Self-adaptive bird repelling method and system based on multi-source information fusion

Abstract

The invention discloses a self-adaptive bird repelling method and system based on multi-source information fusion, relates to the technical field of artificial intelligence and computer vision, and can be applied to the field of power systems and power transmission and distribution engineering. The main purpose is to solve the problem that an existing bird repelling method is low in effectiveness. The method mainly comprises the following steps: acquiring real-time environment information, and determining a plurality of monitoring execution objects according to the real-time environment information; calling real-time monitoring data of the monitoring execution object, and performing bird multi-dimensional identification according to the real-time monitoring data; determining a target bird repelling strategy according to the real-time environment information and the recognition result under the condition that the recognition result is a bird; and sending a bird repelling execution instruction to at least one type of target bird repelling assembly according to the target bird repelling strategy, so as to repel birds near the target power equipment based on the target bird repelling assembly. The method is mainly used for self-adaptive bird repelling in a multi-source information fusion mode.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and computer vision technology, and can be applied to the fields of power systems and power transmission and distribution engineering. In particular, it relates to an adaptive bird deterrence method and system based on multi-source information fusion. Background Technology

[0002] In the field of power system operation and maintenance, bird control refers to a series of protective measures taken to prevent bird activity from endangering the safety of power equipment. Birds, especially large birds, tend to regard power facilities such as transmission towers and substation structures as ideal roosting or nesting sites. However, their activities can cause "bird-line conflicts": bird droppings can reduce the insulation strength of insulator air gaps, leading to "bird droppings flashover" tripping; nesting materials may scatter in wind and rain, short-circuiting live parts; and large birds, during flight or fighting, may directly bridge conductors with different potentials, causing "bird short circuits." These bird-induced faults seriously threaten the stable operation of the power grid; therefore, taking effective bird control measures is a necessary step to ensure the reliability of power supply.

[0003] Currently, existing bird control methods can be mainly divided into physical barriers and frightening methods. Common measures include installing bird spikes and bird barriers to prevent birds from landing, or deploying sound, light, and electronic bird control systems to frighten them. However, these methods have significant limitations: static physical barriers (such as bird spikes) are prone to creating blind spots, allowing birds to gradually adapt and find footholds; single deterrent methods (such as fixed-frequency sound waves) are easily ineffective due to bird habituation and lack specificity, and their effectiveness is greatly reduced in inclement weather (such as rain, fog, and night). Existing technologies generally suffer from the problems of "passive defense, single methods, and lack of intelligence," failing to accurately respond to bird species, behavioral intentions, and environmental changes, resulting in unsustainable and incomplete bird control effects and low bird control effectiveness. Summary of the Invention

[0004] In view of this, the present invention provides an adaptive bird deterrence method and system based on multi-source information fusion, the main purpose of which is to solve the problem of low effectiveness of existing bird deterrence methods.

[0005] According to one aspect of the present invention, an adaptive bird deterrence method based on multi-source information fusion is provided, comprising: Obtain real-time environmental information and determine multiple monitoring execution objects based on the real-time environmental information; Retrieve real-time monitoring data of the monitored object and perform multi-dimensional bird identification based on the real-time monitoring data; If the identification result is a bird, a target bird deterrence strategy is determined based on the real-time environmental information and the identification result; According to the target bird deterrence strategy, bird deterrence execution instructions are sent to at least one type of target bird deterrence component to drive away birds near the target power equipment based on the target bird deterrence component.

[0006] Furthermore, the real-time environmental information includes normal visibility, no light, low visibility, and high humidity low visibility; The determination of multiple monitoring execution objects based on the real-time environmental information includes: When the atmospheric transmission state is normal visibility, the first optical imaging component is used as the main monitoring target, and the first sensor and the second sensor are used as auxiliary monitoring targets. When the atmospheric transmission state is a dark state, the second optical imaging component is used as the main monitoring target, and the first sensor, the third sensor and the second sensor are used as auxiliary monitoring targets. When the atmospheric transmission condition is low visibility, the second optical imaging component is used as the primary monitoring target, and the first sensor is used as the auxiliary monitoring target. When the atmospheric transmission conditions are high humidity and low visibility, the first sensor is used as the primary monitoring target, and the second and third sensors are used as auxiliary monitoring targets. The first optical imaging component is used to capture images in the visible light band, the second optical imaging component is used to capture infrared radiation information, the first sensor is used to detect moving targets at a distance, the second sensor is used to detect physical contact at close range, and the third sensor is used to acquire audio signals.

[0007] Furthermore, the step of retrieving real-time monitoring data of the monitoring target and performing multi-dimensional bird identification based on the real-time monitoring data includes: The system acquires first real-time monitoring data of the auxiliary monitoring execution object in real time. If the first real-time monitoring data meets a preset risk threshold, a start command is generated and sent to the main monitoring execution object so that the main monitoring execution object can collect an image of the target monitoring location. The second real-time monitoring data of the main monitoring execution object is obtained, and multi-dimensional bird identification is performed based on the first real-time monitoring data and the second real-time monitoring data. The multi-dimensional bird identification includes at least one of body posture recognition, behavior recognition, movement trajectory recognition and sound recognition.

[0008] Furthermore, the multi-dimensional identification of birds based on the first real-time monitoring data and the second real-time monitoring data includes: When the first optical imaging component is the primary monitoring target and the first and second sensors are the secondary monitoring targets, the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, and the bird visible light image collected by the first optical imaging component are input into the pre-trained first multi-source prediction model to generate body type, behavior category, and movement trajectory through the first multi-source prediction model. With the second optical imaging component as the primary monitoring target and the first, third, and second sensors as auxiliary monitoring targets, the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, the bird sound signal collected by the third sensor, and the bird infrared thermal radiation image collected by the second optical imaging component are input into the pre-trained second multi-source prediction model to generate body type, behavior category, and movement trajectory through the second multi-source prediction model. When the second optical imaging component is the primary monitoring target and the first sensor is the secondary monitoring target, the bird infrared thermal radiation image acquired by the second optical imaging component and the long-distance bird movement trajectory acquired by the first sensor are input into the second multi-source prediction model to generate body type, behavior category and movement trajectory through the second multi-source prediction model. With the first sensor serving as the primary monitoring target and the second and third sensors serving as auxiliary monitoring targets, the long-distance bird movement trajectory, the bird landing vibration data, and the bird sound signal are input into a pre-trained third multi-source prediction model to generate body type, behavior category, and movement trajectory through the third multi-source prediction model.

[0009] Furthermore, the real-time environmental information includes normal visibility, no light, low visibility, and high humidity low visibility, and the recognition results include body type and behavior category; The step of determining the target bird-repelling strategy based on the real-time environmental information and recognition results includes: Based on the real-time environmental information and behavior category, the target execution component is matched from the execution component sub-library of the bird deterrence strategy database; Based on body type and behavior category, the target operation parameters of each target execution component are matched from the execution parameter sub-library of the bird deterrence strategy database; The target bird deterrence strategy is determined based on the target execution component and the target operating parameters.

[0010] Further, the step of matching the target execution component from the execution component sub-library of the bird deterrence strategy database based on the real-time environmental information and behavior category includes: When the real-time environmental information is in a dark state and the behavior category is not transit, the strong light flashing light and ultrasonic bird repeller are used as target execution components. In situations where real-time environmental information is low visibility and the behavior category is non-transit, ultrasonic bird repellers and sonic bird repellers are used as target execution components. When the real-time environmental information is high humidity and low visibility, and the behavior category is roosting, acoustic bird deterrents and mechanical bird deterrents are used as target execution components. When the real-time environmental information is normal visibility and the behavior category is nest building, the acoustic bird deterrent and the mechanical bird deterrent are used as target execution components.

[0011] Furthermore, the matching of target operating parameters for each target execution component from the execution parameter sub-library of the bird deterrence strategy database based on body type and behavior category includes: When the target execution components include a high-intensity strobe light and / or an ultrasonic bird repeller, if the body type is large and the behavior category is perching, the maximum light intensity and the highest frequency are used as the target configuration parameters; if the body type is medium or small and the behavior category is flocking or nesting, the medium brightness and low frequency are used as the target configuration parameters. When the target execution component includes a sonic bird repeller, if the body type is large and the behavior category is roosting, then a strong impact sound and maximum volume are used as the target configuration parameters; if the body type is medium or small and the behavior category is flocking or nesting, then a medium volume and a sonic frequency matching the current bird species are used as the target configuration parameters. When the target execution component includes a mechanical bird deterrent, if the behavior category is nesting or roosting, then medium-frequency continuous or high-frequency intermittent is used as the target configuration parameter; if the behavior category is landing, then high-frequency and short-duration are used as the target configuration parameters.

[0012] Furthermore, after the bird deterrent component drives away birds near the target electrical equipment, the method further includes: Analysis of the expulsion effect based on real-time data collected from the monitored targets; If the bird deterrence effect is effective, the real-time monitoring data, identification results and target bird deterrence strategy are sent to the cloud platform so that the real-time monitoring data, identification results and target bird deterrence strategy can be updated to the positive sample database based on the cloud platform. If the bird deterrence effect is ineffective, the current real-time monitoring data, current identification results, and current target bird deterrence strategy are sent to the cloud platform so that the bird deterrence strategy database can be updated through the cloud platform based on the positive sample library and the current real-time monitoring data, current identification results, and current target bird deterrence strategy.

[0013] Furthermore, the method also includes: When the geographical location of the target power equipment is within the migration corridor and the system time enters a preset window period before the migration period, the scanning frequency of the first sensor is increased and the confidence level of bird identification in the identification results is reduced. The system sends early warning signals to adjacent poles downstream of the migration route via the communication network, so that the adjacent poles can adjust their bird deterrence strategies to the migration season mode in advance.

[0014] According to another aspect of the present invention, an adaptive bird deterrence system based on multi-source information fusion is provided, comprising: a power supply module, a perception layer, a decision layer, and an execution layer, wherein the perception layer, the decision layer, and the execution layer communicate with each other through a communication bus; The power supply module includes a battery and a solar panel. The sensing layer includes a first optical imaging component for capturing visible light band images, a second optical imaging component for capturing infrared radiation information, a first sensor for detecting moving targets at a distance, a second sensor for detecting physical contact at close range, and a third sensor for acquiring audio signals. The decision-making layer includes a central processing unit and a cloud platform; The execution layer includes a high-intensity flashing light, an ultrasonic bird repeller, a sonic bird repeller, and a mechanical bird repeller; The decision layer is used to perform operations corresponding to the adaptive bird-repelling method based on multi-source information fusion described above.

[0015] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages: This invention provides an adaptive bird deterrence method and system based on multi-source information fusion. In this embodiment, real-time environmental information is acquired, and multiple monitoring targets are identified based on this information. Real-time monitoring data of the monitored targets is retrieved, and multi-dimensional bird identification is performed based on this data. If the identification result is birds, a target bird deterrence strategy is determined based on the real-time environmental information and the identification result. Bird deterrence execution instructions are sent to at least one type of target bird deterrence component according to the target bird deterrence strategy to drive away birds near the target power equipment. Through a multi-modal collaborative bird deterrence mechanism, the adaptability of birds to a single deterrence method is overcome, significantly reducing the probability of bird-related failures. Secondly, a hierarchical decision-making mechanism is adopted to achieve precise control of bird deterrence intensity, optimizing system energy consumption while ensuring protective effectiveness. Thirdly, a closed-loop learning mechanism continuously improves the strategy library, ensuring the long-term effectiveness of the protective effect. The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the specification. Furthermore, in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 The flowchart of an adaptive bird-repelling method based on multi-source information fusion provided by an embodiment of the present invention is shown. Figure 2 This invention provides a flowchart illustrating the logic for generating bird deterrence execution commands based on three monitoring channels, according to an embodiment of the present invention. Figure 3(a) shows a front view of a schematic diagram of the installation method of a system provided in an embodiment of the present invention on a single-loop straight tower; Figure 3(b) shows a top view of a schematic diagram of the installation method of a system provided in an embodiment of the present invention on a single-loop straight tower; Figure 4 The diagram shows a front view illustrating the installation method of a system in a dual-loop tension tower according to an embodiment of the present invention. Figure 5 The diagram shows a block diagram of an adaptive bird deterrent system based on multi-source information fusion provided by an embodiment of the present invention. Detailed Implementation

[0017] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0018] To address the issue of low effectiveness of existing bird control methods, this invention provides an adaptive bird control method based on multi-source information fusion. This method is applied to an adaptive bird control system based on multi-source information fusion, which includes a power supply module, a perception layer, a decision layer, and an execution layer. The perception layer, decision layer, and execution layer communicate with each other via a communication bus. The decision layer serves as the execution entity of the adaptive bird control method based on multi-source information fusion. Figure 1 As shown, the method includes: 101. Obtain real-time environmental information and determine multiple monitoring execution objects based on the real-time environmental information.

[0019] In this embodiment of the invention, the sensing layer integrates multiple sensors (such as light sensors, temperature and humidity sensors, radar, vibration sensors, sound sensors, infrared thermal imaging cameras, and visible light cameras). The light and temperature / humidity sensors actively acquire the current environmental state (such as no light, low visibility, normal visibility, etc.). Based on this environmental information, the most suitable monitoring target, i.e., the sensor combination, is selected and activated. For example, at night (no light), the monitoring target is automatically switched to the thermal imaging camera and radar, while the visible light camera is disabled. By adaptively switching the monitoring target according to the real-time environmental state, all sensors are prevented from operating at full power 24 / 7, significantly reducing the overall system power consumption. This is crucial for field equipment that relies on solar energy and batteries, extending its continuous operating time.

[0020] 102. Retrieve the real-time monitoring data of the monitored object and perform multi-dimensional bird identification based on the real-time monitoring data.

[0021] In this embodiment of the invention, the real-time monitoring data is multi-source heterogeneous data, requiring spatiotemporal alignment and feature extraction of the raw data streams output from the multi-source sensors. Then, a pre-trained multi-source recognition model is used to fuse and predict features from radar point cloud trajectories, optical image features, acoustic signature spectra, and vibration signals. This model is a multi-input multi-output model, capable of outputting multi-dimensional prediction results including biological category (bird / non-bird), body type, behavior category, and motion trajectory vector. Multi-sensor fusion effectively overcomes the limitations and blind spots of a single sensor, cross-validating each other and greatly reducing the risk of false alarms (such as misidentifying a falling plastic bag as a bird) and missed alarms.

[0022] 103. If the identification result is birds, determine the target bird-repelling strategy based on the real-time environmental information and the identification result.

[0023] In this embodiment of the invention, environmental state parameters and bird characteristic parameters are combined for retrieval, and a corresponding equipment scheduling scheme, i.e., a target bird deterrence strategy, is matched from a predefined bird deterrence strategy library. The bird deterrence strategy library is built based on historical operation and maintenance data. The bird deterrence strategies in the library are composite strategies, clearly defining which equipment to use and with what parameters. A unique and optimal target bird deterrence strategy is dynamically generated. For example, in a rainy night environment, for a large bird landing, a strong sonic bird deterrent is activated, and a specific predator's call is played at maximum volume. Through the matching mechanism of composite strategies, precise and differentiated handling is achieved, ensuring that the intensity of the deterrence action is perfectly matched to the risk level. This allows for rapid and effective responses under high threats, as well as energy-saving and eco-friendly methods under low threats, avoiding over-determination or ineffective deterrence. Furthermore, since the bird deterrence strategy is generated in real-time based on dynamic environment and behavior, it has higher randomness, effectively preventing birds from adapting to fixed deterrence patterns and ensuring long-term deterrence effectiveness.

[0024] 104. Send bird deterrence execution instructions to at least one type of target bird deterrence component according to the target bird deterrence strategy, so as to drive away birds near the target power equipment based on the target bird deterrence component.

[0025] In this embodiment of the invention, according to the target bird-repelling component defined by the target bird-repelling strategy, its activation / deactivation sequence, and operational parameter configuration, bird-repelling execution commands are sent to the corresponding acoustic, optical, and mechanical bird-repelling components via an industrial communication protocol. The bird-repelling execution commands include device address codes, action type codes, and intensity parameter values, ensuring that the multimodal bird-repelling components can operate collaboratively according to the predetermined scheme. Through multimodal (sound, light, physical) coordinated stimulation, birds can be influenced from multiple sensory channels, significantly improving the success rate of bird repelling.

[0026] In one embodiment of the present invention, for further explanation and limitation, the step of determining multiple monitoring execution objects based on the real-time environmental information includes: When the atmospheric transmission state is normal visibility, the first optical imaging component is used as the main monitoring target, and the first sensor and the second sensor are used as auxiliary monitoring targets. When the atmospheric transmission state is a dark state, the second optical imaging component is used as the main monitoring target, and the first sensor, the third sensor and the second sensor are used as auxiliary monitoring targets. When the atmospheric transmission condition is low visibility, the second optical imaging component is used as the primary monitoring target, and the first sensor is used as the auxiliary monitoring target. When the atmospheric transmission conditions are high humidity and low visibility, the first sensor is used as the primary monitoring target, and the second and third sensors are used as auxiliary monitoring targets.

[0027] The real-time environmental information includes normal visibility, no light, low visibility, and high humidity low visibility. The first optical imaging component is used to capture images in the visible light band, the second optical imaging component is used to capture infrared radiation information, the first sensor is used for long-range moving target detection, the second sensor is used for short-range physical contact detection, and the third sensor is used for audio signal acquisition.

[0028] In this embodiment of the invention, integrated environmental sensing units (such as light sensors and meteorological sensors) quantify and analyze on-site conditions, classifying complex weather phenomena into four distinct atmospheric transmission states: normal visibility (corresponding to sunny days), no light (corresponding to darkness and night), low visibility (corresponding to haze or foggy weather), and high humidity and low visibility (corresponding to rainy days). Based on this state determination, the system dynamically schedules different sensor combinations to serve as primary and secondary monitoring targets. Under normal visibility conditions, the system prioritizes the information-rich visible light camera (first optical imaging component) for accurate identification, with radar (first sensor) providing remote early warning and vibration sensor (second sensor) acting as the last line of defense. When entering a dark state, where visible light fails, the system immediately switches to a thermal imaging camera (second optical imaging component) unaffected by light, and integrates radar trajectory data, audio information from a sound sensor (third sensor), and contact signals from a vibration sensor for comprehensive judgment. In low visibility conditions (such as fog or haze), the system continues to rely primarily on the penetrating thermal imaging, with all-weather radar providing auxiliary detection. Under the highest-risk conditions of high humidity and low visibility (heavy rain), the system prioritizes the most reliable radar as the main sensor, responsible for overall monitoring and early warning, while simultaneously coordinating with vibration and sound sensors for close-range confirmation to address insulation flashover faults that are highly likely to occur at this time.

[0029] It should be noted that by adaptively switching between the primary and secondary sensor combinations for different atmospheric transmission conditions, this system ensures that an optimal sensing system is always in control under any lighting and weather conditions. This effectively overcomes the functional limitations of a single sensor in specific environments, achieving reliable 24-hour uninterrupted monitoring. Furthermore, by allocating tasks as needed based on environmental conditions and clearly defining the primary and secondary sensors, high-power components (such as optical imaging equipment) are only used as the primary sensor when their performance is at its highest. This significantly reduces the overall energy consumption of the system while ensuring monitoring performance.

[0030] In one embodiment of the present invention, for further explanation and limitation, the step of retrieving real-time monitoring data of the monitoring execution object and performing multi-dimensional bird identification based on the real-time monitoring data includes: The system acquires first real-time monitoring data of the auxiliary monitoring execution object in real time. If the first real-time monitoring data meets a preset risk threshold, a start command is generated and sent to the main monitoring execution object so that the main monitoring execution object can collect an image of the target monitoring location. The second real-time monitoring data of the main monitoring execution object is obtained, and multi-dimensional bird identification is performed based on the first real-time monitoring data and the second real-time monitoring data.

[0031] In this embodiment of the invention, multi-dimensional bird identification includes at least one of body posture recognition, behavior recognition, movement trajectory recognition, and sound recognition. A hierarchical triggering and collaborative verification mechanism optimizes the identification process and energy efficiency. Specifically, auxiliary monitoring targets (such as radar, vibration sensors, and sound sensors) with low continuous power consumption and high targeting but relatively simple information dimensions are continuously operated, and their first real-time monitoring data (such as radar point cloud trajectories, vibration signal characteristics, and specific acoustic signatures) is acquired in real time. The system sets preset risk thresholds for these data (e.g., radar detects a target entering a 500-meter range, vibration sensors detect impact signals of a specific frequency, or sound sensors identify bird warning calls). Only when any one or more combinations of data from these auxiliary sensors meet the threshold, indicating a potential risk, will the system generate and send a start command to the main monitoring target (such as a visible light or thermal imaging camera) with higher power consumption and richer information dimensions. The start command includes the target monitoring location. After the main sensor is awakened, it adjusts the shooting angle according to the target monitoring location and collects second real-time monitoring data. The system simultaneously acquires its high-quality second real-time monitoring data (such as high-definition visible light images or thermal imaging images) and fuses the data from the main and auxiliary sensors. Ultimately, based on this fused data, multi-dimensional bird identification is performed: using the main sensor data to perform body shape recognition (distinguishing between large / medium / small birds) and behavior recognition (determining whether they are landing, roosting, or nesting) through AI visual algorithms, combining radar data to complete movement trajectory recognition (calculating speed and direction), and using sound sensor data to assist in sound recognition (identifying species or stress state).

[0032] As an application example, different low-power sensors can perform parallel sensing, such as... Figure 2The diagram shows the logic flow for generating bird deterrence commands based on three monitoring channels. The three channels operate simultaneously: Radar Monitoring Channel: As the primary long-range early warning system, the radar sensor continuously scans the airspace. Upon detecting a suspicious moving target, it doesn't directly trigger bird deterrence; instead, it guides optical instruments, such as visible light or thermal imaging, to precisely scan and identify the target airspace. Only after the optical system confirms the target as a bird does the system generate a bird deterrence command. Vibration Monitoring Channel: As crucial for close-range contact detection, a vibration (or infrared beam) sensor continuously monitors the physical contact of key parts of the tower (such as crossarms). When a suspicious vibration signal is detected, it also guides optical instruments to scan and confirm the vibration source. This design effectively compensates for the radar's blind spot for landed birds and generates a command after confirmation by the optical system. Sound Monitoring Channel: As an auxiliary identification method, a sound sensor continuously collects environmental audio. When it analyzes sound signatures matching bird characteristics (such as calls or flapping wings), it also guides optical instruments to scan the sound source area, providing important assistance, especially in dense vegetation or when visibility is obstructed, and generates a command after optical confirmation. By operating three monitoring channels in parallel—radar, vibration, and sound—complementary detection blind spots can be achieved, improving identification accuracy.

[0033] It should be noted that by using a low-power auxiliary sensor as a "sentinel" for initial screening and triggering, the system avoids the ineffective energy consumption caused by the main sensor running idle for a long time. At the same time, the fusion of sensor data ensures the comprehensiveness and accuracy of the identification results.

[0034] In one embodiment of the present invention, for further explanation and limitation, the step of performing multi-dimensional bird identification based on the first real-time monitoring data and the second real-time monitoring data includes: When the first optical imaging component is the primary monitoring target and the first and second sensors are the secondary monitoring targets, the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, and the bird visible light image collected by the first optical imaging component are input into the pre-trained first multi-source prediction model to generate body type, behavior category, and movement trajectory through the first multi-source prediction model. With the second optical imaging component as the primary monitoring target and the first, third, and second sensors as auxiliary monitoring targets, the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, the bird sound signal collected by the third sensor, and the bird infrared thermal radiation image collected by the second optical imaging component are input into the pre-trained second multi-source prediction model to generate body type, behavior category, and movement trajectory through the second multi-source prediction model. When the second optical imaging component is the primary monitoring target and the first sensor is the secondary monitoring target, the bird infrared thermal radiation image acquired by the second optical imaging component and the long-distance bird movement trajectory acquired by the first sensor are input into the second multi-source prediction model to generate body type, behavior category and movement trajectory through the second multi-source prediction model. With the first sensor serving as the primary monitoring target and the second and third sensors serving as auxiliary monitoring targets, the long-distance bird movement trajectory, the bird landing vibration data, and the bird sound signal are input into the pre-trained third multi-source prediction model to generate body type, behavior category, and movement trajectory through the third multi-source prediction model.

[0035] In this embodiment of the invention, under normal visibility conditions, the system operates in an ideal perception environment. High-detail visible light images (from the first optical imaging component) are the most reliable basis for AI to identify body type (e.g., distinguishing between black storks and birds) and behavioral categories (e.g., determining nesting and roosting). Radar motion trajectories (first sensor) are used for early warning and trajectory prediction. When the radar detects a distant target, the model can activate the optical system in advance to prepare for identification and align and fuse the radar trajectory with the optical identification results on a timeline, generating a more coherent and predictive motion trajectory. Vibration data (second sensor) provides crucial behavioral confirmation. When the optical system identifies a landing intention but with low confidence, the landing impact signal collected by the vibration sensor at the corresponding time point can provide decisive physical evidence for the behavioral judgment, greatly reducing misjudgment.

[0036] In the absence of light, the system loses visible light information, and the second multi-source prediction model relies on thermal imaging and deeply fuses other data for compensation. Thermal images (from the second optical imaging component) become the primary source of information for identification. The model utilizes tuned algorithms to focus on analyzing the morphological features (body type) of the heat source's outline and its motion patterns (behavioral category, such as wing flapping or stationary). Radar data continues to provide long-range trajectories and is associated with the thermally imaged target. Sound signals (from the third sensor) play a prominent role in this mode. Through voiceprint recognition tuning, the model can associate specific calls with targets in the thermal image, aiding in species identification (e.g., confirming the heat source is a calling bird rather than a silent bat or drone). Vibration data continues to serve as the final verification method for close-range encounters and key behaviors (such as landing). This multi-source cross-validation effectively overcomes the inherent limitations of thermal imaging in terms of insufficient detail in body type and species identification.

[0037] In low-visibility conditions such as fog and light rain, visible light becomes ineffective, and sound and vibration are easily affected by environmental interference. In these situations, the system employs a more robust strategy: a simplified application of the second multi-source prediction model. This model primarily relies on thermal imaging and radar, two of the best-performing sensors in all weather conditions. Thermal imaging provides weather-independent target thermal radiation information for basic target identification and morphological analysis. Radar provides precise motion trajectories. In this mode, the system may temporarily reduce the requirements for fine-grained species identification and subtle behavioral judgments, prioritizing the reliable detection and tracking of the core event of birds approaching.

[0038] During the rainy season and flood season, in conditions of high humidity and low visibility, optical methods become largely ineffective. The system then employs a third multi-source prediction model, an emergency identification scheme based on non-optical sensors. Radar (the first sensor) becomes the primary sensor, providing the target's trajectory and approximate size for preliminary assessment of its size (large / small). Vibration data (the second sensor) becomes crucial. Once the radar detects an approaching target, the model pays close attention to vibration signals. The triggering of vibration sensors is directly interpreted by the model as the highest-risk landing behavior category. Sound signals (the third sensor) serve as an auxiliary measure, employing enhanced noise reduction algorithms and specific voiceprint filtering to attempt to capture bird flapping or calls amidst wind and rain noise, providing supplementary identification information for radar detection. In this mode, the model's output may have lower confidence, but because it corresponds to the highest risk of failure (bird droppings flashover in high humidity), its warning threshold is adjusted accordingly to ensure safety.

[0039] It should be noted that the first, second, and third multi-source prediction models mentioned above all employ a deep learning-based multimodal fusion network architecture. Their core idea is to achieve unified representation and joint inference of heterogeneous sensor data through a shared feature encoding layer, multiple dedicated feature extraction branches, and a multi-task prediction head. The multi-path feature extraction branches are feature extraction sub-networks built for each type of input data (such as optical images, radar point cloud trajectories, vibration waveforms, and audio spectra). For example, convolutional neural networks are used to process image and spectral data, while recurrent neural networks or temporal convolutional networks are used to process trajectory and waveform data, extracting high-dimensional feature vectors from the raw data. The feature fusion module aligns and concatenates feature vectors from different branches, and performs cross-modal feature interaction and fusion through one or more fully connected layers, forming a unified joint feature representation containing multi-source information. Based on the fused joint features, the multi-task prediction head connects multiple task-specific output layers in parallel to simultaneously generate recognition results for body type, behavior category, and motion trajectory. The above models construct a large-scale, high-quality multi-source sample library based on historical operational data. Each sample contains raw multi-sensor data collected synchronously at a specific timestamp, along with accurate body type, behavior category, and motion trajectory labels annotated by experts or verified through multiple validations.

[0040] In one embodiment of the present invention, for further explanation and limitation, the step of determining the target bird-repelling strategy based on the real-time environmental information and the recognition result includes: Based on the real-time environmental information and behavior category, the target execution component is matched from the execution component sub-library of the bird deterrence strategy database; Based on body type and behavior category, the target operation parameters of each target execution component are matched from the execution parameter sub-library of the bird deterrence strategy database; The target bird deterrence strategy is determined based on the target execution component and the target operating parameters.

[0041] In this embodiment of the invention, real-time environmental information includes normal visibility, no light, low visibility, and high humidity with low visibility. The identification results include body type and behavior category. The system uses the current real-time environmental information (e.g., no light) and the identified behavior category (e.g., landing) as joint query conditions, inputting them into an execution component sub-library. This sub-library is essentially a decision matrix that defines which type(s) of physical equipment should be prioritized when responding to different bird behaviors under different environmental backgrounds. For example, the query conditions (no light, landing) might match the target execution components as a high-intensity flashing light and a sonic bird deterrent. This process ensures that the selected tools are best suited to the current environmental constraints and behavioral threats.

[0042] After determining which devices will be used as the target execution components, the system needs to configure operating parameters for each component. At this point, the system uses the identified body type (e.g., large birds) and behavior category (e.g., landing) as another set of query conditions, inputting them into the execution parameter sub-library. This sub-library has pre-set detailed parameter tables for each device, targeting different targets and behaviors. For example, for the condition (large birds, landing), the parameters matched for a high-intensity flashing light are maximum brightness and maximum flashing frequency, while the parameters matched for a sonic bird deterrent are maximum volume and predator alarm sound. This ensures that the intensity of the deterrence is precisely matched to the risk level of the target. Finally, the system binds the list of target execution components matched in the first step with the target operating parameters corresponding to each component matched in the second step, assembling a complete target bird deterrence strategy that can be immediately deployed to hardware for execution.

[0043] It should be noted that the complex strategy formulation process is broken down into two relatively independent query steps: equipment selection and parameter configuration. This approach is logically clear, computationally inefficient, and has a fast response time, meeting the requirements of real-time control. Furthermore, by combining environmental information, behavioral categories, and body types for decision-making, the final bird deterrence strategy can simultaneously consider external conditions, bird intentions, and the birds' own threats, effectively avoiding both insufficient and excessive deterrence.

[0044] In one embodiment of the present invention, for further explanation and limitation, the step of matching the target execution component from the execution component sub-library of the bird deterrence strategy database based on the real-time environmental information and behavior category includes: When the real-time environmental information is in a dark state and the behavior category is not transit, the strong light flashing light and ultrasonic bird repeller are used as target execution components. In situations where real-time environmental information is low visibility and the behavior category is non-transit, ultrasonic bird repellers and sonic bird repellers are used as target execution components. When the real-time environmental information is high humidity and low visibility, and the behavior category is roosting, acoustic bird deterrents and mechanical bird deterrents are used as target execution components. When the real-time environmental information is normal visibility and the behavior category is nest building, the acoustic bird deterrent and the mechanical bird deterrent are used as target execution components.

[0045] In this embodiment of the invention, the core logic for matching target execution components from the execution component sub-library, i.e., the specific construction logic of the execution component sub-library, is as follows: In the absence of light, to address non-transiting behavior (bird activity that poses a direct or potential security threat to power equipment): When the environment is dark, visible light deterrence is ineffective. The system prioritizes using a high-intensity flashing light, leveraging its strong impact on the bird's visual system in the dark to achieve effective deterrence. Simultaneously, it coordinates with an ultrasonic bird repeller, using sound as a combined deterrent to ensure effective deterrence. In low visibility conditions, to address non-transiting behavior: Under conditions such as fog and rain, the penetration and effective distance of optical means (including strong light and lasers) are limited. In this case, the system turns to acoustic means, which are less affected by weather, simultaneously activating both ultrasonic and sonic bird repellers. This utilizes sound waves of different frequency bands to cover a wider range of bird-sensitive spectrums, and the collaborative work of the two devices compensates for the limitations of a single acoustic means. Addressing roosting behavior in high humidity and low visibility conditions: This combination indicates the highest risk level (high humidity easily triggers flashovers, and birds have already established a stable presence), requiring the most direct, reliable, and intensive deterrence. Therefore, a combination of acoustic bird deterrents (playing high-intensity alarm sounds) and mechanical bird deterrents (generating physical vibrations) is chosen. Through the dual mode of "strong sound + vibration"—both acoustic deterrence and direct physical interference—the aim is to dismantle the birds' roosting status in the shortest possible time. Addressing nesting behavior in normal visibility conditions: Nesting is a continuous high-risk behavior requiring sustained intervention. The system selects a combination of acoustic bird deterrents (providing continuous auditory interference) and mechanical bird deterrents (directly disrupting the physical stability of nests), aiming to fundamentally destroy the nesting environment and achieve precise, targeted removal.

[0046] In one embodiment of the present invention, for further explanation and limitation, the step of matching the target operating parameters of each target execution component from the execution parameter sub-library of the bird deterrence strategy database based on body type and behavior category includes: When the target execution components include a high-intensity strobe light and / or an ultrasonic bird repeller, if the body type is large and the behavior category is perching, the maximum light intensity and the highest frequency are used as the target configuration parameters; if the body type is medium or small and the behavior category is flocking or nesting, the medium brightness and low frequency are used as the target configuration parameters. When the target execution component includes a sonic bird repeller, if the body type is large and the behavior category is roosting, then a strong impact sound and maximum volume are used as the target configuration parameters; if the body type is medium or small and the behavior category is flocking or nesting, then a medium volume and a sonic frequency matching the current bird species are used as the target configuration parameters. When the target execution component includes a mechanical bird deterrent, if the behavior category is nesting or roosting, then medium-frequency continuous or high-frequency intermittent is used as the target configuration parameter; if the behavior category is landing, then high-frequency and short-duration are used as the target configuration parameters.

[0047] In this embodiment of the invention, the core logic of matching target execution parameters from the parameter component sub-library based on body type and behavior category, that is, the specific construction logic of the parameter component sub-library, is as follows: For high-intensity light / ultrasonic devices, targeting high-risk targets (large bird roosts): The droppings of large birds are the main risk factor for bird droppings flashes, and their large size makes them difficult to disperse. Therefore, the system is configured with the highest light intensity and highest frequency ultrasound to quickly resolve the issue through the strongest visual and auditory stimulation. For low- to medium-risk targets (small and medium-sized bird flocks / nesting): Small and medium-sized birds pose a relatively lower risk, but their flocking activities or nesting behaviors still require intervention. In this case, medium brightness and low frequency parameters are used, achieving effective interference and dispersal while avoiding excessive fright and energy waste, demonstrating eco-friendliness. The low-frequency mode is also less likely for birds to adapt quickly.

[0048] For sonic bird deterrents, when dealing with high-risk targets: use the highest intensity, employing strong impact sounds (such as predator alarms or pops) and maximum volume to create a powerful acoustic impact. When dealing with low- to medium-risk targets: use a more targeted medium volume and play specific deterrent sound frequencies matched to the current bird species. Based on the different sensitivities of different birds to specific frequencies, this approach can significantly reduce noise pollution and interference with non-target birds while ensuring effectiveness.

[0049] For mechanical bird deterrents, addressing continuous behaviors (nesting / perching): This requires configuring a medium-frequency continuous or high-frequency intermittent vibration mode. The aim is to disrupt the stability of nesting and the comfort of perching through uninterrupted or regular physical disturbance, preventing the bird from completing its work or settling comfortably. Addressing transient behaviors (landing): Configuring high-frequency and short-duration strong vibrations to provide a strong negative feedback the moment the bird's claws contact the pole, preventing it from perching at the root of the behavior.

[0050] In one embodiment of the invention, for further explanation and limitation, after the step of using the target bird deterrent component to drive away birds near the target electrical equipment, the method further includes: Analysis of the expulsion effect based on real-time data collected from the monitored targets; If the bird deterrence effect is effective, the real-time monitoring data, identification results and target bird deterrence strategy are sent to the cloud platform so that the real-time monitoring data, identification results and target bird deterrence strategy can be updated to the positive sample database based on the cloud platform. If the bird deterrence effect is ineffective, the current real-time monitoring data, current identification results, and current target bird deterrence strategy are sent to the cloud platform so that the bird deterrence strategy database can be updated through the cloud platform based on the positive sample library and the current real-time monitoring data, current identification results, and current target bird deterrence strategy.

[0051] In this embodiment of the invention, within a preset time period after the bird deterrence command is executed, the system again invokes the monitoring execution objects (such as radar, optical cameras, and vibration sensors) of the perception layer to collect subsequent behavioral data of the birds. Based on this data, the system automatically analyzes the data using preset quantitative indicators (such as whether the target has flown away from the protected area, whether it has not returned within a specified time, and whether the flock has dispersed), and determines whether the deterrence action is effective or ineffective. When the deterrence is effective, the system treats this event as a successful positive sample. The real-time monitoring data that triggered this task, the identification results, and the ultimately proven successful bird deterrence strategy are packaged and uploaded to the cloud platform. The cloud platform then updates this data to the positive sample database, thereby consolidating and strengthening the effective knowledge.

[0052] When bird deterrence is ineffective (e.g., birds are not deterred or return within a short time), the current real-time monitoring data, identification results, and the proven ineffective bird deterrence strategy need to be marked as invalid cases and uploaded. Upon receiving invalid cases, the cloud platform initiates an optimization algorithm: by comparing the current strategy with a proven effective strategy used under similar monitoring data and identification results in the positive sample library, it identifies the shortcomings of the current strategy. Based on this analysis, the cloud platform can update the bird deterrence strategy database. For example, it can modify the execution component sub-library to add or replace more effective execution components under specific conditions. It can also adjust the execution parameter sub-library to optimize device operating parameters in specific scenarios (e.g., increasing volume or changing sound wave type). Furthermore, it can generate new optimized strategies for specific difficult-to-deter birds or behaviors.

[0053] In one embodiment of the present invention, for further explanation and limitation, the method further includes: When the geographical location of the target power equipment is within the migration corridor and the system time enters a preset window period before the migration period, the scanning frequency of the first sensor is increased and the confidence level of bird identification in the identification results is reduced. The system sends early warning signals to adjacent poles downstream of the migration route via the communication network, so that the adjacent poles can adjust their bird deterrence strategies to the migration season mode in advance.

[0054] In this embodiment of the invention, the geographic information of the target power equipment is continuously compared with the geographic information database of bird migration routes pre-stored in the cloud platform, while the system time is also compared with the bird migration period calendar. When it is determined that the equipment is located within the migration route and the current time is the start of the migration season, the system triggers the migration season alert mode. After the mode is activated, the system adjusts its local perception strategy in advance to improve capture and response efficiency. Specifically, by increasing the scanning frequency of the first sensor (radar), the airspace is scanned earlier and more frequently, thereby enabling more timely detection of approaching flocks of birds from a distance, providing early warning time for the entire joint defense network. The confidence threshold for bird identification in the identification results is lowered, i.e., a sensitive strategy is adopted when identifying birds. Some distant and vague targets that might normally be filtered out due to low confidence levels will be included in the monitoring and early warning range in advance under this mode, greatly reducing the risk of missed detection. When the local radar first identifies a large flock of birds and confirms its direction of movement, the decision-making level of the tower will send an early warning signal to adjacent towers downstream of the migration path through a communication network (such as 4G / 5G or a dedicated power grid). This signal contains at least information such as the size, speed, direction, and estimated arrival time of the flock. Upon receiving this warning, downstream poles can adjust their bird control strategies to migratory patterns in advance, thereby achieving global optimization of protection effectiveness.

[0055] In specific application examples, it can be done as shown in Figure 3- Figure 4 The diagram illustrates the installation methods of the system on a single-loop straight-line tower and on a double-loop tension tower, showcasing the installation of sensors, bird deterrents, radar, and image acquisition components in an adaptive bird deterrent system based on multi-source information fusion. In the diagram, 1 represents a typical mechanical bird deterrent—a bird spike; 2 represents a radar-video integrated unit, i.e., an integrated component combining radar and visible light image acquisition; 3 represents any one or more sensors, including sound sensors, infrared sensors, and vibration sensors; and 4 represents the bird-pulling line. The top view of the system's installation on the double-loop tension tower can be referenced from the top view of the system's installation on the single-loop straight-line tower.

[0056] This invention provides an adaptive bird deterrence method based on multi-source information fusion. In this embodiment, real-time environmental information is acquired, and multiple monitoring targets are identified based on this information. Real-time monitoring data of the monitored targets is retrieved, and multi-dimensional bird identification is performed based on this data. If the identification result is birds, a target bird deterrence strategy is determined based on the real-time environmental information and the identification result. Bird deterrence execution instructions are sent to at least one type of target bird deterrence component according to the target bird deterrence strategy to drive away birds near the target power equipment. Through a multi-modal collaborative bird deterrence mechanism, the adaptability of birds to a single deterrence method is overcome, significantly reducing the probability of bird-related failures. Secondly, a hierarchical decision-making mechanism is adopted to achieve precise control of bird deterrence intensity, optimizing system energy consumption while ensuring protective effectiveness. Thirdly, a closed-loop learning mechanism continuously improves the strategy library, ensuring the long-term effectiveness of the protective effect. Furthermore, as a response to the above Figure 1 The implementation of the method shown in this invention provides an adaptive bird deterrent system based on multi-source information fusion, such as... Figure 5 As shown, the system includes: a power supply module 210, a perception layer 220, a decision layer 230, and an execution layer 240. The perception layer 220, the decision layer 230, and the execution layer 240 communicate with each other through a communication bus. The power supply module 210 includes a battery and a solar panel, and the sensing layer 220 includes a first optical imaging component for capturing visible light band images, a second optical imaging component for capturing infrared radiation information, a first sensor for detecting moving targets at a distance, a second sensor for detecting physical contact at close range, and a third sensor for acquiring audio signals. The decision-making layer 230 includes a central processing unit and a cloud platform; The execution layer 240 includes a high-intensity flashing light, an ultrasonic bird repeller, a sonic bird repeller, and a mechanical bird repeller. The central processing unit is used to acquire real-time environmental information and determine multiple monitoring execution objects based on the real-time environmental information; retrieve real-time monitoring data of the monitoring execution objects and perform multi-dimensional bird identification based on the real-time monitoring data; if the identification result is birds, determine a target bird repelling strategy based on the real-time environmental information and the identification result; and send bird repelling execution instructions to at least one type of target bird repelling component based on the target bird repelling component to drive away birds near the target power equipment.

[0057] Furthermore, the real-time environmental information includes normal visibility, no light, low visibility, and high humidity low visibility. The central processing unit is also configured to: when the atmospheric transmission state is normal visibility, use the first optical imaging component as the primary monitoring object and the first and second sensors as auxiliary monitoring objects; when the atmospheric transmission state is no light, use the second optical imaging component as the primary monitoring object and the first, third, and second sensors as auxiliary monitoring objects; when the atmospheric transmission state is low visibility, use the second optical imaging component as the primary monitoring object and the first sensor as an auxiliary monitoring object; when the atmospheric transmission state is high humidity low visibility, use the first sensor as the primary monitoring object and the second and third sensors as auxiliary monitoring objects. The first optical imaging component is used to capture visible light band images, the second optical imaging component is used to capture infrared radiation information, the first sensor is used for long-range moving target detection, the second sensor is used for short-range physical contact detection, and the third sensor is used for audio signal acquisition.

[0058] Furthermore, the central processing unit is also used to acquire first real-time monitoring data of the auxiliary monitoring execution object in real time; when the first real-time monitoring data meets a preset risk threshold, generate and send a start command to the main monitoring execution object so that the main monitoring execution object can collect an image of the target monitoring location; acquire second real-time monitoring data of the main monitoring execution object; and perform multi-dimensional bird identification based on the first real-time monitoring data and the second real-time monitoring data, wherein the multi-dimensional bird identification includes at least one of body posture recognition, behavior recognition, movement trajectory recognition and sound recognition.

[0059] Furthermore, the central processing unit is also configured to, when the first optical imaging component is the primary monitoring target and the first sensor and the second sensor are auxiliary monitoring targets, input the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, and the visible light image of the bird collected by the first optical imaging component into a pre-trained first multi-source prediction model, so as to generate body type, behavior category, and movement trajectory through the first multi-source prediction model; and when the second optical imaging component is the primary monitoring target and the first sensor, the third sensor, and the second sensor are auxiliary monitoring targets, input the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, the bird sound signal collected by the third sensor, and the bird infrared thermal radiation image collected by the second optical imaging component into a pre-trained first multi-source prediction model. The trained second multi-source prediction model generates body type, behavior category, and movement trajectory. With the second optical imaging component as the primary monitoring object and the first sensor as the auxiliary monitoring object, the bird infrared thermal radiation image acquired by the second optical imaging component and the long-distance bird movement trajectory acquired by the first sensor are input into the second multi-source prediction model to generate body type, behavior category, and movement trajectory. With the first sensor as the primary monitoring object and the second and third sensors as auxiliary monitoring objects, the long-distance bird movement trajectory, the bird landing vibration data, and the bird sound signal are input into the pre-trained third multi-source prediction model to generate body type, behavior category, and movement trajectory.

[0060] Furthermore, the central processing unit is also configured to match target execution components from the execution component sub-library of the bird deterrence strategy database based on the real-time environmental information and behavior category; match target operating parameters of each target execution component from the execution parameter sub-library of the bird deterrence strategy database based on body type and behavior category; and determine the target bird deterrence strategy based on the target execution component and the target operating parameters.

[0061] Furthermore, the central processing unit is also configured to use a high-intensity flashing light and an ultrasonic bird repeller as target execution components when the real-time environmental information is in a dark state and the behavior category is non-transit; to use an ultrasonic bird repeller and a sonic bird repeller as target execution components when the real-time environmental information is low visibility and the behavior category is non-transit; to use a sonic bird repeller and a mechanical bird repeller as target execution components when the real-time environmental information is high humidity and low visibility and the behavior category is perching; and to use a sonic bird repeller and a mechanical bird repeller as target execution components when the real-time environmental information is normal visibility and the behavior category is nesting.

[0062] Furthermore, the central processing unit is also configured to, when the target execution component includes a high-intensity flashing light and / or an ultrasonic bird repeller, use the maximum light intensity and highest frequency as target configuration parameters if the bird's body type is large and the behavior category is roosting, and use medium brightness and low frequency as target configuration parameters if the bird's body type is small to medium and the behavior category is flocking or nesting; when the target execution component includes an acoustic bird repeller, use a strong impact sound and maximum volume as target configuration parameters if the bird's body type is large and the behavior category is roosting, and use medium volume and sound frequency matching the current bird species as target configuration parameters if the bird's body type is small to medium and the behavior category is flocking or nesting; when the target execution component includes a mechanical bird repeller, use continuous medium frequency or intermittent high frequency as target configuration parameters if the behavior category is nesting or roosting, and use high frequency and short duration as target configuration parameters if the behavior category is landing.

[0063] Furthermore, the central processing unit is also used to analyze the bird deterrence effect based on real-time data collected from the monitored target; if the bird deterrence effect is effective, the real-time monitoring data, identification results, and target bird deterrence strategy are sent to the cloud platform so that the real-time monitoring data, identification results, and target bird deterrence strategy are updated to the positive sample database based on the cloud platform; if the bird deterrence effect is ineffective, the current real-time monitoring data, current identification results, and current target bird deterrence strategy are sent to the cloud platform. The cloud platform is used to update the bird deterrence strategy database based on the positive sample library, current real-time monitoring data, current identification results, and current target bird deterrence strategy.

[0064] Furthermore, the central processing unit is also configured to, when the geographical location of the target power equipment is within the migration corridor and the system time enters a preset window period before the migration period, increase the scanning frequency of the first sensor and reduce the confidence level of bird identification in the identification results; and send a warning signal to adjacent poles downstream of the migration path through the communication network so that the adjacent poles can adjust their bird deterrence strategy to the migration period mode in advance.

[0065] This invention provides an adaptive bird deterrence system based on multi-source information fusion. In this embodiment, real-time environmental information is acquired, and multiple monitoring targets are identified based on this information. Real-time monitoring data of the monitored targets is retrieved, and multi-dimensional bird identification is performed based on this data. If the identification result is birds, a target bird deterrence strategy is determined based on the real-time environmental information and the identification result. Bird deterrence execution instructions are sent to at least one type of target bird deterrence component according to the target bird deterrence strategy to drive away birds near the target power equipment. Through a multi-modal collaborative bird deterrence mechanism, the system overcomes the adaptability of birds to a single deterrence method, significantly reducing the probability of bird-related failures. Secondly, a hierarchical decision-making mechanism enables precise control of bird deterrence intensity, optimizing system energy consumption while ensuring protective effectiveness. Thirdly, a closed-loop learning mechanism continuously improves the strategy library, ensuring the long-term effectiveness of the protective effect. It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive bird-repelling method based on multi-source information fusion, characterized in that, include: Obtain real-time environmental information and determine multiple monitoring execution objects based on the real-time environmental information; Retrieve real-time monitoring data of the monitored object and perform multi-dimensional bird identification based on the real-time monitoring data; If the identification result is a bird, a target bird deterrence strategy is determined based on the real-time environmental information and the identification result; According to the target bird deterrence strategy, bird deterrence execution instructions are sent to at least one type of target bird deterrence component to drive away birds near the target power equipment based on the target bird deterrence component.

2. The method according to claim 1, characterized in that, The real-time environmental information includes normal visibility, no light, low visibility, and high humidity with low visibility. The determination of multiple monitoring execution objects based on the real-time environmental information includes: When the atmospheric transmission state is normal visibility, the first optical imaging component is used as the main monitoring target, and the first sensor and the second sensor are used as auxiliary monitoring targets. When the atmospheric transmission state is a dark state, the second optical imaging component is used as the main monitoring target, and the first sensor, the third sensor and the second sensor are used as auxiliary monitoring targets. When the atmospheric transmission condition is low visibility, the second optical imaging component is used as the primary monitoring target, and the first sensor is used as the auxiliary monitoring target. When the atmospheric transmission conditions are high humidity and low visibility, the first sensor is used as the primary monitoring target, and the second and third sensors are used as auxiliary monitoring targets. The first optical imaging component is used to capture images in the visible light band, the second optical imaging component is used to capture infrared radiation information, the first sensor is used to detect moving targets at a distance, the second sensor is used to detect physical contact at close range, and the third sensor is used to acquire audio signals.

3. The method according to claim 2, characterized in that, The process of retrieving real-time monitoring data of the monitored object and performing multi-dimensional bird identification based on the real-time monitoring data includes: The system acquires first real-time monitoring data of the auxiliary monitoring execution object in real time. If the first real-time monitoring data meets a preset risk threshold, a start command is generated and sent to the main monitoring execution object so that the main monitoring execution object can collect an image of the target monitoring location. The second real-time monitoring data of the main monitoring execution object is obtained, and multi-dimensional bird identification is performed based on the first real-time monitoring data and the second real-time monitoring data. The multi-dimensional bird identification includes at least one of body posture recognition, behavior recognition, movement trajectory recognition and sound recognition.

4. The method according to claim 3, characterized in that, The multi-dimensional identification of birds based on the first real-time monitoring data and the second real-time monitoring data includes: When the first optical imaging component is the primary monitoring target and the first and second sensors are the secondary monitoring targets, the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, and the bird visible light image collected by the first optical imaging component are input into the pre-trained first multi-source prediction model to generate body type, behavior category, and movement trajectory through the first multi-source prediction model. With the second optical imaging component as the primary monitoring target and the first, third, and second sensors as auxiliary monitoring targets, the long-distance bird movement trajectory collected by the first sensor, the bird landing vibration data collected by the second sensor, the bird sound signal collected by the third sensor, and the bird infrared thermal radiation image collected by the second optical imaging component are input into the pre-trained second multi-source prediction model to generate body type, behavior category, and movement trajectory through the second multi-source prediction model. When the second optical imaging component is the primary monitoring target and the first sensor is the secondary monitoring target, the bird infrared thermal radiation image acquired by the second optical imaging component and the long-distance bird movement trajectory acquired by the first sensor are input into the second multi-source prediction model to generate body type, behavior category and movement trajectory through the second multi-source prediction model. With the first sensor serving as the primary monitoring target and the second and third sensors serving as auxiliary monitoring targets, the long-distance bird movement trajectory, the bird landing vibration data, and the bird sound signal are input into a pre-trained third multi-source prediction model to generate body type, behavior category, and movement trajectory through the third multi-source prediction model.

5. The method according to claim 1, characterized in that, The real-time environmental information includes normal visibility, no light, low visibility, and high humidity low visibility; the recognition results include body type and behavior category. The step of determining the target bird-repelling strategy based on the real-time environmental information and recognition results includes: Based on the real-time environmental information and behavior category, the target execution component is matched from the execution component sub-library of the bird deterrence strategy database; Based on body type and behavior category, the target operation parameters of each target execution component are matched from the execution parameter sub-library of the bird deterrence strategy database; The target bird deterrence strategy is determined based on the target execution component and the target operating parameters.

6. The method according to claim 5, characterized in that, The process of matching the target execution component from the execution component sub-library of the bird deterrence strategy database based on the real-time environmental information and behavior category includes: When the real-time environmental information is in a dark state and the behavior category is not transit, the strong light flashing light and ultrasonic bird repeller are used as target execution components. In situations where real-time environmental information is low visibility and the behavior category is non-transit, ultrasonic bird repellers and sonic bird repellers are used as target execution components. When the real-time environmental information is high humidity and low visibility, and the behavior category is roosting, acoustic bird deterrents and mechanical bird deterrents are used as target execution components. When the real-time environmental information is normal visibility and the behavior category is nest building, the acoustic bird deterrent and the mechanical bird deterrent are used as target execution components.

7. The method according to claim 5, characterized in that, The target operating parameters for each target execution component are matched from the execution parameter sub-library of the bird deterrence strategy database based on body type and behavior category, including: When the target execution components include a high-intensity strobe light and / or an ultrasonic bird repeller, if the body type is large and the behavior category is perching, the maximum light intensity and the highest frequency are used as the target configuration parameters; if the body type is medium or small and the behavior category is flocking or nesting, the medium brightness and low frequency are used as the target configuration parameters. When the target execution component includes a sonic bird repeller, if the body type is large and the behavior category is roosting, then a strong impact sound and maximum volume are used as the target configuration parameters; if the body type is medium or small and the behavior category is flocking or nesting, then a medium volume and a sonic frequency matching the current bird species are used as the target configuration parameters. When the target execution component includes a mechanical bird deterrent, if the behavior category is nesting or roosting, then medium-frequency continuous or high-frequency intermittent is used as the target configuration parameter; if the behavior category is landing, then high-frequency and short-duration are used as the target configuration parameters.

8. The method according to claim 5, characterized in that, After the bird deterrent component based on the target device drives away birds near the target power equipment, the method further includes: Analysis of the expulsion effect based on real-time data collected from the monitored targets; If the bird deterrence effect is effective, the real-time monitoring data, identification results and target bird deterrence strategy are sent to the cloud platform so that the real-time monitoring data, identification results and target bird deterrence strategy can be updated to the positive sample database based on the cloud platform. If the bird deterrence effect is ineffective, the current real-time monitoring data, current identification results, and current target bird deterrence strategy are sent to the cloud platform so that the bird deterrence strategy database can be updated through the cloud platform based on the positive sample library and the current real-time monitoring data, current identification results, and current target bird deterrence strategy.

9. The method according to claim 1, characterized in that, The method further includes: When the geographical location of the target power equipment is within the migration corridor and the system time enters a preset window period before the migration period, the scanning frequency of the first sensor is increased and the confidence level of bird identification in the identification results is reduced. The system sends early warning signals to adjacent poles downstream of the migration route via the communication network, so that the adjacent poles can adjust their bird deterrence strategies to the migration season mode in advance.

10. An adaptive bird deterrent system based on multi-source information fusion, characterized in that, It includes a power supply module, a sensing layer, a decision-making layer, and an execution layer, and the sensing layer, decision-making layer, and execution layer communicate with each other through a communication bus; The power supply module includes a battery and a solar panel. The sensing layer includes a first optical imaging component for capturing visible light band images, a second optical imaging component for capturing infrared radiation information, a first sensor for detecting moving targets at a distance, a second sensor for detecting physical contact at close range, and a third sensor for acquiring audio signals. The decision-making layer includes a central processing unit and a cloud platform; The execution layer includes a high-intensity flashing light, an ultrasonic bird repeller, a sonic bird repeller, and a mechanical bird repeller; The decision layer is used to perform the operations corresponding to the adaptive bird-repelling method based on multi-source information fusion as described in any one of claims 1-7.

Images