include at least the transmission line equipment operating status, historical fault or bird damage records, and meteorological environmental characteristics for the corresponding time period. After obtaining the real-time threat dimension indicators and historical environmental dimension indicators, time-dimensional alignment and weight configuration are performed on the two types of indicators. A comprehensive bird damage risk value is constructed through weighted fusion to comprehensively characterize the degree of bird damage risk to the transmission line. The weight configuration is used to balance the impact of real-time bird activity threats and historical operating environment on the risk assessment results. Furthermore, a risk level mapping rule is constructed to define the correspondence between the comprehensive bird damage risk value and different risk level intervals. Based on this rule, the comprehensive bird damage risk value is mapped to the corresponding bird damage risk level, and a risk description is performed around each risk level, extracting core risk features that characterize the current risk causes and impacts. Finally, multi-level early warning analysis is performed on the transmission lines based on the core risk features, generating corresponding graded early warning instructions for different bird damage risk levels. These instructions at least indicate the early warning level, the affected transmission line section, and the corresponding response priority, providing a basis for subsequent linkage control and operation and maintenance decisions for bird damage prevention devices.

[0048] The execution of the graded early warning command triggers the prevention and control device to intervene in bird damage prevention and control on the transmission line.

[0049] After generating the tiered early warning command, the bird control devices in the corresponding sections of the transmission line are triggered and controlled according to the command to implement targeted prevention and control interventions for bird activities. The control devices include at least one or more of the following: sound and light bird deterrent devices, bionic predator devices, and ultrasonic transmitters, and can be configured and linked for different bird risk levels.

[0050] Specifically, when executing tiered early warning commands, the system first analyzes the warning level, target transmission line section, and corresponding core risk characteristics contained in the command. Based on the warning level, a prevention and intervention strategy is determined. When the bird risk level is low, a low-intensity prevention mode is triggered, using intermittent activation of sound and light bird deterrent devices or low-power ultrasonic transmitters to gently drive away birds. When the bird risk level is moderate, a combined prevention mode is triggered, using coordinated activation of sound and light bird deterrent devices and bionic predator devices to enhance the deterrent effect on birds. When the bird risk level is high, a high-intensity prevention mode is triggered, using synchronous or sequential activation of sound and light bird deterrent devices, bionic predator devices, and ultrasonic transmitters to continuously drive away birds from the target area.

[0051] During the operation of the prevention and control device, the working parameters of the device are set according to the graded early warning instructions, including the start-up sequence, working duration, power level and working frequency, and the prevention and control effect is monitored within the preset time window. When the intensity of bird activity decreases or the bird damage risk level decreases, the working intensity of the prevention and control device is automatically reduced or the prevention and control intervention is terminated. When bird activity continues to be detected, the working parameters of the prevention and control device can be maintained or adjusted to achieve dynamic prevention and control of bird damage risk to transmission lines.

[0052] By employing the above methods, the coordinated execution of tiered early warning commands and prevention and control devices can be achieved, ensuring that bird damage prevention and control interventions are matched with bird damage risk levels, thereby improving the pertinence, timeliness, and overall operational safety of bird damage prevention and control for transmission lines.

[0053] In summary, the embodiments of this application have at least the following technical effects: First, a dual-layer positioning method is used to traverse the transmission line. Based on the dual positioning information, a multi-dimensional sensor group is deployed to collect data, obtaining real-time multimodal bird monitoring data. Next, a deep learning fusion recognition model is constructed. The real-time multimodal bird monitoring data is synchronized to the deep learning fusion recognition model for preliminary recognition, obtaining initial recognition results. Further, multi-directional subdivision transfer verification is performed based on the initial recognition results, generating multiple verification values. The initial recognition results are updated based on these verification values, generating a multi-directional bird recognition dataset. Then, the regional operation logs of the transmission line are introduced. Dynamic evaluation is performed based on the regional operation logs and the multi-directional bird recognition dataset to obtain bird damage risk levels and classify the transmission line for early warning, generating graded early warning commands. Finally, the graded early warning commands are executed to trigger prevention and control devices to intervene in bird damage prevention and control of the transmission line. This method solves the technical problems of insufficient monitoring accuracy and low reliability of recognition results in existing technologies, leading to untimely bird damage risk warnings. It achieves the technical effect of accurate perception and dynamic assessment of bird activity through multimodal monitoring and fusion recognition, thereby improving the timeliness and effectiveness of bird damage prevention and control early warnings for transmission lines.

[0054] Example 2, based on the same inventive concept as the bird damage prevention and early warning method for transmission lines in the foregoing examples, such as... Figure 2 As shown, this application provides a bird damage prevention and early warning system for power transmission lines, wherein the system includes: Data Acquisition Component 11: Traverses the transmission line for dual-layer positioning, deploys a multi-dimensional sensor group based on the dual positioning information to collect data, and obtains real-time bird multimodal monitoring data; Preliminary Identification Component 12: Constructs a deep learning fusion identification model, synchronizes the real-time bird multimodal monitoring data to the deep learning fusion identification model for preliminary identification, and obtains initial identification results; Result Verification Component 13: Performs multi-directional subdivision migration verification based on the initial identification results, generates multiple verification values, updates the initial identification results based on the multiple verification values, and generates a multi-directional bird identification dataset; Risk Assessment Component 14: Introduces the regional operation log of the transmission line, performs dynamic assessment based on the regional operation log combined with the multi-directional bird identification dataset, obtains the bird damage risk level, classifies the transmission line for early warning, and generates a graded early warning instruction; Prevention and Intervention Component 15: Executes the graded early warning instruction to trigger the prevention and control device to intervene in bird damage prevention and control of the transmission line.

[0055] Furthermore, the data acquisition component 11 is used to perform the following methods: The process involves: traversing the transmission lines to perform regional geographic structure analysis and constructing a geographic information layer; traversing the transmission lines to perform equipment location distribution analysis and constructing an equipment topology layer; performing a dual-layer overlay analysis based on the geographic information layer and the equipment topology layer to generate overlay analysis results; calculating spatial location weights based on the overlay analysis results to determine composite risk weights; performing dual-layer positioning based on the composite risk weights to determine dual positioning information; performing spatial analysis on the composite risk weights based on the dual positioning information to generate weighted spatial distribution data; setting up a sensor deployment array based on the weighted spatial distribution data; deploying multi-dimensional sensors according to the sensor deployment array to construct a multi-dimensional sensor group; and collecting data from the transmission lines using the multi-dimensional sensor group to obtain real-time bird multimodal monitoring data.

[0056] Furthermore, the data acquisition component 11 is used to perform the following methods: Spatial analysis is performed based on power transmission lines to construct a three-dimensional spatial coordinate system. The geographic information layer and the equipment topology layer are mapped to the three-dimensional spatial coordinate system for correlation analysis to generate correlation identification results. The correlation identification results are verified. If the verification is successful, dual-layer density analysis is performed on the correlation identification results to generate layer density parameters. Based on the layer density parameters, the geographic information layer and the equipment topology layer are traversed to identify priorities and construct a priority list. The geographic information layer and the equipment topology layer are linked for positioning according to the priority list, and the overlapping position information is extracted as the dual positioning information.

[0057] Furthermore, the preliminary identification component 12 is used to perform the following method: A deep learning fusion recognition model is constructed, which includes a visual recognition subnetwork, an acoustic recognition subnetwork, a vibration recognition subnetwork, and a cross-modal attention fusion module set up in parallel. Real-time multimodal bird monitoring data is synchronized to the deep learning fusion recognition model for preliminary classification, generating video streams, audio streams, and tower vibration signals. The video stream is synchronized to the visual recognition subnetwork to generate video recognition results, the audio stream is synchronized to the acoustic recognition subnetwork to generate audio recognition results, and the tower vibration signals are synchronized to the vibration recognition subnetwork to generate vibration recognition results. The video recognition results, audio recognition results, and vibration recognition results are synchronized to the cross-modal attention fusion module for adaptive weight fusion to obtain the initial recognition results.

[0058] Furthermore, the preliminary identification component 12 is used to perform the following method: The video stream is processed frame by frame by the visual recognition subnetwork, multi-scale feature maps are extracted and fused to obtain multi-frame detection results; target tracking is performed based on the multi-frame detection results to generate multiple continuous trajectory segments of bird targets as video recognition results; the audio stream is processed frame by frame, multi-frame features are extracted, and a time-frequency feature matrix is ​​constructed based on the multi-frame features; the time-frequency feature matrix is ​​synchronized to the acoustic recognition subnetwork for acoustic discrimination to generate audio recognition results; a short-time Fourier transform is performed based on the tower vibration signal to generate a time-spectrum map; the time-spectrum map is synchronized to the vibration recognition subnetwork for bird correlation judgment to generate vibration recognition results.

[0059] Furthermore, the preliminary identification component 12 is used to perform the following method: The video recognition results, audio recognition results, and vibration recognition results are synchronized to the cross-modal attention fusion module for feature analysis, generating visual features, acoustic features, and vibration features. The visual features have visual feature vectors, the acoustic features have acoustic feature vectors, and the vibration features have vibration feature vectors. Based on the visual features and acoustic features, cross-fusion analysis is performed to calculate a first cross-attention weight. Based on the visual features and vibration features, cross-fusion analysis is performed to calculate a second cross-attention weight. The acoustic feature vector and the vibration feature vector are weighted according to the first cross-attention weight and the second cross-attention weight to generate a weighted result. The visual feature vector is concatenated to the weighted result to construct a fused feature vector. Based on the fused feature vector, regression analysis is performed to construct the initial recognition result.

[0060] Furthermore, the result verification component 13 is used to perform the following method: A multi-directional subdivision transfer verification process is initiated based on the initial recognition results. This process includes at least a fine-grained visual verification direction, a spatiotemporal behavior verification direction, and a cross-modal consistency verification direction, all of which are parallel directions. Based on the initial recognition results, localization and cropping are performed to determine regions of high detail interest (GPIs). These GPIs are then verified using the fine-grained visual verification direction to generate a first verification value. Based on the initial recognition results, continuous dynamic recording of bird targets is performed to construct motion trajectory sequences and behavioral state time sequences. A bird behavior knowledge graph is introduced, and the motion trajectory sequences and behavioral state time sequences are mapped to the bird behavior knowledge graph for matching, generating multiple behavior matching results. These multiple behavior matching results are then verified using the spatiotemporal behavior verification direction to generate a second verification value. The video recognition results, audio recognition results, and vibration recognition results are compared to calculate a multi-modal consistency metric. Finally, the multi-modal consistency metric is verified using the cross-modal consistency verification direction to generate a third verification value.

[0061] Furthermore, the risk assessment component 14 is used to perform the following methods: Real-time threat dimension indicators are extracted based on the multi-directional bird identification dataset, and historical environmental dimension indicators are extracted based on the regional operation log. The real-time threat dimension indicators and the historical environmental dimension indicators are weighted and fused to construct a comprehensive bird damage risk value. A risk level mapping rule is constructed, and the comprehensive bird damage risk value is mapped to a bird damage risk level based on the risk level. Risk is described based on the bird damage risk level, and core risk features are defined. Multi-level early warning analysis is performed on the transmission line according to the core risk features, and the graded early warning instructions are generated.

[0062] Example 3, Figure 3 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention, showing a block diagram of an exemplary electronic device suitable for implementing the embodiments of the present invention. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of the present invention. Figure 3 As shown, the electronic device includes a processor 21, a memory 22, an input device 23, and an output device 24; the number of processors 21 in the electronic device can be one or more. Figure 3 Taking a processor 21 as an example, the processor 21, memory 22, input device 23, and output device 24 in an electronic device can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0063] The memory 22, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the bird damage prevention and early warning method for power transmission lines in this embodiment of the invention. The processor 21 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 22, thereby realizing the aforementioned bird damage prevention and early warning method for power transmission lines.

[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A bird damage prevention and early warning method for power transmission lines, characterized in that, The method includes: The transmission lines are traversed to perform dual-layer positioning. Based on the dual positioning information, a multi-dimensional sensor group is deployed to collect data and obtain real-time multimodal bird monitoring data. A deep learning fusion recognition model is constructed, and the real-time bird multimodal monitoring data is synchronized to the deep learning fusion recognition model for preliminary recognition to obtain initial recognition results; Based on the initial recognition results, multi-directional subdivision transfer verification is performed to generate multiple verification values. The initial recognition results are then updated according to the multiple verification values ​​to generate a multi-directional bird recognition dataset. The regional operation log of the transmission line is introduced. Based on the regional operation log and the multi-directional bird recognition dataset, dynamic evaluation is performed to obtain the bird damage risk level and to classify the transmission line for early warning, generating a graded early warning instruction. The execution of the graded early warning command triggers the prevention and control device to intervene in bird damage prevention and control on the transmission line.

2. The bird damage prevention and early warning method for power transmission lines as described in claim 1, characterized in that, A dual-layer positioning method is used to traverse the transmission line. Based on the dual positioning information, a multi-dimensional sensor group is deployed to collect data and obtain real-time multimodal bird monitoring data. The method includes: Traverse the transmission lines to perform regional geographic structure analysis and construct geographic information layers; Traverse the transmission lines to analyze the location distribution of line equipment and construct an equipment topology layer; Based on the geographic information layer and the device topology layer, a dual-layer overlay analysis is performed to generate the overlay analysis results. Based on the results of the overlay analysis, spatial location weights are calculated to determine composite risk weights. Dual-layer positioning is performed based on the composite risk weights to determine dual positioning information; Based on the dual positioning information, spatial analysis is performed on the composite risk weights to generate weight spatial distribution data. The sensor deployment array is set according to the weighted spatial distribution data, and the multidimensional sensors are deployed according to the sensor deployment array to construct a multidimensional sensor group. The multi-dimensional sensor array is used to collect data from the power transmission line to obtain real-time multimodal bird monitoring data.

3. The bird damage prevention and early warning method for power transmission lines as described in claim 2, characterized in that, Based on the composite risk weights, dual-layer localization is performed to determine dual localization information. The method includes: Spatial analysis based on power transmission lines is performed to construct a three-dimensional spatial coordinate system. The geographic information layer and the device topology layer are mapped to the three-dimensional spatial coordinate system for correlation analysis, generating correlation identification results; Verify the association recognition result. If the verification passes, perform dual-layer density analysis on the association recognition result to generate layer density parameters. Based on the layer density parameter, the geographic information layer and the device topology layer are traversed respectively to identify the priority and construct a priority list; The geographic information layer and the device topology layer are linked for positioning according to the priority list, and the overlapping location information is extracted as the dual positioning information.

4. The bird damage prevention and early warning method for transmission lines as described in claim 1, characterized in that, A deep learning fusion recognition model is constructed, and the real-time bird multimodal monitoring data is synchronized to the deep learning fusion recognition model for preliminary recognition to obtain initial recognition results. The method includes: A deep learning fusion recognition model is constructed, wherein the deep learning fusion recognition model is configured in parallel with a visual recognition subnetwork, an acoustic recognition subnetwork, a vibration recognition subnetwork, and a cross-modal attention fusion module; The real-time bird multimodal monitoring data is synchronized to the deep learning fusion recognition model for preliminary classification, and video stream, audio stream, and tower vibration signal are generated. The video stream is synchronized to the visual recognition sub-network to generate video recognition results, the audio stream is synchronized to the acoustic recognition sub-network to generate audio recognition results, and the tower vibration signal is synchronized to the vibration recognition sub-network to generate vibration recognition results. The video recognition result, the audio recognition result, and the vibration recognition result are synchronized to the cross-modal attention fusion module for adaptive weight fusion to obtain the initial recognition result.

5. The bird damage prevention and early warning method for transmission lines as described in claim 4, characterized in that, The method includes synchronizing the video stream to a visual recognition subnetwork to generate a video recognition result, synchronizing the audio stream to an acoustic recognition subnetwork to generate an audio recognition result, and synchronizing the tower vibration signal to a vibration recognition subnetwork to generate a vibration recognition result. The video stream is processed frame by frame by the visual recognition sub-network, multi-scale feature maps are extracted and fused to obtain multi-frame detection results; Based on the multi-frame detection results, target tracking is performed to generate multiple continuous trajectory segments of bird targets as video recognition results; The audio stream is segmented into frames, multi-frame features are extracted, and a time-frequency feature matrix is ​​constructed based on the multi-frame features. The time-frequency feature matrix is ​​synchronized to the acoustic recognition sub-network for acoustic discrimination to generate audio recognition results; A short-time Fourier transform is performed on the tower vibration signal to generate a time-frequency spectrum. The time-spectrum image is synchronized to the vibration recognition sub-network to determine bird correlation and generate vibration recognition results.

6. The bird damage prevention and early warning method for transmission lines as described in claim 4, characterized in that, The method involves synchronizing the video recognition result, the audio recognition result, and the vibration recognition result to the cross-modal attention fusion module for adaptive weight fusion to obtain the initial recognition result. The video recognition results, audio recognition results, and vibration recognition results are synchronized to the cross-modal attention fusion module for feature analysis to generate visual features, acoustic features, and vibration features. The visual features have visual feature vectors, the acoustic features have acoustic feature vectors, and the vibration features have vibration feature vectors. Based on the visual features and the acoustic features, a cross-fusion analysis is performed to calculate the first cross-attention weight; Based on the visual features and the vibration features, a cross-fusion analysis is performed to calculate the second cross-attention weight; The acoustic feature vector and the vibration feature vector are weighted according to the first cross-attention weight and the second cross-attention weight to generate a weighted result; The visual feature vectors are concatenated to the weighted result to construct a fused feature vector; Regression analysis is performed based on the fused feature vector to construct the initial recognition result.

7. The bird damage prevention and early warning method for power transmission lines as described in claim 4, characterized in that, Based on the initial recognition result, multi-directional subdivision transfer validation is performed to generate multiple validation values. The initial recognition result is then updated based on the multiple validation values ​​to generate a multi-directional bird recognition dataset. The method includes: A multi-directional subdivision transfer verification process is initiated on the initial recognition result. The multi-directional subdivision transfer verification process includes at least a fine-grained visual verification direction, a spatiotemporal behavior verification direction, and a cross-modal consistency verification direction. The fine-grained visual verification direction, the spatiotemporal behavior verification direction, and the cross-modal consistency verification direction are parallel directions. Based on the initial recognition result, localization and cropping are performed to determine the region of interest in high detail. The region of interest in high detail is then verified based on the fine-grained visual verification direction to generate a first verification value. Based on the initial identification results, bird targets are continuously and dynamically recorded to construct motion trajectory sequences and behavioral state time series. A bird behavior knowledge graph is introduced, and the movement trajectory sequence and the behavioral state time sequence are mapped to the bird behavior knowledge graph for matching, generating multiple behavior matching results; Based on the spatiotemporal behavior verification direction, multiple behavior matching results are verified to generate a second verification value; The video recognition results, the audio recognition results, and the vibration recognition results are compared to calculate a multimodal consistency metric. The multimodal consistency metric is validated based on the cross-modal consistency validation direction to generate a third validation value.

8. The bird damage prevention and early warning method for transmission lines as described in claim 1, characterized in that, Based on the regional operation log and the multi-directional bird recognition dataset, a dynamic assessment is performed to obtain the bird damage risk level, and a warning classification is applied to the transmission line to generate a graded warning instruction. The method includes: Real-time threat dimension indicators are extracted based on the multi-directional bird identification dataset, and historical environmental dimension indicators are extracted based on the regional operation log. The real-time threat dimension indicators and the historical environment dimension indicators are weighted and fused to construct a comprehensive bird damage risk value; Construct risk level mapping rules, map the comprehensive value of bird damage risk to bird damage risk level based on the risk level mapping rules, describe the risk based on the bird damage risk level, and define the core characteristics of the risk; Based on the core risk characteristics, a multi-level early warning analysis is performed on the transmission lines to generate the graded early warning instructions.

9. A bird damage prevention and early warning system for power transmission lines, characterized in that, For implementing the bird damage prevention and early warning method for power transmission lines according to any one of claims 1-8, the system comprises: Data acquisition component: Traverse the transmission line to perform dual-layer positioning, deploy multi-dimensional sensor groups based on dual positioning information to collect data and obtain real-time multimodal bird monitoring data; Preliminary identification component: Construct a deep learning fusion identification model, synchronize the real-time bird multimodal monitoring data to the deep learning fusion identification model for preliminary identification, and obtain initial identification results; Result verification component: Performs multi-directional subdivision migration verification based on the initial recognition result, generates multiple verification values, updates the initial recognition result according to the multiple verification values, and generates a multi-directional bird recognition dataset; Risk assessment component: Introduces regional operation logs of transmission lines, performs dynamic assessments based on the regional operation logs and the multi-directional bird recognition dataset, obtains bird damage risk levels, classifies and issues early warnings for transmission lines, and generates graded early warning instructions; Prevention and intervention component: Executes the graded early warning command to trigger the prevention and control device to carry out bird damage prevention and intervention on the transmission line.

10. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the bird damage prevention and early warning method for power transmission lines as described in any one of claims 1-8.

Images