A method for crop acoustic sensing and directional acoustic control based on drone swarms

Abstract

This invention discloses a method for acoustic sensing and directional acoustic control of crops based on a drone swarm, including constructing an acoustic parameter model of the crop growth area. This invention abandons traditional chemical pesticides, utilizing sound waves for physical control, avoiding environmental pollution and pesticide residues, and meeting the needs of green agriculture development. By using an acoustic sensor array to penetrate the crop canopy, it captures the acoustic characteristics and weak physiological signals of early signs of pests and diseases. Combined with multi-dimensional environmental data, it solves the problems of obstruction and lag in optical methods, achieving early intervention with "seeing the subtle signs." Furthermore, it adopts a two-layer drone swarm architecture of "wide-area patrol + close-plant precision control." The first layer is responsible for large-area scanning and initial intervention, while the second layer penetrates deep into the canopy and plant gaps for directional acoustic compensation, completely eliminating acoustic shadows and signal blind spots. A spatiotemporal collaborative algorithm is used to coordinate the timing and location of swarm operations, achieving early perception, layered intervention, and dynamic optimization of pests and diseases.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, specifically to a method for crop acoustic sensing and directional acoustic control based on drone swarms. Background Technology

[0002] As modern agriculture develops towards precision, intelligence, and green practices, crop pest and disease control and growth regulation have become crucial links in ensuring food security and improving the quality of agricultural products. Traditional agricultural management methods mainly rely on manual field inspections and experience-based judgment, which are labor-intensive, inefficient, and highly subjective. While subsequent development of ground machinery operations has improved efficiency, it has limitations such as soil compaction, mechanical damage to crops, and the inability to enter the field during the later stages of crop growth.

[0003] In recent years, drone technology has been widely used in agricultural plant protection, mainly focusing on pesticide spraying, sowing and fertilization, and crop monitoring based on vision / multispectral imaging. However, existing drone agricultural applications still face the following significant technical bottlenecks: 1. Limited and Lagging Perception Dimensions: Existing agricultural drones primarily rely on optical cameras (RGB, multispectral, hyperspectral) for crop phenotypic monitoring. However, optical methods are highly susceptible to lighting conditions (such as strong light and shadow, nighttime) and leaf shading, making it difficult to penetrate dense crop canopies. More importantly, optical imaging typically only identifies pests and diseases after they have manifested visible symptoms (such as spots, wilting), resulting in significant time lag. In reality, crops emit specific acoustic signals (such as cavitation sounds, friction sounds, and stress sound emissions) and weak physiological acoustic emissions in the early stages of biotic stress (such as insect chewing, pathogen infection) or abiotic stress (such as drought, salinity). Current technologies lack effective means to collect and analyze these "crop sounds," making early intervention during the "golden window" of stress occurrence impossible.

[0004] 2. Inefficient and Environmentally Unfriendly Control Methods: Current drone operations mostly rely on chemical pesticide spraying or physical impact (such as weeding). The overuse of chemical pesticides has led to serious environmental pollution, increased pest resistance, and excessive pesticide residues. Physical weeding, on the other hand, is energy-intensive and can easily damage crops. Acoustic wave control, as a physical agricultural method, has the potential to be residue-free, resistance-free, and energy-efficient. However, existing acoustic control equipment is mostly ground-based or handheld, which has the following drawbacks: Limited scope of application: unable to cover large areas of farmland; Lack of directionality: Sound waves are emitted in all directions, resulting in low energy utilization and easy noise pollution to non-target areas; Fixed parameters: The frequency, sound pressure level and beam direction of the sound wave cannot be dynamically adjusted according to the crop type, growth stage and real-time environment (temperature, humidity, wind speed), resulting in unstable intervention effect and even causing mechanical damage to the crop (such as sound-induced damage). Summary of the Invention

[0005] To address the aforementioned technical problems, this paper provides a method for crop acoustic sensing and directional acoustic control based on unmanned aerial vehicle (UAV) swarms. This technical solution solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect of the present invention, a method for crop acoustic sensing and directional acoustic control based on unmanned aerial vehicle (UAV) swarms is provided, comprising the following steps: A sound wave parameter model of the crop growth area is constructed. The model includes a sound wave response feature library of different crop growth stages, a sound wave feature library of disease and pest precursors, and the corresponding set of optimal intervention sound wave parameters. Deploy a drone swarm to conduct multi-level spatial coverage operations on target farmland. The swarm includes a first-level drone that performs wide-area patrols and a second-level drone that performs near-plant fine-tuning interventions. By using an acoustic sensor array and environmental sensors mounted on a drone, acoustic response signals, micro-meteorological parameters, and crop physiological and ecological data in the crop canopy and gaps are collected in real time to generate a multi-dimensional spatiotemporal sensing dataset. The multidimensional spatiotemporal sensing dataset is cleaned and its features are fused to extract abnormal acoustic fingerprints and environmental features that reflect crop stress status. The extracted features are input into the acoustic parameter model for reasoning and analysis to determine the current crop growth status and pest and disease risk level, and generate acoustic intervention strategy instructions for different spatial locations. According to the intervention strategy instructions, the sound wave emission frequency, sound pressure level and beam direction of the first-level UAV are dynamically adjusted, while the second-level UAV is guided to enter the interior of the crop canopy or the gaps between plants to perform directional sound wave compensation in the signal-obstructed area. Establish communication links between drone clusters, and coordinate the operation sequence and spatial location of each drone based on spatiotemporal collaborative algorithms to ensure that acoustic intervention is continuous in time and has no blind spots in space.

[0007] Preferably, the construction of the acoustic parameter model of the crop growth area specifically includes the following steps: Historical farmland acoustic data, crop physiological experimental data, and acoustic samples of pests and diseases were collected to establish an original acoustic feature database. The data included the background noise spectrum of healthy crops, the stress sound emission spectrum of damaged crops, the chewing sound of pests, the oviposition sound, and the time-frequency characteristics of the release sound of pathogen spores. Wavelet packet transform decomposition was performed on the original acoustic data to extract the energy entropy, zero-crossing rate and Mel frequency cepstral coefficients of each frequency band. Principal component analysis was used to reduce the dimensionality of the high-dimensional features and retain the key feature vectors with a cumulative contribution rate greater than 95%. A deep learning model based on the concatenation of CNN and LSTM is constructed, where CNN is used to extract the spatial texture features of the spectrogram, and LSTM is used to capture the acoustic temporal evolution pattern during the crop growth cycle. A transfer learning mechanism was introduced to transfer the acoustic features of pests and diseases under standard laboratory conditions to a complex field environment model. Generative adversarial networks were used to generate interference data simulating field wind noise, rain noise, and background insect sounds to train the model adversarially and enhance its robustness. A three-dimensional mapping relationship table of "pest and disease type - acoustic fingerprint - intervention parameters" is established. The particle swarm optimization algorithm is used to search for and lock the optimal intervention sound wave frequency, sound pressure level and modulation mode with the objective function of maximizing the recovery rate of crop physiological indicators and minimizing energy consumption after intervention, thus forming the optimal sound wave parameter model.

[0008] Preferably, the generation of the multidimensional spatiotemporal perception dataset specifically includes the following steps: The raw audio stream is collected by an acoustic sensor array, and an adaptive filtering algorithm is used to remove drone rotor self-noise, motor electromagnetic noise and wind-induced noise in real time, while retaining the effective acoustic signal inside the crop canopy. Collect micrometeorological parameters and crop physiological data, and use Kalman filtering to synchronize the time and estimate the state of multi-sensor data to correct sensor clock drift. A three-dimensional spatial grid based on a geographic information system is established, and acoustic data, meteorological data, and physiological data are mapped to corresponding grid coordinates to form a four-dimensional raw data matrix with timestamps and spatial coordinates; Outlier detection is performed on the four-dimensional data matrix. The isolated forest algorithm is used to identify and remove outliers caused by sensor failure or sudden strong interference. Kriging interpolation is used to spatially complete the missing data. The signal-to-noise ratio and coherence coefficient of the acoustic signal are calculated, invalid data segments with coherence coefficients below a preset threshold are removed, and finally a high-quality multidimensional spatiotemporal sensing dataset is generated.

[0009] Preferably, the extraction of anomalous acoustic fingerprints and environmental correlation features reflecting crop stress states specifically includes the following steps: Short-time Fourier transform is performed on the cleaned acoustic data to generate a high-resolution spectrogram. The texture and shape features in the spectrogram are then extracted using computer vision algorithms. An environmental-acoustic coupling model was constructed to analyze the influence of micro-meteorological parameters on sound wave propagation attenuation. Sound wave refraction correction coefficients under different temperature and humidity gradients were calculated, and sound pressure level data were normalized. The sound pressure level normalization correction formula is as follows: ; in, To normalize the sound pressure level, For the original measured sound pressure level, The distance between the sound source and the sensor. For temperature and relative humidity The relevant air absorption coefficient, This is the system calibration constant; A multimodal fusion algorithm was used to analyze the correlation between acoustic features and crop physiological features, calculate the Pearson correlation coefficient and mutual information value, and screen out the "sensitive feature combination" that is strongly correlated with the occurrence of pests and diseases. An abnormal acoustic fingerprint database is defined. Using the DBSCAN density clustering algorithm in unsupervised learning, the currently collected acoustic features are compared with a healthy crop benchmark fingerprint database. The Euclidean distance and Dynamic Time Warped (DTW) distance are calculated. The DTW distance calculation formula is as follows: ; in, For the sequence X, the first... The first frame of sequence Y Cumulative distance between frames The local distance between the feature vectors of the two frames is denoted as , and the Euclidean distance is denoted as . When the distance metric exceeds the adaptive threshold and at least two indicators in the sensitive feature combination trigger an alert, it is determined that there is an abnormal acoustic fingerprint, and the stress type is determined based on the cluster center of the feature vector.

[0010] Preferably, the step of inputting the extracted features into the acoustic parameter model for inference analysis specifically includes the following steps: The abnormal acoustic fingerprint and environmental association features are vectorized and fed into a pre-trained random forest classifier as input. The output is the probability distribution of the current crop growth status and the risk level of pests and diseases. The final classification probability formula of the random forest is: ; in, For the number of decision trees, For the first Trees for input features Predicted as category The indicator function (0 or 1); Based on the risk level, the expert rule base in the sound wave parameter model is called. If it is determined to be high risk, the "targeted extermination" mode is activated, matching the resonance frequency for the auditory system of a specific pest. If the risk is determined to be moderate or abiotic stress, the "growth regulation" mode is activated, matching specific frequency sound waves that promote crop gene expression, and using ray tracing models to predict the propagation path and attenuation of sound waves in the crop canopy in combination with current micrometeorological data. Based on the prediction results, the required sound pressure level compensation is dynamically calculated to ensure that the energy density of the sound wave when it reaches the target leaf or stem is within the effective intervention range. The sound energy attenuation compensation formula is as follows: ; in, For the target point sound intensity, To transmit sound intensity, The medium absorption coefficient, For sound wave frequency, For transmission distance, The directional gain function is related to the beam pointing angle. Related; The system generates digital intervention strategy instructions that include target coordinates, transmission frequency, sound pressure level, beamwidth, duration, and modulation code, and performs hash verification on the instructions to prevent transmission errors.

[0011] Preferably, guiding the second-level drone into the crop canopy or between plants specifically includes the following steps: Based on the wide-area sound field distribution map transmitted back by the first-level UAV, image segmentation algorithms are used to identify acoustic signal blind spots or areas with severe signal attenuation. Based on a 3D point cloud map, a fast exploratory random tree (RRT) path planning algorithm is used to plan a collision-free flight path for the second-level UAV, entering the canopy interior from a safe altitude. The expansion bias formula of the RRT algorithm is: ; in, The nearest node in the tree. The vector is the random sample point and its nearest node. To extend the step size; During the drone's descent, real-time monitoring of airflow disturbance data around the aircraft is conducted, and the rotor speed is adjusted using a model predictive control (MPC) algorithm to maintain the aircraft's attitude stability in turbulent conditions. The cost function of MPC is defined as: ; in, To predict the time domain, To control the time domain, For predicting output (attitude angle). For reference trajectory, To control the increment (rotor speed change). , It is a weighted matrix; The directional acoustic transducer mounted on the second-level UAV is activated, and beamforming technology is used to focus the acoustic energy on a local area with a diameter of less than 10cm. During the intervention, a near-field microphone array is used to monitor the reflected sound waves at the target location in real time. The emission phase is dynamically fine-tuned through a feedback control loop, and the sound shadow in the shielded area is eliminated by using the principle of destructive interference.

[0012] Preferably, the dynamic adjustment of the acoustic wave emission frequency, sound pressure level, and beam direction of the first-level UAV specifically includes the following steps: Real-time reception of environmental sensor data, analysis of the influence of wind speed vector and temperature gradient on sound wave refraction, and calculation of the optimal launch elevation angle under current atmospheric conditions using the sound propagation equation; By monitoring changes in crop canopy density, leaf area index is calculated using lidar point data. The acoustic frequency is dynamically adjusted based on canopy density, and the canopy penetration coefficient τ is defined as follows: ; in, Leaf area index, For crop-specific extinction coefficients, when When the value is less than 0.3, it is considered a dense region, and the frequency is switched to the low frequency band (f < 5kHz). Closed-loop power control is implemented. Based on the intensity of the received crop-reflected sound waves, the sound pressure level output is adjusted in real time using a PID control algorithm. The PID control law formula is as follows: ; in, To set the deviation between the echo intensity and the actual feedback, , , These are the proportional, integral, and differential coefficients, respectively. Frequency hopping spread spectrum technology is used to quickly switch between multiple intervention frequencies, avoiding the development of pests' adaptation due to long-term use of a single frequency; Record the parameter changes and crop physiological response data for each adjustment to generate a parameter adjustment log.

[0013] Preferably, the coordination of the operational timing and spatial location of each UAV based on the spatiotemporal cooperative algorithm includes: Establish a self-organizing network for unmanned aerial vehicle (UAV) swarms, employing a hybrid access protocol combining time division multiple access (TDMA) and frequency division multiple access (FDMA), and divide the control channel into a data channel; A time-space synchronization protocol is introduced, and the absolute time and space coordinates of each UAV are aligned using GPS / RTK modules. The operation time is divided into continuous time slices, and the farmland is divided into several hexagonal operation sub-areas. A task allocation algorithm based on game theory is adopted, with the optimization objectives of minimizing total cluster energy consumption and coverage overlap rate. The number ratio of drones in each sub-region is dynamically calculated, and the total cluster energy consumption cost function is defined as follows: ; in, For the number of drones, For flight power consumption, For acoustic emission power consumption, For acoustic emission duty cycle, For homework time, To cover the overlapping area, This is the overlap penalty coefficient; When a drone needs to leave due to insufficient power or malfunction, a dynamic rescheduling mechanism is triggered, and neighboring drones automatically take over the operation sub-area it is responsible for through negotiation. Voronoi diagrams are used to divide the acoustic field coverage responsibility area of ​​each UAV, and overlapping buffer zones are set at the boundaries of the responsibility areas. By exchanging acoustic state data at the boundaries, the beam pointing of adjacent UAVs is coordinated.

[0014] In a second aspect of the invention, a crop acoustic sensing and directional acoustic control system based on a drone swarm is also provided, comprising: The construction module is used to construct an acoustic parameter model of the crop growth area. The model includes an acoustic response feature library for different crop growth stages, an acoustic feature library of disease and pest precursors, and the corresponding set of optimal intervention acoustic parameters. The drone module is used to deploy a drone swarm to perform multi-level spatial coverage operations on target farmland. The swarm includes a first-level drone that performs wide-area patrols and a second-level drone that performs near-plant fine-tuning interventions. The acquisition module is used to collect acoustic response signals, micro-meteorological parameters and crop physiological and ecological data in real time through an acoustic sensor array and environmental sensors mounted on the UAV, and generate a multi-dimensional spatiotemporal perception dataset. The data processing module is used to clean and fuse features of the multidimensional spatiotemporal sensing dataset, and extract abnormal acoustic fingerprints and environmental correlation features that reflect crop stress status. The determination module is used to input the extracted features into the acoustic parameter model for reasoning and analysis, determine the current crop growth status and pest and disease risk level, and generate acoustic intervention strategy instructions for different spatial locations. The compensation module is used to dynamically adjust the acoustic emission frequency, sound pressure level and beam direction of the first-level UAV according to the intervention strategy instructions, and at the same time guide the second-level UAV to enter the interior of the crop canopy or the gaps between plants to perform directional acoustic compensation for the signal-obstructed area. The coordination module is used to establish communication links between drone clusters and coordinate the operation sequence and spatial position of each drone based on a spatiotemporal cooperative algorithm to ensure that the acoustic intervention is continuous in time and has no blind spots in space.

[0015] In a third aspect of the invention, an electronic device is also provided. The electronic device includes at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of the first aspect of the invention.

[0016] Compared with existing technologies, this invention provides a method for crop acoustic sensing and directional acoustic control based on UAV swarms, which has the following beneficial effects: This invention abandons traditional chemical pesticides and utilizes sound waves for physical regulation, avoiding environmental pollution and pesticide residues, meeting the needs of green agriculture development, and does not produce pesticide resistance. Through an acoustic sensor array penetrating the crop canopy, it captures the acoustic characteristics of early signs of pests and diseases (such as insect chewing sounds and plant stress sounds) and weak physiological signals. Combined with multi-dimensional environmental data, it solves the problems of obstruction and lag in optical methods, achieving early intervention with "seeing the subtle signs." Furthermore, it adopts a two-layer drone swarm architecture of "wide-area patrol + close-plant precision control," with the first layer responsible for large-area coverage. The first step involves scanning and initial intervention, followed by a second-level in-depth approach to directional acoustic compensation within the canopy and plant gaps, completely eliminating acoustic shadows and signal blind spots. Based on real-time cleaned spatiotemporal data and acoustic parameter models, the system dynamically adjusts the transmission frequency, sound pressure level, and beam direction to precisely focus acoustic energy on the target area, ensuring effective intervention while avoiding mechanical damage to crops. Furthermore, a spatiotemporal collaborative algorithm coordinates the timing and location of cluster operations, ensuring continuous acoustic intervention in time and eliminating blind spots in space, significantly improving the efficiency and uniformity of operations across large areas of farmland. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the method flow of S101-S107 in this invention; Figure 2 This is a schematic diagram of the method flow for S201-S205 in this invention; Figure 3 This is a schematic diagram of the method flow for S301-S305 in this invention; Figure 4This is a schematic diagram of the method flow for S401-S405 in this invention; Figure 5 This is a schematic diagram of the method flow for S501-S505 in this invention. Detailed Implementation

[0018] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0019] Example 1 Please refer to Figure 1 As shown, in a first aspect of the present invention, a method for crop acoustic sensing and directional acoustic control based on unmanned aerial vehicle (UAV) swarms is provided, comprising the following steps: S101. Construct a sound wave parameter model for the crop growth area. The model includes a sound wave response feature library for different crop growth stages, a sound wave feature library for disease and pest precursors, and the corresponding set of optimal intervention sound wave parameters. S102. Deploy a drone swarm to conduct multi-level spatial coverage operations on the target farmland. The swarm includes a first-level drone that performs wide-area patrols and a second-level drone that performs near-plant fine-tuning interventions. S103. By using an acoustic sensor array and environmental sensors mounted on the drone, acoustic response signals, micro-meteorological parameters and crop physiological and ecological data in the crop canopy and gaps are collected in real time to generate a multi-dimensional spatiotemporal perception dataset. S104. Clean and fuse the multidimensional spatiotemporal perception dataset to extract abnormal acoustic fingerprints and environmental features that reflect crop stress status. S105. Input the extracted features into the acoustic parameter model for reasoning analysis, determine the current crop growth status and pest and disease risk level, and generate acoustic intervention strategy instructions for different spatial locations. S106. According to the intervention strategy instructions, dynamically adjust the sound wave emission frequency, sound pressure level and beam direction of the first-level UAV, and at the same time guide the second-level UAV to enter the interior of the crop canopy or the gap between plants to perform directional sound wave compensation for the signal-obscured area. S107. Establish communication links between drone clusters, coordinate the operation sequence and spatial location of each drone based on spatiotemporal collaborative algorithms, and ensure that acoustic intervention is continuous in time and has no blind spots in space.

[0020] As will be understood by those skilled in the art, this invention abandons traditional chemical pesticides and utilizes sound waves for physical regulation, avoiding environmental pollution and pesticide residues, meeting the needs of green agricultural development, and does not produce pesticide resistance; by using an acoustic sensor array to penetrate the crop canopy, it captures the acoustic characteristics of early signs of pests and diseases (such as the chewing sounds of pests and the stress sounds of plants) and weak physiological signals, combined with multi-dimensional environmental data, solving the problems of obstruction and lag in optical methods, and achieving early intervention by "seeing the big picture from small details"; and by adopting a two-layer UAV cluster architecture of "wide-area patrol + close-plant precision control", the first The first layer is responsible for large-area scanning and initial intervention, while the second layer delves into the canopy and plant gaps to perform directional acoustic compensation, completely eliminating acoustic shadows and signal blind spots. Based on real-time cleaned spatiotemporal data and acoustic parameter models, the transmission frequency, sound pressure level, and beam direction are dynamically adjusted to ensure that acoustic energy is precisely focused on the target area, ensuring the intervention effect while avoiding mechanical damage to crops. The spatiotemporal collaborative algorithm coordinates the timing and location of cluster operations, ensuring that acoustic intervention is continuous in time and without blind spots in space, significantly improving the operational efficiency and uniformity of large-area farmland.

[0021] Please refer to Figure 2 As shown, the construction of an acoustic parameter model for the crop growth area includes the following steps: S201. Collect historical farmland acoustic data, crop physiological experimental data, and acoustic samples of pests and diseases to establish an original acoustic feature database. The data includes the background noise spectrum of healthy crops, the stress sound emission spectrum of damaged crops, the chewing sound of pests, the oviposition sound, and the time-frequency characteristics of the release sound of pathogen spores. S202. Perform wavelet packet transform decomposition on the original acoustic data, extract the energy entropy, zero-crossing rate and Mel frequency cepstral coefficients of each frequency band, and use principal component analysis to reduce the dimensionality of high-dimensional features, retaining key feature vectors with a cumulative contribution rate greater than 95%. S203. Construct a deep learning model based on the concatenation of CNN and LSTM, where CNN is used to extract the spatial texture features of the spectrogram, and LSTM is used to capture the acoustic temporal evolution pattern during the crop growth cycle. S204. Introduce a transfer learning mechanism to transfer the acoustic features of pests and diseases under standard laboratory conditions to a complex field environment model. Use generative adversarial networks to generate interference data that simulates field wind noise, rain noise, and background insect sounds to train the model adversarially and enhance its robustness. S205. Establish a three-dimensional mapping relationship table of "pest and disease type - acoustic fingerprint - intervention parameters". Use particle swarm optimization algorithm with the objective function of maximizing the recovery rate of crop physiological indicators and minimizing energy consumption after intervention to search for and lock the optimal intervention sound wave frequency, sound pressure level and modulation mode to form the optimal sound wave parameter model.

[0022] Please refer to Figure 3As shown, generating a multidimensional spatiotemporal perception dataset specifically includes the following steps: S301: Acquires raw audio stream through acoustic sensor array, and uses adaptive filtering algorithm to remove drone rotor self-noise, motor electromagnetic noise and wind-induced noise in real time, while retaining effective acoustic signals inside crop canopy. S302. Collect micro-meteorological parameters and crop physiological data, and use Kalman filtering to synchronize the time and estimate the state of multi-sensor data to correct sensor clock drift. S303. Establish a three-dimensional spatial grid based on a geographic information system, and map acoustic data, meteorological data and physiological data to the corresponding grid coordinates to form a four-dimensional raw data matrix with timestamps and spatial coordinates. S304. Perform outlier detection on the four-dimensional data matrix, use the isolated forest algorithm to identify and remove outliers caused by sensor failure or sudden strong interference, and use Kriging interpolation to spatially complete the missing data. S305. Calculate the signal-to-noise ratio and coherence coefficient of the acoustic signal, remove invalid data segments with coherence coefficients below a preset threshold, and finally generate a high-quality multidimensional spatiotemporal sensing dataset.

[0023] Please refer to Figure 4 As shown, the extraction of anomalous acoustic fingerprints and environmental correlation features reflecting crop stress states includes the following steps: S401. Perform a short-time Fourier transform on the cleaned acoustic data to generate a high-resolution spectrogram, and use computer vision algorithms to extract texture and shape features from the spectrogram. S402. Construct an environmental-acoustic coupling model, analyze the influence of micro-meteorological parameters on sound wave propagation attenuation, calculate the sound wave refraction correction coefficient under different temperature and humidity gradients, and normalize the sound pressure level data; the sound pressure level normalization correction formula is: ; in, To normalize the sound pressure level, For the original measured sound pressure level, The distance between the sound source and the sensor. For temperature and relative humidity The relevant air absorption coefficient, This is the system calibration constant; S403. Using a multimodal fusion algorithm, acoustic features are correlated with crop physiological features, Pearson correlation coefficient and mutual information value are calculated, and "sensitive feature combinations" that are strongly correlated with the occurrence of pests and diseases are selected. S404. Define an abnormal acoustic fingerprint database. Using the DBSCAN density clustering algorithm in unsupervised learning, compare the currently collected acoustic features with the healthy crop benchmark fingerprint database, and calculate the Euclidean distance and the Dynamic Time Warped (DTW) distance. The DTW distance calculation formula is as follows: ; in, For the sequence X, the first... The first frame of sequence Y Cumulative distance between frames The local distance between the feature vectors of the two frames is denoted as , and the Euclidean distance is denoted as . S405. When the distance metric exceeds the adaptive threshold and at least two indicators in the sensitive feature combination trigger an early warning, it is determined that there is an abnormal acoustic fingerprint, and the stress type is determined based on the cluster center of the feature vector.

[0024] The extracted features are input into the acoustic wave parameter model for inference and analysis, specifically including the following steps: The abnormal acoustic fingerprint and environmental association features are vectorized and fed into a pre-trained random forest classifier as input. The output is the probability distribution of the current crop growth status and the risk level of pests and diseases. The final classification probability formula of the random forest is: ; in, For the number of decision trees, For the first Trees for input features Predicted as category The indicator function (0 or 1); Based on the risk level, the expert rule base in the sound wave parameter model is called. If it is determined to be high risk, the "targeted extermination" mode is activated, matching the resonance frequency for the auditory system of a specific pest. If the risk is determined to be moderate or abiotic stress, the "growth regulation" mode is activated, matching specific frequency sound waves that promote crop gene expression, and using ray tracing models to predict the propagation path and attenuation of sound waves in the crop canopy in combination with current micrometeorological data. Based on the prediction results, the required sound pressure level compensation is dynamically calculated to ensure that the energy density of the sound wave when it reaches the target leaf or stem is within the effective intervention range. The sound energy attenuation compensation formula is as follows: ; in, For the target point sound intensity, To transmit sound intensity, The medium absorption coefficient, For sound wave frequency, For transmission distance, The directional gain function is related to the beam pointing angle. Related; The system generates digital intervention strategy instructions that include target coordinates, transmission frequency, sound pressure level, beamwidth, duration, and modulation code, and performs hash verification on the instructions to prevent transmission errors.

[0025] Guiding the second-level drone into the crop canopy or between plants involves the following steps: Based on the wide-area sound field distribution map transmitted back by the first-level UAV, image segmentation algorithms are used to identify acoustic signal blind spots or areas with severe signal attenuation. Based on a 3D point cloud map, a fast exploratory random tree (RRT) path planning algorithm is used to plan a collision-free flight path for the second-level UAV, entering the canopy interior from a safe altitude. The expansion bias formula of the RRT algorithm is: ; in, The nearest node in the tree. The vector is the random sample point and its nearest node. To extend the step size; During the drone's descent, real-time monitoring of airflow disturbance data around the aircraft is conducted, and the rotor speed is adjusted using a model predictive control (MPC) algorithm to maintain the aircraft's attitude stability in turbulent conditions. The cost function of MPC is defined as: ; in, To predict the time domain, To control the time domain, For predicting output (attitude angle). For reference trajectory, To control the increment (rotor speed change). , It is a weighted matrix; The directional acoustic transducer mounted on the second-level UAV is activated, and beamforming technology is used to focus the acoustic energy on a local area with a diameter of less than 10cm. During the intervention, a near-field microphone array is used to monitor the reflected sound waves at the target location in real time. The emission phase is dynamically fine-tuned through a feedback control loop, and the sound shadow in the shielded area is eliminated by using the principle of destructive interference.

[0026] Dynamically adjusting the acoustic emission frequency, sound pressure level, and beam direction of the first-level UAV includes the following steps: Real-time reception of environmental sensor data, analysis of the influence of wind speed vector and temperature gradient on sound wave refraction, and calculation of the optimal launch elevation angle under current atmospheric conditions using the sound propagation equation; By monitoring changes in crop canopy density, leaf area index is calculated using lidar point data. The acoustic frequency is dynamically adjusted based on canopy density, and the canopy penetration coefficient τ is defined as follows: ; in, Leaf area index, For crop-specific extinction coefficients, when When the value is less than 0.3, it is considered a dense region, and the frequency is switched to the low frequency band (f < 5kHz). Closed-loop power control is implemented. Based on the intensity of the received crop-reflected sound waves, the sound pressure level output is adjusted in real time using a PID control algorithm. The PID control law formula is as follows: ; in, To set the deviation between the echo intensity and the actual feedback, , , These are the proportional, integral, and differential coefficients, respectively. Frequency hopping spread spectrum technology is used to quickly switch between multiple intervention frequencies, avoiding the development of pests' adaptation due to long-term use of a single frequency; Record the parameter changes and crop physiological response data for each adjustment to generate a parameter adjustment log.

[0027] Please refer to Figure 5 As shown, the coordination of the operational timing and spatial location of each UAV based on the spatiotemporal cooperative algorithm includes: S501. Establish a self-organizing network for UAV swarms, using a hybrid access protocol of time division multiple access and frequency division multiple access, and divide the control channel and data channel. S502 introduces a time-space synchronization protocol, uses GPS / RTK modules to align the absolute time and space coordinates of each UAV, divides the operation time into continuous time slices, and divides the farmland into several hexagonal operation sub-areas; S503. A task allocation algorithm based on game theory is adopted, with the optimization objectives of minimizing total cluster energy consumption and minimizing coverage overlap rate. The number ratio of drones in each sub-region is dynamically calculated, and the total cluster energy consumption cost function is defined as follows: ; in, For the number of drones, For flight power consumption, For acoustic emission power consumption, For acoustic emission duty cycle, For homework time, To cover the overlapping area, This is the overlap penalty coefficient; S504. When a drone needs to exit due to insufficient power or malfunction, a dynamic rescheduling mechanism is triggered, and neighboring drones automatically take over the operation sub-area it is responsible for through negotiation. S505. The acoustic field coverage responsibility area of ​​each UAV is divided using Voronoi diagrams. An overlapping buffer zone is set at the boundary of the responsibility area. The beam pointing of adjacent UAVs is coordinated by exchanging acoustic state data at the boundary.

[0028] In a second aspect of the invention, a crop acoustic sensing and directional acoustic control system based on a drone swarm is also provided, comprising: The building module is used to build acoustic parameter models of crop growth areas. The models include acoustic response feature libraries for different crop growth stages, acoustic feature libraries for disease and pest precursors, and corresponding optimal intervention acoustic parameter sets. The drone module is used to deploy drone swarms to carry out multi-level spatial coverage operations on target farmland. The swarm includes first-level drones that perform wide-area patrols and second-level drones that perform near-plant fine-tuning interventions. The acquisition module is used to collect acoustic response signals, micro-meteorological parameters and crop physiological and ecological data in real time from the acoustic sensor array and environmental sensors mounted on the UAV, and generate a multi-dimensional spatiotemporal sensing dataset. The data processing module is used to clean and fuse features of the multidimensional spatiotemporal sensing dataset, and extract abnormal acoustic fingerprints and environmental correlation features that reflect crop stress status. The judgment module is used to input the extracted features into the acoustic parameter model for reasoning and analysis, determine the current crop growth status and pest and disease risk level, and generate acoustic intervention strategy instructions for different spatial locations. The compensation module is used to dynamically adjust the acoustic emission frequency, sound pressure level and beam direction of the first-level UAV according to the intervention strategy instructions, and at the same time guide the second-level UAV into the interior of the crop canopy or the gaps between plants to perform directional acoustic compensation in the signal-obstructed area. The coordination module is used to establish communication links between drone clusters. Based on the spatiotemporal cooperative algorithm, it coordinates the operation sequence and spatial position of each drone to ensure that the acoustic intervention is continuous in time and has no blind spots in space.

[0029] In a third aspect of the invention, an electronic device is also provided. The electronic device includes at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the method of the first aspect of the invention.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for crop acoustic sensing and directional acoustic control based on unmanned aerial vehicle (UAV) swarms, characterized in that, Includes the following steps: A sound wave parameter model of the crop growth area is constructed. The model includes a sound wave response feature library of different crop growth stages, a sound wave feature library of disease and pest precursors, and the corresponding set of optimal intervention sound wave parameters. Deploy a drone swarm to conduct multi-level spatial coverage operations on target farmland. The swarm includes a first-level drone that performs wide-area patrols and a second-level drone that performs near-plant fine-tuning interventions. By using an acoustic sensor array and environmental sensors mounted on a drone, acoustic response signals, micro-meteorological parameters, and crop physiological and ecological data in the crop canopy and gaps are collected in real time to generate a multi-dimensional spatiotemporal sensing dataset. The multidimensional spatiotemporal sensing dataset is cleaned and its features are fused to extract abnormal acoustic fingerprints and environmental features that reflect crop stress status. The extracted features are input into the acoustic parameter model for reasoning and analysis to determine the current crop growth status and pest and disease risk level, and generate acoustic intervention strategy instructions for different spatial locations. According to the intervention strategy instructions, the sound wave emission frequency, sound pressure level and beam direction of the first-level UAV are dynamically adjusted, while the second-level UAV is guided to enter the interior of the crop canopy or the gaps between plants to perform directional sound wave compensation in the signal-obstructed area. Establish communication links between drone clusters, and coordinate the operation sequence and spatial location of each drone based on spatiotemporal collaborative algorithms to ensure that acoustic intervention is continuous in time and has no blind spots in space.

2. The method for crop acoustic sensing and directional acoustic control based on UAV swarms according to claim 1, characterized in that, The construction of the acoustic parameter model for the crop growth area specifically includes the following steps: Historical farmland acoustic data, crop physiological experimental data, and acoustic samples of pests and diseases were collected to establish an original acoustic feature database. The data included the background noise spectrum of healthy crops, the stress sound emission spectrum of damaged crops, the chewing sound of pests, the oviposition sound, and the time-frequency characteristics of the release sound of pathogen spores. Wavelet packet transform decomposition was performed on the original acoustic data to extract the energy entropy, zero-crossing rate and Mel frequency cepstral coefficients of each frequency band. Principal component analysis was used to reduce the dimensionality of the high-dimensional features and retain the key feature vectors with a cumulative contribution rate greater than 95%. A deep learning model based on the concatenation of CNN and LSTM is constructed, where CNN is used to extract the spatial texture features of the spectrogram, and LSTM is used to capture the acoustic temporal evolution pattern during the crop growth cycle. A transfer learning mechanism was introduced to transfer the acoustic features of pests and diseases under standard laboratory conditions to a complex field environment model. Generative adversarial networks were used to generate interference data simulating field wind noise, rain noise, and background insect sounds to train the model adversarially and enhance its robustness. A three-dimensional mapping table of "pest and disease type - acoustic fingerprint - intervention parameters" is established. The particle swarm optimization algorithm is used to search for and lock the optimal intervention sound wave frequency, sound pressure level and modulation mode with the objective function of maximizing the recovery rate of crop physiological indicators and minimizing energy consumption after intervention, thus forming the optimal sound wave parameter model.

3. The method for crop acoustic sensing and directional acoustic control based on UAV swarms according to claim 2, characterized in that, The generation of the multidimensional spatiotemporal perception dataset specifically includes the following steps: The raw audio stream is collected by an acoustic sensor array, and an adaptive filtering algorithm is used to remove drone rotor self-noise, motor electromagnetic noise and wind-induced noise in real time, while retaining the effective acoustic signal inside the crop canopy. Collect micrometeorological parameters and crop physiological data, and use Kalman filtering to synchronize the time and estimate the state of multi-sensor data to correct sensor clock drift. A three-dimensional spatial grid based on a geographic information system is established, and acoustic data, meteorological data, and physiological data are mapped to corresponding grid coordinates to form a four-dimensional raw data matrix with timestamps and spatial coordinates; Outlier detection is performed on the four-dimensional data matrix. The isolated forest algorithm is used to identify and remove outliers caused by sensor failure or sudden strong interference. Kriging interpolation is used to spatially complete the missing data. The signal-to-noise ratio and coherence coefficient of the acoustic signal are calculated, invalid data segments with coherence coefficients below a preset threshold are removed, and finally a high-quality multidimensional spatiotemporal sensing dataset is generated.

4. The method for crop acoustic sensing and directional acoustic control based on UAV swarms according to claim 3, characterized in that, The extraction of anomalous acoustic fingerprints and environmental correlation features reflecting crop stress states specifically includes the following steps: Short-time Fourier transform is performed on the cleaned acoustic data to generate a high-resolution spectrogram. The texture and shape features in the spectrogram are then extracted using computer vision algorithms. An environmental-acoustic coupling model was constructed to analyze the influence of micro-meteorological parameters on sound wave propagation attenuation. Sound wave refraction correction coefficients under different temperature and humidity gradients were calculated, and sound pressure level data were normalized. The sound pressure level normalization correction formula is as follows: ； in, To normalize the sound pressure level, For the original measured sound pressure level, The distance between the sound source and the sensor. For temperature and relative humidity The relevant air absorption coefficient, This is the system calibration constant; A multimodal fusion algorithm was used to analyze the correlation between acoustic features and crop physiological features, calculate the Pearson correlation coefficient and mutual information value, and screen out "sensitive feature combinations" that are strongly correlated with the occurrence of pests and diseases. An abnormal acoustic fingerprint database is defined. Using the DBSCAN density clustering algorithm in unsupervised learning, the currently collected acoustic features are compared with a healthy crop benchmark fingerprint database. The Euclidean distance and Dynamic Time Warped (DTW) distance are calculated. The DTW distance calculation formula is as follows: ； in, For the sequence X, the first Frame and sequence Y Cumulative distance between frames The local distance between the feature vectors of the two frames is denoted as , and the Euclidean distance is denoted as . When the distance metric exceeds the adaptive threshold and at least two indicators in the sensitive feature combination trigger an alert, it is determined that there is an abnormal acoustic fingerprint, and the stress type is determined based on the cluster center of the feature vector.

5. The method for crop acoustic sensing and directional acoustic control based on UAV swarms according to claim 4, characterized in that, The step of inputting the extracted features into the acoustic parameter model for inference analysis specifically includes the following steps: The abnormal acoustic fingerprint and environmental association features are vectorized and fed into a pre-trained random forest classifier as input. The output is the probability distribution of the current crop growth status and the risk level of pests and diseases. The final classification probability formula of the random forest is: ； in, For the number of decision trees, For the first Trees for input features Predicted as category The indicator function (0 or 1); Based on the risk level, the expert rule base in the sound wave parameter model is called. If it is determined to be high risk, the "targeted extermination" mode is activated, matching the resonance frequency for the auditory system of a specific pest. If the risk is determined to be moderate or abiotic stress, the "growth regulation" mode is activated, matching specific frequency sound waves that promote crop gene expression, and using ray tracing models to predict the propagation path and attenuation of sound waves in the crop canopy in combination with current micrometeorological data. Based on the prediction results, the required sound pressure level compensation is dynamically calculated to ensure that the energy density of the sound wave when it reaches the target leaf or stem is within the effective intervention range. The sound energy attenuation compensation formula is as follows: ； in, For the target point sound intensity, To transmit sound intensity, The medium absorption coefficient, For sound wave frequency, For transmission distance, The directional gain function is related to the beam pointing angle. Related; The system generates digital intervention strategy instructions that include target coordinates, transmission frequency, sound pressure level, beamwidth, duration, and modulation code, and performs hash verification on the instructions to prevent transmission errors.

6. The method for crop acoustic sensing and directional acoustic control based on UAV swarms according to claim 5, characterized in that, The process of guiding the second-level drone into the crop canopy or between plants specifically includes the following steps: Based on the wide-area sound field distribution map transmitted back by the first-level UAV, image segmentation algorithms are used to identify acoustic signal blind spots or areas with severe signal attenuation. Based on a 3D point cloud map, a fast exploratory random tree (RRT) path planning algorithm is used to plan a collision-free flight path for the second-level UAV, entering the canopy interior from a safe altitude. The expansion bias formula of the RRT algorithm is: ； in, The nearest node in the tree. This is a vector of random sampling points and their nearest nodes. To extend the step size; During the drone's descent, real-time monitoring of airflow disturbance data around the aircraft is conducted, and the rotor speed is adjusted using a model predictive control (MPC) algorithm to maintain the aircraft's attitude stability in turbulent conditions. The cost function of MPC is defined as: ； in, To predict the time domain, To control the time domain, For predicting output (attitude angle). For reference trajectory, To control the increment (rotor speed change). , It is a weighted matrix; The directional acoustic transducer mounted on the second-level UAV is activated, and beamforming technology is used to focus the acoustic energy on a local area with a diameter of less than 10cm. During the intervention, a near-field microphone array is used to monitor the reflected sound waves at the target location in real time. The emission phase is dynamically fine-tuned through a feedback control loop, and the sound shadow in the shielded area is eliminated by using the principle of destructive interference.

7. The method for crop acoustic sensing and directional acoustic control based on UAV swarms according to claim 6, characterized in that, The dynamic adjustment of the acoustic wave emission frequency, sound pressure level, and beam direction of the first-level UAV specifically includes the following steps: Real-time reception of environmental sensor data, analysis of the influence of wind speed vector and temperature gradient on sound wave refraction, and calculation of the optimal launch elevation angle under current atmospheric conditions using the sound propagation equation; To monitor changes in crop canopy density, leaf area index is calculated using lidar point data. The acoustic frequency is dynamically adjusted based on canopy density, and the canopy penetration coefficient τ is defined as follows: ； in, Leaf area index, For crop-specific extinction coefficients, when When the value is less than 0.3, it is considered a dense region, and the frequency is switched to the low frequency band (f < 5kHz). Closed-loop power control is implemented. Based on the intensity of the received crop-reflected sound waves, the sound pressure level output is adjusted in real time using a PID control algorithm. The PID control law formula is as follows: ； in, To set the deviation between the echo intensity and the actual feedback, , , These are the proportional, integral, and differential coefficients, respectively. Frequency hopping spread spectrum technology is used to quickly switch between multiple intervention frequencies, avoiding the development of pests' adaptation due to long-term use of a single frequency; Record the parameter changes and crop physiological response data for each adjustment to generate a parameter adjustment log.

8. The method for crop acoustic sensing and directional acoustic control based on UAV swarms according to claim 7, characterized in that, The coordination of the operational timing and spatial location of each UAV based on the spatiotemporal cooperative algorithm includes: Establish a self-organizing network for unmanned aerial vehicle (UAV) swarms, employing a hybrid access protocol combining time division multiple access (TDMA) and frequency division multiple access (FDMA), and divide the control channel into a data channel; A time-space synchronization protocol is introduced, and the absolute time and space coordinates of each UAV are aligned using GPS / RTK modules. The operation time is divided into continuous time slices, and the farmland is divided into several hexagonal operation sub-areas. A task allocation algorithm based on game theory is adopted, with the optimization objectives of minimizing total cluster energy consumption and coverage overlap rate. The number ratio of drones in each sub-region is dynamically calculated, and the total cluster energy consumption cost function is defined as follows: ； in, For the number of drones, For flight power consumption, For acoustic emission power consumption, For acoustic emission duty cycle, For homework time, To cover the overlapping area, This is the overlap penalty coefficient; When a drone needs to leave due to insufficient power or malfunction, a dynamic rescheduling mechanism is triggered, and neighboring drones automatically take over the operation sub-area it is responsible for through negotiation. Voronoi diagrams are used to divide the acoustic field coverage responsibility area of ​​each UAV, and overlapping buffer zones are set at the boundaries of the responsibility areas. By exchanging acoustic state data at the boundaries, the beam pointing of adjacent UAVs is coordinated.

9. A crop acoustic wave sensing and directional acoustic control system based on UAV swarms, used to implement the crop acoustic wave sensing and directional acoustic control method based on UAV swarms as described in any one of claims 1-8, characterized in that, include: The construction module is used to construct an acoustic parameter model of the crop growth area. The model includes an acoustic response feature library for different crop growth stages, an acoustic feature library of disease and pest precursors, and the corresponding set of optimal intervention acoustic parameters. The drone module is used to deploy a drone swarm to perform multi-level spatial coverage operations on target farmland. The swarm includes a first-level drone that performs wide-area patrols and a second-level drone that performs near-plant fine-tuning interventions. The acquisition module is used to collect acoustic response signals, micro-meteorological parameters and crop physiological and ecological data in real time through an acoustic sensor array and environmental sensors mounted on the UAV, and generate a multi-dimensional spatiotemporal sensing dataset. The data processing module is used to clean and fuse features of the multidimensional spatiotemporal sensing dataset, and extract abnormal acoustic fingerprints and environmental correlation features that reflect crop stress status. The determination module is used to input the extracted features into the acoustic parameter model for reasoning and analysis, determine the current crop growth status and pest and disease risk level, and generate acoustic intervention strategy instructions for different spatial locations. The compensation module is used to dynamically adjust the acoustic emission frequency, sound pressure level and beam direction of the first-level UAV according to the intervention strategy instructions, and at the same time guide the second-level UAV to enter the interior of the crop canopy or the gaps between plants to perform directional acoustic compensation for the signal-obstructed area. The coordination module is used to establish communication links between drone clusters and coordinate the operation sequence and spatial position of each drone based on a spatiotemporal cooperative algorithm to ensure that the acoustic intervention is continuous in time and has no blind spots in space.

10. An electronic device, comprising at least one processor; and a memory communicatively connected to said at least one processor; characterized in that, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

Images