Acousto-optic fusion expelling system and method based on bird habits

Abstract

The invention relates to an acousto-optic fusion repelling system and method based on bird habits. The system comprises a bird behavior recognition module, an environment sensing module, a bird repelling strategy generation module, an acousto-optic control module and a self-adaptive learning optimization module. The method comprises the following steps: acquiring bird images, sound and environmental parameters in real time; bird types are identified through ResNet + MFCC feature extraction, and a behavior state is analyzed in combination with a CNN-LSTM model; generating an acousto-optic fusion stimulation sequence according to species characteristics, behavior states and environmental parameters; and performing expelling according to a multi-modal sequence and the like. According to the method for generating the acousto-optic bird repelling sequence in combination with the bird behavior characteristics and the environment state, the bird repelling efficiency is improved and the occurrence frequency of bird damage is reduced through a feedback-based adaptive learning optimization algorithm and a dynamic acousto-optic bird repelling strategy matching and execution mechanism; the problems of high adaptability and high maintenance cost of a traditional bird repelling mode are solved; the intelligent, dynamic and sustainable bird repelling strategy is realized.

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary field of artificial intelligence, biobehavior and environmental control, and specifically relates to an acoustic and optical fusion repelling system and method based on bird habits. Background Art

[0002] In areas such as power lines, farmland orchards, and airport runways, bird gatherings often lead to equipment damage, crop damage, or flight safety hazards. Currently, mainstream bird repellent technologies include ultrasound, lasers, bird repellents, and manual intervention, but most methods have the following defects and shortcomings: single stimulation methods do not provide long-lasting bird repellent effects; birds are prone to adaptability, and repeated stimulation is ineffective; the system cannot dynamically adjust its strategy based on bird species, time period, and weather; and there is a lack of intelligent algorithm support, resulting in delayed responses and serious waste of resources. The main reasons for the above defects are:

[0003] 1. Disconnect between identification and removal strategies

[0004] Species-generalization processing: Existing systems generally only detect the presence of birds without distinguishing between species. This results in repelling strategies that do not match the birds' actual sensitivity characteristics. For example, using the same repelling pattern for owls, which are sensitive to sound waves, and sparrows, which are sensitive to light, reduces efficiency by over 40%.

[0005] Static knowledge base: Traditional methods rely on a fixed parameter library and cannot adapt to regional differences in bird habits, such as changes in migratory population behavior.

[0006] 2. Lack of closed-loop optimization mechanism

[0007] One-way execution architecture: Existing systems generally lack feedback channels for repelling birds, making it impossible to adjust strategies based on real-time bird reactions. For example, if a target bird approaches the device instead of leaving, the system will continue with its original strategy until a preset timeout expires.

[0008] High dependence on manual labor: Parameter adjustment requires on-site operation by professionals. For example, a substation's bird repellent system requires manual calibration three times a month, increasing operation and maintenance costs by 200%. Summary of the Invention

[0009] The technical problem to be solved by the present invention is: In order to overcome the above technical problems, the present invention provides a sound and light fusion expulsion system and method based on bird habits. Based on the principles of bird behavior, the bird repellent strategy combines acoustic and optical stimulation mechanisms. By sensing the activity patterns, reaction thresholds and avoidance characteristics of different birds, a sound and light combined stimulation sequence is dynamically generated to achieve intelligent expulsion of birds; real-time strategy optimization can be performed according to environmental changes and bird reactions to improve the bird repellent effect and reduce bird adaptability.

[0010] The technical solution adopted by the present invention to solve the technical problem is: an acoustic and optical fusion repellent system based on bird habits, including a bird behavior recognition module, an environmental perception module, a bird repellent strategy generation module, an acoustic and optical control module and an adaptive learning optimization module, wherein:

[0011] Bird Behavior Recognition Module: This module collects data, identifies bird species and locations using YOLOv5, uses ResNet (residual network) and MFCC (Mel-Frequency Cepstral Coefficient) feature extraction to identify bird sounds, and uses a CNN-LSTM model to determine bird behavior.

[0012] Environmental perception module: including temperature and humidity sensors, light sensors, and wind speed sensors, for real-time monitoring of environmental conditions;

[0013] Bird repellent strategy generation module: This module is used to match the optimal sound and light combination from the knowledge base based on the bird behavior and environmental data collected by the bird behavior recognition module and the environmental perception module, such as the combination of high-frequency sound waves and intermittent strong light;

[0014] Sound and light control module: used to implement bird-repelling strategies, adjust sound wave frequency, light intensity and light flashing frequency;

[0015] Adaptive learning optimization module: This module uses bird responses, such as dwell time and flight trajectory, to provide feedback on optimization strategy weights and update the knowledge base using reinforcement learning. The reinforcement learning algorithm is a Q-learning algorithm whose state space includes bird species, numbers, behavioral states, environmental parameters, and current device status. The action space is a combination of sound and light parameter adjustments, and the reward function is designed based on the success rate of repelling and energy consumption.

[0016] The bird behavior recognition module collects data through a camera and a microphone.

[0017] The knowledge base includes a bird sensitivity database and an environmental adaptation strategy database, wherein:

[0018] Bird sensitivity database: records the sound wave frequency sensitivity range, light source wavelength sensitivity range and flickering pattern;

[0019] Environmental adaptation strategy library: stores the sound and light parameter correction rules under different meteorological conditions.

[0020] Suitable for intelligent bird repellent in power lines, agricultural production areas or airport runways.

[0021] A sound and light fusion driving away method based on bird habits, comprising the following steps:

[0022] Step 1: Real-time collection of bird images, sounds and environmental parameters;

[0023] Step 2: Use YOLOv5 to identify bird species and locations, combine ResNet+MFCC feature extraction to identify bird sounds, and use the CNN-LSTM model to determine the bird's behavior status;

[0024] Step 3: Generate a sound and light fusion stimulation sequence based on species characteristics, behavioral status, and environmental parameters:

[0025] Differentiation of species characteristics: calling the bird sensitivity database to match the optimal sound and light combination;

[0026] Behavioral state adaptation: distinguish bird behaviors and dynamically adjust strategies;

[0027] Environmental adaptation: Optimize adjustments and switch between day and night modes according to weather conditions;

[0028] Step 4: Execute the expulsion according to the multimodal sequence: using sound wave emission → strong light flashing → silent period alternation, and randomize the superimposed parameters;

[0029] Step 5: Combine the DeepSORT algorithm with a Kalman filter to predict bird movement trajectories. Track the birds' departure trajectories and calculate the success rate of repelling them using YOLOv5. (A successful repelling occurs when a bird leaves and does not return within 30 seconds.) Analyze the stress response using the call spectrum. If the alarm sounds increase, the strategy is considered effective.

[0030] Step 6: Use the Q-learning algorithm to adjust the policy parameters in real time and update the knowledge base through the transfer learning cycle. Real-time data annotation: Record the bird species, environmental parameters, policy combinations and results of each expulsion to build an annotated dataset.

[0031] In step 3, behavioral state adaptation enables high-frequency sound waves to disrupt the animal's auditory positioning for foraging behavior, and enables strong light flashes to disrupt the animal's visual comfort for roosting behavior;

[0032] Environmental adaptation is to reduce the sound wave frequency to 1-5kHz on rainy days to reduce air attenuation and increase the light source intensity by 20% to compensate for water mist scattering; increase the sound wave transmission power in strong winds to overcome noise interference, and use a stable light source to prevent flickering from being obscured by wind-driven vegetation; during the day, light stimulation is mainly used, such as 500 lux strobe, supplemented by sound waves; at night, sound waves are mainly used, such as intermittent 8kHz pulses, to avoid strong light pollution.

[0033] The trigger condition control logic in step 3 includes intelligent threshold determination, group size graded response, and immediate response to dangerous behavior; among them, the intelligent threshold determination: the dynamic stay time threshold is dynamically adjusted according to the bird's behavior;

[0034] Group size graded response: single stimulation is enabled for 1-5 individuals, and sound and light alternation + random interference is enabled for more than 5 individuals;

[0035] Instant response to dangerous behavior: Activate the highest intensity strategy when a bird is detected entering a critical area.

[0036] The priority of the sound and light fusion strategy in step 3 is:

[0037] First priority: species identity and behavioral state combination;

[0038] Second priority: environmental adaptation and correction strategies;

[0039] Third priority: If it fails twice in a row, it will switch to the backup combination of ultrasonic + laser scanning.

[0040] The optimization mechanism in step 6 includes:

[0041] Short-term optimization: A Q-learning algorithm is used to adjust policy parameters based on immediate feedback, such as sound wave intensity ±10% and light frequency ±2Hz. For example, if an expulsion fails, the duration of the sound wave will be increased by 20% in the next similar scenario.

[0042] Long-term optimization: Aggregate regional data to train bird sensitivity classification models, and update the knowledge base based on migration seasons, such as adding strategies to drive away migratory birds in autumn.

[0043] The implementation of the Q-learning algorithm in step 6 includes:

[0044] State space definition: including the following variables: bird species, number, behavioral state, light intensity, weather and current sound and light parameters;

[0045] Action space definition: including adjustments to the frequency, intensity, and duration of sound waves, and the wavelength, flickering frequency, and brightness of light sources;

[0046] Reward function design: The reward value R is based on real-time feedback on bird repellent effectiveness. Successful repellent (birds leave and don't return within 30 seconds) is rewarded with R = +10, partial repellent (birds briefly leave and then return) is rewarded with R = +5, invalid repellent (birds don't respond or get too close to the device) is rewarded with R = -3, and energy penalty (excessive intensity or prolonged operation) is rewarded with R = -1. The goal is to maximize cumulative rewards while balancing bird repellent effectiveness with energy consumption.

[0047] The beneficial effects of the present invention are that the sound and light fusion bird repellent system and method based on bird habits are suitable for various scenarios such as agricultural bird prevention, power line protection, shipping airport safety, and ecological conservation. The method generates a sound and light bird repellent sequence by combining bird behavior characteristics and environmental conditions. The adaptive learning optimization algorithm based on feedback and the dynamic sound and light bird repellent strategy matching and execution mechanism are used to improve the efficiency of bird repellent and reduce the frequency of bird damage; overcome the problems of strong adaptability and high maintenance costs of traditional bird repellent methods; realize the intelligence, dynamism and sustainability of bird repellent strategies; the present invention combines bird behavior with intelligent algorithms to construct a sound and light coordinated and learnable and evolutionary bird repellent system, so the present invention has irreplaceable advantages in accuracy, response speed and strategy flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be further described below with reference to the accompanying drawings and examples.

[0049] Figure 1 This is the overall architecture diagram of the sound and light fusion repelling system based on bird habits of the present invention.

[0050] Figure 2 It is a flowchart of the algorithm execution of the sound and light fusion driving away system and method based on bird habits of the present invention. DETAILED DESCRIPTION

[0051] The present invention will now be described in further detail with reference to the accompanying drawings. It should be emphasized that the following description is merely exemplary and is not intended to limit the scope of the present invention and its application.

[0052] like Figure 1 As shown, the present invention provides an acoustic and optical fusion repelling system based on bird habits, including a bird behavior recognition module, an environment perception module, a bird repelling strategy generation module, an acoustic and optical control module, and an adaptive learning optimization module, wherein:

[0053] Bird Behavior Recognition Module: This module collects data, identifies bird species and locations using YOLOv5, uses ResNet (residual network) and MFCC (Mel-Frequency Cepstral Coefficient) feature extraction to identify bird sounds, and uses a CNN-LSTM model to determine bird behavior, such as foraging and roosting.

[0054] Environmental perception module: including temperature and humidity sensors, light sensors, and wind speed sensors, for real-time monitoring of environmental conditions;

[0055] The bird repellent strategy generation module is used to match the optimal sound and light combination, such as high-frequency sound waves + intermittent strong light, from the knowledge base based on the bird behavior and environmental data collected by the bird behavior recognition module and the environmental perception module. The knowledge base is constructed based on multimodal data fusion and domain knowledge integration, and the main data sources include:

[0056]

[0057] Sound and light control module: used to implement bird-repelling strategies, adjusting the sound wave frequency 0.5-20kHz, light intensity 100-1000lux and light flashing frequency 1-5Hz;

[0058] Adaptive learning optimization module: This module uses bird responses, such as dwell time and flight trajectory, to provide feedback on optimization strategy weights and update the knowledge base using reinforcement learning. The reinforcement learning algorithm is a Q-learning algorithm whose state space includes bird species, numbers, behavioral states, environmental parameters, and current device status. The action space is a combination of sound and light parameter adjustments, and the reward function is designed based on the success rate of repelling and energy consumption.

[0059] The bird behavior recognition module collects data through a camera and a microphone.

[0060] The knowledge base includes a bird sensitivity database and an environmental adaptation strategy database, wherein:

[0061] Bird Sensitivity Database: Records sensitivity ranges for sound frequencies, light wavelengths, and flicker patterns, covering parameters such as sound frequency (0.5-20kHz), light wavelength (400-700nm), and flicker pattern (continuous / intermittent). For example, owls are sensitive to low-frequency sound waves (1-3kHz), while sparrows react more strongly to high-frequency flickering light (500-600nm, 10-15Hz).

[0062] Environmental adaptation strategy library: stores the sound and light parameter correction rules under different meteorological conditions.

[0063] Suitable for intelligent bird repellent in power lines, agricultural production areas or airport runways.

[0064] The specific implementation of YOLOv5 in the system of the present invention is:

[0065] (1) Model training and optimization:

[0066] Dataset construction:

[0067] 100,000 annotated images were collected, covering different bird species (such as sparrows, seagulls, and owls), lighting conditions (day and night / rain and fog), and postures (flying / roosting).

[0068] Annotation format: COCO format, including bounding box (bbox) and category label.

[0069] Model improvements:

[0070] Attention mechanism: Embed the CBAM module in Backbone to enhance the ability to extract bird features.

[0071] Small object optimization: Using a higher resolution input of 1280×1280 improves the small object detection accuracy AP@0.5 by 15%.

[0072] Training parameters:

[0073] Pre-trained model: lightweight YOLOv5s.

[0074] Iterations: 300 epochs, batch size = 32.

[0075] Data enhancement: Mosaic enhancement, random cropping, and color jittering.

[0076] (2) Real-time detection process:

[0077] 1. Image input: The camera captures images at 30 fps with a resolution of 1920 × 1080.

[0078] 2. Forward reasoning:

[0079] The input image is scaled to 640×640 and predicted by the YOLOv5 model.

[0080] Output format: [x min ,y min ,x max ,y max ,confidence,class id ].

[0081] 3. Post-processing:

[0082] Non-maximum suppression (NMS) is used, and the IoU threshold is 0.5 to filter overlapping boxes.

[0083] Category filtering: Only bird-related detection results are retained, such as confidence ≥ 0.6.

[0084] (3) Bird tracking and behavior analysis:

[0085] 1. Multi-Object Tracking (MOT):

[0086] Combined with the DeepSORT algorithm, the bird's motion trajectory is predicted through Kalman filtering.

[0087] Re-ID features (appearance + motion information) are used to associate the same object in consecutive frames.

[0088] 2. Behavioral judgment:

[0089] Dwell time: If the center coordinates of the bounding box of the same bird change by <5 pixels / s, it is considered to be perching.

[0090] Repelling effect: Track the number of frames in which birds leave the monitoring area and calculate the success rate.

[0091] (4) Performance indicators:

[0092] Accuracy: mAP@0.5 = 92.3% (Sparrow AP = 94.1%, Seagull AP = 89.7%) on the test set. Speed: Inference speed reaches 45FPS on NVIDIA Jetson Xavier NX.

[0093] Robustness: In rainy / foggy conditions, detection accuracy drops by <8%.

[0094] ResNet+MFCC feature extraction is used for bird sound recognition. The MFCC feature extraction process includes: Pre-emphasis: Apply a first-order high-pass filter to the audio signal, such as H(z) = 1-0.97z -1 , enhance high-frequency components.

[0095] Frame Windowing: Split the audio into 20-40ms frames (with 50% overlap) and apply a Hamming window to reduce spectral leakage.

[0096] Fourier transform: Calculate the power spectrum (FFT) of each frame.

[0097] Mel filter bank: maps the power spectrum through 20-40 triangular filters (Mel scale) to simulate the auditory characteristics of the human ear.

[0098] DCT transform: Take the logarithm of the Mel spectrum and perform discrete cosine transform, retaining the first 13-20 coefficients as MFCC features.

[0099] As a preferred option, the specific values ​​are as follows:

[0100] Sampling rate: 16kHz; frame length: 25ms (400 samples), frame shift: 10ms; MFCC dimensions: 13 (static) + 13 (first-order difference) + 13 (second-order difference) = 39 dimensions

[0101] ResNet is used to build an audio classification model and input processing:

[0102] Convert MFCC features into time-frequency maps (time × Mel-band) as two-dimensional input.

[0103] Normalization: Normalize each band to zero mean and unit variance.

[0104] Model structure:

[0105] Python

[0106] ResNet34(

[0107] stem:Conv2d(1,64,kernel=7,stride=2)#single channel input

[0108] ↓

[0109] ResBlock × 3 (channel number 64 → 128 → 256 → 512)

[0110] ↓

[0111] GlobalAvgPool+FC(512→Number of bird species) )

[0113] The CNN-LSTM model is used for bird behavior recognition, which includes the following steps:

[0114] (1) Input data preparation

[0115] Visual Input:

[0116] The video stream is divided into segments of 16 frames / segment (2 seconds, 30fps);

[0117] Each frame is scaled to 224×224, three-channel RGB.

[0118] Timing alignment:

[0119] Optical flow calculation (TV-L1 algorithm) supplements motion information.

[0120] (2) Model architecture:

[0121] Python

[0122] CNN branch (ResNet50 pre-training):

[0123] Input: (16,224,224,3)

[0124] Output: (16, 2048) # Features per frame

[0125] LSTM branch:

[0126] Input: (16, 2048)

[0127] Structure: Bi-LSTM (units = 512) + Attention

[0128] Output: (512) # Clip-level features

[0129] Fusion layer:

[0130] Concatenate[CNN features, LSTM output] → FC(1024 → bird behavior category)

[0131] like Figure 2 As shown, the present invention provides a sound and light fusion driving away method based on bird habits, comprising the following steps:

[0132] Step 1: Real-time collection of bird images, sounds and environmental parameters;

[0133] Step 2: Use YOLOv5 to identify bird species and locations, combine ResNet+MFCC feature extraction to identify bird sounds, and use the CNN-LSTM model to determine the bird's behavior status;

[0134] Step 3: Generate a sound and light fusion stimulation sequence based on species characteristics, behavioral status, and environmental parameters:

[0135] Differentiation of species characteristics: calling the bird sensitivity database to match the optimal sound and light combination;

[0136] Behavioral state adaptation: distinguish bird behaviors and dynamically adjust strategies;

[0137] Environmental adaptation: Optimize adjustments and switch between day and night modes according to weather conditions;

[0138] Step 4: Execute the expulsion according to the multimodal sequence: using sound wave emission → strong light flashing → silent period alternation, and randomize the superimposed parameters;

[0139] Multimodal stimulation sequences:

[0140] Alternation of sound and light: sound wave (10 seconds) → strong light flash (5 seconds) → silent period (3 seconds, to prevent adaptation).

[0141] Randomization parameters: sound wave frequency ±2kHz, light source flash interval ±1s, to avoid bird habituation.

[0142] Step 5: Combine the DeepSORT algorithm with a Kalman filter to predict bird movement trajectories. Track the birds' departure trajectories and calculate the success rate of repelling them using YOLOv5. A successful repelling is considered if the birds leave and do not return within 30 seconds. The stress response is analyzed based on the call spectrum. If the alarm sounds increase, the strategy is considered effective.

[0143] Step 6: Use the Q-learning algorithm to adjust strategy parameters in real time, and update the knowledge base monthly through a transfer learning cycle. Real-time data annotation: Record the bird species, environmental parameters, strategy combinations, and results of each expulsion to build an annotated dataset.

[0144] In step 3, behavioral state adaptation enables high-frequency sound waves to disrupt the animal's auditory positioning for foraging behavior, and enables strong light flashes to disrupt the animal's visual comfort for roosting behavior;

[0145] Environmental adaptation is to reduce the sound wave frequency to 1-5kHz on rainy days to reduce air attenuation and increase the light source intensity by 20% to compensate for water mist scattering; increase the sound wave transmission power in strong winds to overcome noise interference, and use a stable light source to prevent flickering from being obscured by wind-driven vegetation; during the day, light stimulation is mainly used, such as 500 lux strobe, supplemented by sound waves; at night, sound waves are mainly used, such as intermittent 8kHz pulses, to avoid strong light pollution.

[0146] The trigger condition control logic in step 3 includes intelligent threshold determination, group size graded response, and immediate response to dangerous behavior. Among them, the intelligent threshold determination: the dynamic stay time threshold is dynamically adjusted according to the bird's behavior: 30 seconds for birds of prey and 15 seconds for small birds;

[0147] Group size graded response: single stimulation is enabled for 1-5 individuals, and sound and light alternation + random interference is enabled for more than 5 individuals;

[0148] Instant response to dangerous behavior: Activate the highest intensity strategy when a bird is detected entering a critical area, such as within 1 meter of power equipment.

[0149] The priority of the sound and light fusion strategy in step 3 is:

[0150] First priority: a combination of species characteristics and behavioral states, such as giving priority to blue light for herons;

[0151] Second priority: environmental adaptation and correction strategies, such as low-frequency sound waves in rainy days;

[0152] Third priority: If it fails twice in a row, it will switch to the backup combination of ultrasonic + laser scanning.

[0153] The optimization mechanism in step 6 includes:

[0154] Short-term optimization: A Q-learning algorithm is used to adjust policy parameters based on immediate feedback, such as sound wave intensity ±10% and light frequency ±2Hz. For example, if an expulsion fails, the duration of the sound wave will be increased by 20% in the next similar scenario.

[0155] Long-term optimization: Aggregate regional data to train bird sensitivity classification models, and update the knowledge base based on migration seasons, such as adding strategies to drive away migratory birds in autumn.

[0156] The implementation of the Q-learning algorithm in step 6 includes:

[0157] State Space Definition:

[0158] The state consists of the following variables:

[0159] Bird characteristics: species (such as sparrows, owls), number, behavioral status (foraging / roosting).

[0160] Environmental parameters: light intensity (lux), temperature and humidity, wind speed, weather (sunny / rainy / foggy).

[0161] Device status: current sound and light parameters (such as sound wave frequency 5kHz, light source flashing 10Hz).

[0162] Example state vector:

[0163] [Species = House Sparrow, Population = 3, Behavior = Foraging, Light = 500 lux, Weather = Sunny, Current Sound Wave = 8 kHz, Current Light Source = 12 Hz].

[0164] Action Space Definition:

[0165] Actions are adjustable combinations of sound and light parameters, including:

[0166] Sound wave adjustment: frequency ±2kHz, intensity ±10%, duration ±3 seconds.

[0167] Light source adjustment: wavelength ±50nm (e.g. 500nm→550nm), flicker frequency ±2Hz, brightness ±15%.

[0168] Example actions:

[0169] [Sound wave frequency +1kHz, light source flashing frequency -2Hz].

[0170] Reward Function Design

[0171] The reward value (R) is based on the real-time feedback of the bird-repelling effect:

[0172] Successful expulsion (bird leaves and does not return within 30 seconds): R = +10;

[0173] Partially effective (birds leave briefly and then return): R = +5;

[0174] Invalid (the bird does not respond or is close to the device): R = -3;

[0175] Energy consumption penalty (too high intensity or long running time): R = -1;

[0176] Goal: Maximize cumulative rewards and balance bird-repelling effect with energy consumption.

[0177] The core update formula of Q-learning is:

[0178] Q(st,at)←Q(st,at)+α[rt+1+γamaxQ(st+1,a)-Q(st,at)];

[0179] Among them: α is the learning rate: take 0.1, which controls the influence weight of new experience; γ is the discount factor: take 0.9, which measures the importance of future rewards.

[0180] Update example: If the action at (such as increasing the sound wave frequency) in the current state st leads to successful expulsion (rt+1=+10), the Q value is adjusted towards the future optimal action direction.

[0181] The core idea of ​​the Q-learning algorithm is to evaluate the long-term benefits of taking an action in a specific state through the Q value (action value function), and to approximate the optimal strategy by iteratively updating the Q value table.

[0182] In the present invention, the Q-learning algorithm is used to:

[0183] Short-term strategy optimization: real-time adjustment of sound and light parameters (such as frequency, intensity, duration).

[0184] Long-term knowledge base update: accumulate empirical data and optimize bird sensitivity models.

[0185] Example 1 (Power Line Scenario) Validity Verification Report

[0186] 1. Test environment and objectives

[0187] Scenario: 500kV high-voltage transmission line in a certain province (tower height 45 meters, insulator model FXBW4-110 / 100)

[0188] Problem: A flock of sparrows (approximately 50 birds) inhabits the insulators for extended periods of time, resulting in feces accumulation on the insulator surfaces and causing an average of three flashover failures per year.

[0189] Test objective: To verify the effectiveness of the sound and light fusion bird repellent system in real scenarios.

[0190] 2. System configuration and parameters

[0191]

[0192] 3. Test process and quantitative results

[0193] Phase 1: Baseline Testing (Traditional Ultrasonic Bird Repellent)

[0194] Methods: A commercially available ultrasonic bird repellent (fixed frequency 10 kHz continuous wave output, in line with industry standard GB / T12345-2010) was installed.

[0195] result:

[0196] Expulsion success rate: 68% on the first day, dropped to 22% on the fifth day (obvious adaptability).

[0197] Failure statistics: One guano flashover still occurred during the test.

[0198] Phase 2: Testing of the system of the present invention

[0199] Execution process:

[0200] Object detection: YOLOv5 identified 52 sparrows around the insulator (confidence ≥ 0.9, time taken 1.2 seconds).

[0201] Strategy matching:

[0202] Knowledge base query: Sparrows are sensitive to 12kHz+green frequency light (520nm, 15Hz).

[0203] Environmental adaptation: Wind speed 3m / s → Sound wave intensity increased by 8%.

[0204] Deportation Execution:

[0205] Alternation of sound and light: sound wave 12kHz (10 seconds) → green light 15Hz (5 seconds) → silent period (3 seconds).

[0206] Real-time tracking: DeepSORT showed that 46 sparrows flew away within 40 seconds (removal rate 88.5%).

[0207] Effect feedback:

[0208] Failure case: 6 sparrows briefly left and then returned → triggering the backup strategy (8kHz + UV light) → ultimately all were driven away.

[0209] Performance indicators:

[0210] index Test results Measurement method Bird identification accuracy 93.2% (52 / 56) Manual review of video frames Average removal time 42 seconds (group) Infrared camera trajectory tracking Strategy matching accuracy 91.7% (11 / 12 correct matches) Comparative bird behavior expert assessments Energy consumption 28W / time (including equipment standby) Electricity meter measurement

[0211] 4. Comparative test data

[0212]

[0213]

[0214] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. A sound and light fusion repelling system based on bird habits, characterized in that: It includes bird behavior recognition module, environment perception module, bird repelling strategy generation module, sound and light control module and adaptive learning optimization module, among which: Bird Behavior Recognition Module: This module collects data, identifies bird species and locations using YOLOv5, uses ResNet+MFCC feature extraction to identify bird sounds, and uses the CNN-LSTM model to determine bird behavior. Environmental perception module: including temperature and humidity sensors, light sensors, and wind speed sensors, for real-time monitoring of environmental conditions; Bird repellent strategy generation module: used to match the optimal sound and light combination from the knowledge base based on the bird behavior and environmental data collected by the bird behavior recognition module and the environmental perception module; Sound and light control module: used to implement bird-repelling strategies, adjust sound wave frequency, light intensity and light flashing frequency; Adaptive learning optimization module: This module optimizes strategy weights through bird reaction feedback and uses reinforcement learning to update the knowledge base. The reinforcement learning algorithm is a Q-learning algorithm whose state space includes bird species, numbers, behavioral states, environmental parameters, and current device status. The action space is a combination of sound and light parameter adjustments. The reward function is designed based on the success rate of repelling and energy consumption.

2. The sound and light fusion repelling system based on bird habits according to claim 1, characterized in that: The bird behavior recognition module collects data through a camera and a microphone.

3. The sound and light fusion repelling system based on bird habits according to claim 1, characterized in that: The knowledge base includes a bird sensitivity database and an environmental adaptation strategy database, wherein: Bird sensitivity database: records the sound wave frequency sensitivity range, light source wavelength sensitivity range and flickering pattern; Environmental adaptation strategy library: stores the sound and light parameter correction rules under different meteorological conditions.

4. The sound and light fusion repelling system based on bird habits according to any one of claims 1 to 3, characterized in that: Suitable for intelligent bird repellent in power lines, agricultural production areas or airport runways.

5. A sound and light fusion driving away method based on bird habits, characterized in that: The following steps are involved: Step 1: Real-time collection of bird images, sounds and environmental parameters; Step 2: Use YOLOv5 to identify bird species and locations, combine ResNet+MFCC feature extraction to identify bird sounds, and use the CNN-LSTM model to determine the bird's behavior status; Step 3: Generate a sound and light fusion stimulation sequence based on species characteristics, behavioral status, and environmental parameters: Differentiation of species characteristics: calling the bird sensitivity database to match the optimal sound and light combination; Behavioral state adaptation: distinguish bird behaviors and dynamically adjust strategies; Environmental adaptation: Optimize adjustments and switch between day and night modes according to weather conditions; Step 4: Execute the expulsion according to the multimodal sequence: using sound wave emission → strong light flashing → silent period alternation, and randomize the superimposed parameters; Step 5: Combine the DeepSORT algorithm and Kalman filter to predict the bird's movement trajectory, track the bird's departure trajectory, calculate the success rate of the repelling with YOLOv5, and analyze the stress response based on the call spectrum; Step 6: Use the Q-learning algorithm to adjust the policy parameters in real time and update the knowledge base through the transfer learning cycle.

6. The sound and light fusion driving away method based on bird habits according to claim 5, characterized in that: In step 3, behavioral state adaptation enables high-frequency sound wave interference for foraging behavior and strong light flashing for roosting behavior; Environmental adaptation is to reduce the sound wave frequency to 1-5kHz and increase the light source intensity by 20% on rainy days; increase the sound wave emission power in strong winds and use a stable light source; use light stimulation as the main method during the day, supplemented by sound waves; and use sound waves as the main method at night.

7. The sound and light fusion driving away method based on bird habits according to claim 5, characterized in that: The trigger condition control logic in step 3 includes intelligent threshold determination, group size graded response, and immediate response to dangerous behavior; among them, the intelligent threshold determination: the dynamic stay time threshold is dynamically adjusted according to the bird's behavior; Instant response to dangerous behavior: Activate the highest intensity strategy when a bird is detected entering a critical area.

8. The sound and light fusion driving away method based on bird habits according to claim 5, characterized in that: The priority of the sound and light fusion strategy in step 3 is: First priority: species identity and behavioral state combination; Second priority: environmental adaptation and correction strategies; Third priority: If it fails twice in a row, it will switch to the backup combination of ultrasonic + laser scanning.

9. The sound and light fusion driving away method based on bird habits according to claim 5, characterized in that: The optimization mechanism in step 6 includes: Short-term optimization: Use Q-learning algorithm to adjust strategy parameters based on immediate feedback; Long-term optimization: Aggregate regional data to train bird sensitivity classification models and update the knowledge base based on migration seasons.

10. The sound and light fusion driving away method based on bird habits according to claim 5, characterized in that: The implementation of the Q-learning algorithm in step 6 includes: State space definition: including the following variables: bird species, number, behavioral state, light intensity, weather and current sound and light parameters; Action space definition: including adjustments to the frequency, intensity, and duration of sound waves, and the wavelength, flickering frequency, and brightness of light sources; Reward function design: The reward value is based on real-time feedback of the bird-repelling effect.

Images