An unmanned aerial vehicle flight anti-interference method and system in an emergency scenario

Abstract

The application discloses an unmanned aerial vehicle flight anti-interference method and system in an emergency scene, and relates to the field of unmanned aerial vehicle anti-interference. The method comprises the following steps: collecting environment reference parameters and unmanned aerial vehicle pose information; constructing an initial space vector baseline; when entering a target fire field, collecting fire field environment parameters and real-time state parameters; dynamically correcting the initial space vector baseline to obtain a dynamic space vector baseline, and completing double-clock phase alignment verification; if the double-clock phase alignment verification is passed, collecting fire field double-audio signals; performing multi-stage filtering on the fire field double-audio signals to obtain low-noise double-audio signals; obtaining sound wave detection signals, and completing double-mode acoustic data collaborative fusion to obtain fire field equivalent sound speed; completing sound wave trajectory simulation to inversely calculate non-entity interference data; obtaining double-machine point cloud data, and extracting entity interference data; and generating an anti-interference flight strategy. The application can effectively improve the unmanned aerial vehicle flight anti-interference capability in an emergency scene.

Description

Technical Field

[0001] This application relates to the field of anti-interference for unmanned aerial vehicles (UAVs), and more particularly to a method and system for anti-interference of UAV flight in emergency scenarios. Background Technology

[0002] With the rapid development of drone technology, it is playing an increasingly crucial role in emergency rescue scenarios such as fire rescue, chemical plant accident response, and inspection of areas with strong magnetic interference. However, in practical applications, the extremely harsh and complex environments of emergency scenarios pose severe challenges to the stability, reliability, and mission continuity of drone flights, seriously restricting the effectiveness of drones in fire emergency rescue.

[0003] Especially in emergency rescue missions such as forest fires, urban high-rise building fires, and chemical industrial park fires, when faced with interference such as high temperatures, dense smoke, and irregular airflow, traditional UAV perception and flight control solutions mainly achieve flight anti-interference by upgrading onboard equipment, adjusting communication links, and expanding sensor types. Whether it is equipment upgrades, link adjustments, or sensor expansion, they are all passive responses to interference and cannot dynamically adjust anti-interference strategies according to changes in the intensity and type of interference in real time at the fire site. The anti-interference continuity and flexibility are insufficient, making it difficult to meet the rescue needs of long-term stable flight in complex fire sites, and may still lead to the interruption of rescue missions, UAV flight instability, or even crashes. Summary of the Invention

[0004] This application provides a method and system for anti-interference of UAV flight in emergency scenarios, which is used to improve the anti-interference capability of UAV flight in emergency scenarios.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0006] Firstly, a method for anti-interference of UAV flight in emergency scenarios is provided, applicable to dual-aid UAVs, wherein the dual-aid UAVs include a primary aid UAV and a secondary aid UAV. The method includes:

[0007] Before the dual-aid drones enter the target fire area, environmental baseline parameters and drone pose information of the dual-aid drones are collected.

[0008] Based on the UAV pose information, complete the two-way ranging interaction between the main assistance UAV and the slave assistance UAV, and construct the initial spatial vector baseline of the two assistance UAVs based on the two-way ranging interaction results.

[0009] When the dual-aid drones enter the target fire site, they collect real-time fire environment parameters and real-time status parameters of the dual-aid drones.

[0010] The initial spatial vector baseline is dynamically corrected by combining real-time status parameters, fire environment parameters, and environmental reference parameters to obtain a dynamic spatial vector baseline. The dual clock phase alignment verification of the dual assistance UAVs is then completed based on the dynamic spatial vector baseline.

[0011] If the dual clock phase alignment verification of the dual-aid drone passes, the dual-audio signal of the target fire site will be collected using the microphone array carried by the dual-aid drone.

[0012] A multi-stage filtering algorithm was used to filter the dual-tone audio signal from the fire scene to obtain a low-noise dual-tone audio signal.

[0013] Acquire acoustic detection signals between the main support drone and the slave support drone, and complete the dual-modal acoustic data collaborative fusion of acoustic detection signals and low-noise dual-audio signals to obtain the equivalent sound velocity of the target fire site.

[0014] Based on the equivalent sound velocity of the fire field and using Fermat's principle, the sound wave trajectory of the target fire field is simulated, and the non-physical interference data of the target fire field is derived from the sound wave trajectory simulation results.

[0015] Acquire point cloud data from two aid drones simultaneously, and extract entity interference data of the target fire site based on the point cloud data from the two drones;

[0016] Combining non-physical interference data and physical interference data, an anti-jamming flight strategy for dual-aid UAVs is generated.

[0017] Optionally, the real-time status parameters include real-time ranging parameters and real-time pose parameters. The step of dynamically correcting the initial spatial vector baseline by combining the real-time status parameters, fire environment parameters, and environmental reference parameters to obtain a dynamic spatial vector baseline, and completing the dual-clock phase alignment verification of the dual-aid UAVs based on the dynamic spatial vector baseline, includes the following steps:

[0018] The real-time ranging parameters are scaled and compensated by combining fire scene environmental parameters and baseline environmental parameters.

[0019] The initial spatial vector baseline is dynamically corrected by combining the real-time ranging parameters and real-time pose parameters that have completed scaling compensation, and the dynamic spatial vector baseline is output based on the dynamic correction results.

[0020] Extract the baseline magnitude of the dynamic space vector baseline;

[0021] The acoustic desired time window is constructed by combining the baseline modulus and the pre-acquired reference acoustic propagation velocity;

[0022] Complete the bidirectional timestamp exchange between the primary and secondary aid drones, and calculate the link transmission delay parameters of the two aid drones based on the bidirectional timestamp exchange results.

[0023] Collect dual clock pulse signals from the dual-aid UAV, and construct the clock phase deviation sequence of the dual-aid UAV based on the dual clock pulse signals;

[0024] The clock deviation time series of the dual-aid UAVs was calculated by combining the link transmission delay parameters and the clock phase deviation sequence.

[0025] The clock synchronization and synchronization stability of the dual-aid UAVs were verified based on the clock deviation time series and the acoustic expectation time window, respectively. The dual clock phase alignment of the dual-aid UAVs was verified by combining the clock synchronization and synchronization stability verification results.

[0026] Optionally, the dual-tone audio signal from the fire scene can be filtered in multiple stages using a filtering algorithm to obtain a low-noise dual-tone audio signal, including the following steps:

[0027] Real-time acquisition of dual motor status information of dual-assistance drones;

[0028] The rotor harmonic frequencies of the dual-assistance UAVs were calculated based on the dual-motor status information.

[0029] The target defect filter is obtained by configuring the basic filtering parameters of the defect filter based on the rotor harmonic frequency. The basic filtering parameters include the filter order and attenuation depth parameters.

[0030] The target defect filter is used to filter the dual-tone signal of the fire scene to obtain the initial dual-tone signal;

[0031] The harmonic frequency range is determined based on the rotor harmonic frequency, and the rotor harmonic component and the residual noise component of the initial dual-tone signal of the fire scene are extracted based on the harmonic frequency range.

[0032] The residual harmonic noise rate of the initial dual-tone signal was calculated by combining the residual noise component and the rotor harmonic component.

[0033] If the residual harmonic noise rate is greater than the preset residual rate threshold, the basic filtering parameters of the defective filter are adjusted according to the residual harmonic noise rate, and the initial dual-tone signal is repeatedly filtered using the defective filter with the adjusted basic filtering parameters until the residual harmonic noise rate of the initial dual-tone signal is less than or equal to the preset residual rate threshold.

[0034] If the residual harmonic noise rate is less than or equal to the preset residual rate threshold, the rotor eddy current noise of the initial dual-tone signal is filtered using a filtering algorithm to obtain a low-noise dual-tone signal.

[0035] Optionally, the rotor vortex noise filtering of the initial dual-tone signal is performed using a filtering algorithm to obtain a low-noise dual-tone signal, including the following steps:

[0036] The stationarity of the initial dual-tone signal is detected by using the short-time autocorrelation function, and the initial stationary signal segment of the initial dual-tone signal is extracted based on the stationarity detection result.

[0037] Calculate the dual power spectral density parameters of the initial stationary signal segment;

[0038] The initial dual-tone signal is decomposed into frequency-point signal components, and the dual signal-to-noise ratio parameters of the frequency-point signal components are calculated.

[0039] The frequency domain gain compensation coefficient of the frequency point signal component is determined based on the dual signal-to-noise ratio parameters;

[0040] The frequency filtering coefficients of the frequency signal components are calculated by combining the dual power spectral density parameters and dual signal-to-noise ratio parameters;

[0041] By combining the frequency domain gain compensation coefficient and the frequency point filtering coefficient, the rotor eddy current noise of the initial dual-tone signal is filtered to obtain a low-noise dual-tone signal.

[0042] Optionally, the collaborative fusion of dual-modal acoustic data, including acoustic detection signals and low-noise dual-audio signals, to obtain the equivalent sound velocity of the target fire site includes the following steps:

[0043] Based on the pre-acquired acoustic wave characteristics of the transmitting end, acoustic wave distortion analysis is performed on the acoustic wave detection signal, and the arrival time of the first wave and the total phase offset of the acoustic wave detection signal are extracted based on the acoustic wave distortion analysis results.

[0044] The initial acoustic velocity was calculated by combining the arrival time of the first wave and the dynamic spatial vector baseline.

[0045] The initial acoustic wave velocity is corrected by using the total phase offset to obtain the detected acoustic wave velocity;

[0046] Cross-correlation algorithm is used to perform cross-correlation calculation on low-noise dual-tone signals point by point, and dual-tone cross-correlation curve is constructed based on the cross-correlation calculation results;

[0047] The peak values ​​of the cross-correlation curves of the two-tone audio frequencies are detected using a peak detection algorithm, and a candidate peak list is obtained.

[0048] Traverse the candidate peak list and calculate the theoretical propagation path length of all cross-correlation peaks based on the dynamic spatial vector baseline and the acoustic propagation principle;

[0049] The core effective peaks in the candidate peak list are selected based on the theoretical propagation path length;

[0050] The noise delay deviation between the two aid drones is determined based on the peak coordinate value corresponding to the core effective peak value in the dual-audio cross-correlation curve.

[0051] The passive noise velocity is calculated by combining the noise delay bias and the dynamic space vector baseline;

[0052] The equivalent sound velocity of the target fire site is obtained by weighted fusion of the velocity of the detected sound wave and the velocity of the passive noise.

[0053] Optionally, performing acoustic distortion analysis on the acoustic detection signal based on the pre-acquired acoustic wave characteristics of the transmitting end, and extracting the first arrival time and total phase offset of the acoustic detection signal based on the acoustic distortion analysis results includes the following steps:

[0054] The acoustic wave detection signal is converted from analog to digital using an analog-to-digital converter to obtain a digital acoustic wave signal;

[0055] The digital acoustic signal is demodulated using orthogonal demodulation technology to obtain a complex acoustic signal.

[0056] Based on the pre-acquired acoustic wave characteristics of the transmitting end, the acoustic wave distortion parameters of the complex signal are extracted, including the acoustic wave energy attenuation coefficient and the acoustic wave phase offset.

[0057] The amplitude of the digital acoustic signal is corrected by using the acoustic energy attenuation coefficient, resulting in a corrected acoustic signal.

[0058] An edge detection algorithm is used to detect the amplitude gradient of the acoustic wave correction signal, and the arrival time of the first wave of the acoustic wave correction signal is determined based on the amplitude gradient detection results.

[0059] Phase calibration of the complex acoustic signal is performed using the acoustic phase offset, and folded phase identification is performed on the phase-calibrated complex acoustic signal to obtain folded phase data.

[0060] Phase folding compensation is performed on the complex acoustic signal based on the folded phase data, and the total phase offset of the complex acoustic signal is extracted based on the phase folding compensation result.

[0061] Optionally, the simulation of the sound wave trajectory of the target fire scene based on the equivalent sound velocity of the fire scene and using Fermat's principle, and the inversion of non-physical interference data of the target fire scene based on the sound wave trajectory simulation results, includes the following steps:

[0062] Based on the dynamic operating space between the two aid drones and the equivalent sound speed at the fire site, a refraction distribution model was iteratively constructed using Fermat's principle.

[0063] The grid sound velocity of all grid cells in the dynamic grid space is calculated by combining the refraction distribution model and the sound wave energy attenuation coefficient.

[0064] The lattice sound velocity is mapped to lattice temperature and lattice density using the ideal gas sound velocity formula and the ideal gas law, respectively.

[0065] A raster state data field for dynamic raster space is constructed by combining raster temperature and raster density;

[0066] The refractive gradient components of all grid cells are calculated based on the refractive distribution model, and all grid cells are divided into interference interface grids and interference internal grids based on the refractive gradient components.

[0067] A region growing algorithm is used to perform connected component analysis on all grid cells. Based on the connected component analysis results and a surface reconstruction algorithm, all interfering interface grids are fitted to the initial interference space.

[0068] Multidimensional interference parameters of the initial interference space are extracted based on the grid state data field and refraction distribution model.

[0069] Non-physical interference regions within the dynamic grid space are marked based on multi-dimensional interference parameters, and the multi-dimensional interference parameters and non-physical interference regions are integrated into non-physical interference data of the target fire site.

[0070] Optionally, the refraction distribution model is constructed iteratively based on the dynamic operating space between the two aid drones and the equivalent sound velocity at the fire site, using Fermat's principle, and includes the following steps:

[0071] The dynamic operational space between the two aid drones is processed by rasterization to obtain a dynamic raster space;

[0072] Based on the equivalent sound velocity at the fire scene and using the ideal gas sound velocity formula, an initial refraction distribution model is simulated in the dynamic grid space.

[0073] Based on the initial refraction distribution model and using Fermat's principle, the simulation of sound wave propagation in a non-uniform medium was completed, and the sound wave bending trajectory data was obtained.

[0074] Calculate the predicted phase of the acoustic wave trajectory data, and combine the predicted phase of the acoustic wave with the total phase offset to calculate the trajectory phase residual;

[0075] The initial refraction distribution model is iteratively optimized based on the trajectory phase residual and using the fast travel method until the trajectory phase residual is less than the residual threshold, thus obtaining the refraction distribution model.

[0076] Optionally, acquiring point cloud data from two aid drones simultaneously, and extracting entity interference data of the target fire site based on the point cloud data from the two drones, includes the following steps:

[0077] Preprocess dual-machine point cloud data;

[0078] The preprocessed dual-machine point cloud data is spatiotemporally aligned with non-physical interference data to obtain synchronized point cloud data.

[0079] Based on the dynamic spatial vector baseline, dual-view point cloud stitching of synchronized point cloud data is completed to obtain stitched point cloud data;

[0080] Based on the non-entity interference data, the entity attributes of the merged point cloud data are marked in different regions to obtain the point cloud entity attributes;

[0081] Cluster analysis is performed on the stitched point cloud data based on the entity attributes of the point cloud, and the entity interference data of the target fire site is determined based on the cluster analysis results.

[0082] Secondly, this application provides an anti-interference system for unmanned aerial vehicle (UAV) flights in emergency scenarios, comprising:

[0083] The memory is configured to store instructions; and

[0084] The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the anti-interference method for UAV flight in emergency scenarios according to any one of the preceding statements.

[0085] The above technical solution provides an initial position reference for establishing two-way ranging by collecting environmental baseline parameters and UAV pose information before entering the target fire area. This avoids ranging errors caused by unknown initial positions, ensures the starting accuracy of the dual-UAV collaborative operation, and also provides an initial benchmark for subsequent dynamic correction of the spatial vector baseline. Two-way ranging interaction and the construction of the initial spatial vector baseline provide a spatial coordination basis for the subsequent synchronous collection of dual-audio-frequency signals and point cloud data from the fire area by the two support UAVs. After entering the target fire area, real-time collection of fire area environmental parameters and UAV real-time status parameters provides a correction basis for subsequent initial spatial vector baseline correction, and also provides environmental impact factors for dual-clock phase alignment verification. Then, dual-clock phase alignment verification solves the problem of timestamp mismatch in fire area dual-audio-frequency signals and dual-UAV point cloud data that may be caused by clock asynchrony between the two UAVs, providing a technical foundation for subsequent data analysis and fusion. Then, by collecting dual-audio-frequency (DET) signals from the fire scene, core data is provided for subsequent identification of non-physical interference types and location of interference. Simultaneously, the DET signals are filtered to effectively remove strong noise generated by the rotor rotation and motor operation of the dual-assistance UAV, preventing it from masking the real acoustic signals related to the fire scene, resulting in low-noise DET signals. Next, the low-noise DET signals are fused with acoustic wave detection signals to obtain the equivalent sound velocity of the fire scene, providing accurate input data for subsequent acoustic trajectory simulation. Then, acoustic trajectory simulation is performed using the equivalent sound velocity of the fire scene, transforming intangible interference into quantifiable data such as temperature and density distribution, providing clear interference targets for strategy generation. At the same time, dual-UAV point cloud data is used to capture physical interference in the target fire scene, distinguishing between non-physical and physical interference. This lays the data foundation for developing differentiated anti-interference flight strategies for different interference types, ensuring stable flight and continuous mission operation of the UAV in complex fire scenes, and preventing flight instability, crashes, or mission interruptions caused by interference. In summary, this application effectively improves the anti-interference capability of UAVs in complex fire environments by accurately identifying non-physical and physical interference in the target fire scene and formulating differentiated anti-interference flight strategies accordingly, thus ensuring the smooth execution of fire rescue missions.

[0086] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0087] Figure 1 A flowchart illustrating an anti-interference method for UAV flight in an emergency scenario, provided as an embodiment of this application;

[0088] Figure 2 A flowchart illustrating a multi-stage filtering method for dual-tone signals in a fire scene, provided as an embodiment of this application;

[0089] Figure 3This is a flowchart illustrating a non-physical interference data inversion method provided in an embodiment of this application. Detailed Implementation

[0090] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0091] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0092] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0093] Figure 1 The illustration schematically shows a flowchart of an anti-interference method for unmanned aerial vehicle (UAV) flight in an emergency scenario according to an embodiment of this application. Figure 1 As shown in the figure, this application provides a method for anti-interference of UAV flight in emergency scenarios, applied to dual-aid UAVs, wherein the dual-aid UAVs include a primary aid UAV and a secondary aid UAV. The method may include the following steps:

[0094] S101. Before the dual-aid drones enter the target fire area, collect environmental baseline parameters and drone pose information of the dual-aid drones.

[0095] In this embodiment, the baseline environmental parameters refer to parameters such as temperature, humidity, air pressure, and smoke concentration under standard atmospheric conditions, collected by multiple types of sensors pre-loaded on the dual-aid UAVs before entering the target fire area. Simultaneously with the acquisition of these baseline environmental parameters, the attitude acquisition equipment of the dual-aid UAVs, including an IMU (Inertial Measurement Unit), GPS positioning module, and visual positioning module, is activated. The dual-aid UAVs are controlled to enter a stable hovering state, and with a sampling period of 5ms, the attitude information of the primary and secondary aid UAVs is simultaneously acquired. This includes three-dimensional position coordinates (based on the geodetic coordinate system: east x, north y, vertical altitude z) and three-dimensional attitude angles, including roll, pitch, and yaw angles. During the acquisition process, timestamps are synchronized via a low-latency communication link between the two UAVs to ensure complete consistency in the acquisition time of each frame of attitude data. The acquired attitude information of the primary and secondary aid UAVs is then integrated into the overall UAV attitude information of the dual-aid UAVs. In addition, the emergency scenarios in this application refer to scenarios where sudden safety accidents such as fires, chemical leaks, and explosions occur, and complex and harsh environments such as high temperatures, dense smoke, strong turbulence, and building collapses exist, requiring emergency rescue, disaster reconnaissance, and other emergency response tasks. In particular, these include typical fire rescue scenarios such as forest fires, urban high-rise building fires, chemical industrial park fires, and tunnel fires, which are collectively referred to as target fire sites in this application.

[0096] S102. Based on the UAV pose information, complete the two-way ranging interaction between the main assistance UAV and the slave assistance UAV, and construct the initial spatial vector baseline of the two assistance UAVs based on the two-way ranging interaction results.

[0097] In this embodiment, based on the initial pose reference data stored by both drones, the primary assistance drone and the secondary assistance drone establish a low-latency wireless communication link (such as an ultra-wideband (UWB) link) and agree on a two-way ranging interaction protocol: the format of the ranging request signal and response signal is clearly defined, including device identifier, timestamp, and pose reference information. The ranging trigger period can be set to 20ms to ensure compatibility with the pose data acquisition period. Simultaneously, based on the relative distance range in the initial pose data, the transmit power and receive sensitivity of the ranging module are configured to avoid measurement errors caused by excessively strong or weak ranging signals. The primary assistance drone sends a ranging request signal at the agreed-upon period, containing the primary assistance drone's current precise pose information and a transmission timestamp T1. Upon receiving the request signal, the secondary assistance drone immediately records the reception timestamp T2 and simultaneously sends a response signal containing its own precise pose information, reception timestamp T2, and transmission timestamp T3. Upon receiving the response signal, the primary assistance drone records the reception timestamp T4, completing one full two-way ranging interaction. To ensure data reliability, multiple sets of two-way ranging interactions were continuously executed, and the arithmetic mean of the obtained two-way ranging data was taken to obtain the initial relative distance reference value between the two aid UAVs. This value reflects the true physical distance between the two UAVs under interference-free conditions.

[0098] A local coordinate system is established with the three-dimensional position coordinates of the primary support UAV as the origin. This local coordinate system maintains the same attitude as the geodetic coordinate system, differing only in its origin. The three-dimensional position coordinates of the secondary support UAV are transformed into this local coordinate system to obtain the relative three-dimensional coordinates of the secondary support UAV relative to the primary support UAV. These relative three-dimensional coordinates directly reflect the relative positional relationship between the two UAVs in the initial state and are the core component of the initial spatial vector baseline. Next, the relative three-dimensional coordinates are integrated into a three-dimensional spatial vector. The direction of this vector is from the primary support UAV to the secondary support UAV, and the magnitude of the vector is the initial relative distance reference value obtained from two-way ranging. Simultaneously, the initial attitude angle information of both support UAVs—roll angle, pitch angle, and yaw angle—is bound together to clarify the relative attitude relationship between the two UAVs in the initial state, ultimately forming the initial spatial vector baseline. This baseline completely contains all information regarding the initial relative position and attitude of the two UAVs.

[0099] This step is completed outside the fire interference zone to ensure that the collected baseline data is free from various fire interferences, providing an accurate reference benchmark for all subsequent dynamic corrections and data verifications. It is the fundamental premise of the entire anti-interference system.

[0100] S103. When the dual-aid drones enter the target fire site, the fire site environmental parameters and the real-time status parameters of the dual-aid drones are collected in real time.

[0101] In this embodiment, when the dual-aid UAVs enter the target fire area, the fire environment parameters of the target fire area are collected by multiple types of sensors pre-loaded on the dual-aid UAVs. These parameters include real-time temperature, real-time humidity, real-time smoke concentration, and real-time air pressure. The multiple types of sensors include temperature sensors and humidity sensors. Simultaneously, real-time status parameters are collected, including real-time ranging parameters and real-time pose parameters. The real-time ranging parameters refer to the distance between the primary and secondary aid UAVs obtained through laser ranging, ultra-wideband (UWB) ranging, or other methods. The method and equipment for collecting the real-time pose parameters are consistent with the method and equipment for collecting UAV pose information.

[0102] S104. Combine real-time status parameters, fire environment parameters, and environmental reference parameters to dynamically correct the initial spatial vector baseline, obtain the dynamic spatial vector baseline, and complete the dual-clock phase alignment verification of the dual-aid UAVs based on the dynamic spatial vector baseline.

[0103] In this embodiment, the environmental deviation between the fire environment parameters and the reference environmental parameters is calculated. The real-time ranging parameters in the real-time state parameters are then corrected based on the environmental deviation to obtain the corrected ranging parameters. The initial spatial vector baseline is the three-dimensional spatial vector at the initial deployment of the dual-aid UAVs. Since the flight attitude and relative position of the dual-aid UAVs continuously change when performing fire assistance missions in the target fire area, it is necessary to double-correct the initial spatial vector baseline using both the corrected ranging parameters and the real-time attitude parameters to obtain a dynamic spatial vector baseline that conforms to the dynamic changes of the fire area. The baseline modulus of the initial spatial vector baseline is updated based on the corrected ranging parameters, and the relative attitude relationship of the dual-aid UAVs based on the initial spatial vector baseline is updated based on the real-time attitude parameters. Next, the baseline modulus of the dynamic spatial vector baseline is extracted, and the baseline modulus is multiplied by the reference acoustic propagation speed pre-stored in the airborne system to obtain the theoretical reference time for sound wave propagation between the two UAVs. Then, the upper and lower limits of the time window are calculated based on the theoretical reference time. Using the theoretical reference time as the center and the upper and lower limits as boundaries, an acoustic expected time window is constructed and output. This time window serves as the core time determination basis for subsequent clock synchronization and stability verification. Next, the link transmission delay parameters are extracted through information exchange between the primary and secondary support UAVs. The clock pulse acquisition module of both support UAVs is activated to collect their clock pulse signals. The initial deviation time series of the secondary support UAV is extracted using the link transmission delay parameters. The initial deviation time series is then corrected using the link transmission delay parameters to obtain the clock deviation time series.

[0104] The upper and lower limits of the acoustic desired time window are extracted and used as the threshold range for clock synchronization. Each clock time deviation value in the clock deviation time series is iterated through, and each deviation value is checked to see if it falls within the threshold range. The proportion of deviation values ​​falling within the threshold range to the total number of deviation values ​​is calculated to obtain the synchronization compliance rate. If the synchronization compliance rate is greater than or equal to the preset compliance rate threshold, the clock synchronization verification is considered passed; otherwise, it fails. Next, the slope of the trend line for all clock time deviation values ​​in the clock deviation time series is calculated, and synchronization stability is verified based on the trend line slope. If both the synchronization stability verification and the clock synchronization verification pass, the dual-clock phase alignment verification of the dual-aid UAV is considered passed. Otherwise, if either the synchronization stability verification or the clock synchronization verification fails, dual-clock alignment needs to be performed again until both verifications pass.

[0105] By constructing a dynamic spatial vector baseline, the spatial reference of the two drones can follow the changes in the fire environment and the drone's own status in real time, solving the problem that static baselines cannot adapt to dynamic interference in the fire scene. This maintains the relative positioning accuracy of the two drones within an effective range and avoids errors in subsequent acoustic acquisition and point cloud stitching caused by spatial deviations. Dual clock phase alignment is a prerequisite for all subsequent synchronous acquisition operations, avoiding mismatches in audio and point cloud timestamps caused by clock asynchrony. It also solves the problems of invalid acquisition data and fusion deviation caused by time domain interference, ensuring that the multi-source data acquired by the two drones are accurately aligned on the time axis.

[0106] S105. If the dual clock phase alignment verification of the dual-aid UAV passes, the dual-audio signal of the target fire site is collected using the microphone array carried by the dual-aid UAV.

[0107] In this embodiment, after confirming that the dual-clock phase alignment verification is successful, the microphone arrays on both drones are activated to synchronously acquire acoustic signals from the target fire site, forming a dual-audio signal. The dual-audio signal includes the main audio signal acquired by the main support drone and the secondary audio signal acquired by the secondary support drone. Simultaneously, the timestamps of the dual-audio signals are synchronized via a low-latency link between the two drones to ensure that the acquisition times of the two drones are completely consistent, avoiding subsequent data distortion caused by time differences.

[0108] S106. Use a filtering algorithm to perform multi-stage filtering on the dual-tone signal of the fire scene to obtain a low-noise dual-tone signal.

[0109] In this embodiment, the basic filtering parameters of the defective filter are first configured based on the rotor harmonic frequencies of the dual-tone audio signal from the fire scene. The defective filter with the configured basic filtering parameters is then used for initial filtering to obtain an initial dual-tone audio signal. If the residual harmonic noise rate of the initial dual-tone audio signal is greater than a preset residual rate threshold, the basic filtering parameters of the defective filter are adjusted according to the residual harmonic noise rate, i.e., the filter order and attenuation depth parameters are increased, and the initial dual-tone audio signal is filtered again until the residual harmonic noise rate of the initial dual-tone audio signal is less than or equal to the preset residual rate threshold. Next, rotor eddy current noise filtering is applied to the initial dual-tone audio signal after the initial filtering to obtain a low-noise dual-tone audio signal. This completely eliminates the masking effect of the UAV's own rotor noise on the acoustic signal from the fire scene, solves the problem of difficulty in extracting effective signals due to its own interference, and obtains a pure low-noise dual-tone audio signal, providing a high-quality data foundation for subsequent dual-modal acoustic fusion.

[0110] S107. Acquire the acoustic detection signal between the main assistance drone and the slave assistance drone, and complete the dual-modal acoustic data collaborative fusion of the acoustic detection signal and the low-noise dual-audio signal to obtain the equivalent sound velocity of the target fire site.

[0111] In this embodiment, the velocity of the detected sound wave and the velocity of the passive noise are calculated using the acoustic detection signal and the low-noise dual-tone signal, respectively. Then, the velocity of the detected sound wave and the velocity of the passive noise are weighted and fused to obtain the equivalent sound velocity of the target fire. This step solves the problem of velocity calculation errors caused by the susceptibility of a single acoustic mode to interference from the fire environment medium, resulting in an accurate and robust equivalent sound velocity of the fire. This provides core acoustic parameters for subsequent identification of non-physical interference and improves the acoustic detection's resistance to medium interference. The acoustic detection signal is susceptible to signal distortion, phase shift, and energy attenuation caused by non-uniform fire environment media, such as high-temperature heat flow and gas plumes. Passive fire noise, i.e., the effective acoustic signal of the fire, utilizes the fire's own acoustic signal and has stronger resistance to medium attenuation and distortion. The two complement each other, forming a core basis for subsequent acoustic trajectory simulation and inversion of non-physical interference based on Fermat's Last Theorem, avoiding errors in subsequent non-physical interference identification due to inaccurate acoustic parameters.

[0112] S108. Based on the equivalent sound velocity of the fire field and using Fermat's principle, the sound wave trajectory of the target fire field is simulated, and the non-physical interference data of the target fire field is derived from the sound wave trajectory simulation results.

[0113] In this embodiment, a dynamic grid space is first constructed based on the dynamic operational space between the two support drones. Then, the corresponding equivalent sound velocity of the fire field is assigned to all grid cells within the dynamic grid space. Next, the acoustic refraction coefficient of all grid cells is calculated. The acoustic refraction coefficient is mapped to each grid cell, and a corresponding acoustic refraction coefficient is matched to each grid cell, generating a continuous spatial distribution model of the refraction coefficient within the dynamic grid space, resulting in an initial refraction distribution model. Then, the initial refraction distribution model is iteratively optimized using Fermat's principle to obtain a refraction distribution model. The final refraction coefficient of the grid cells within the refraction distribution model is extracted. Then, based on the final refraction coefficient, the initial grid sound velocity, grid temperature, and grid density corresponding to each grid cell are calculated, forming a grid state data field for the dynamic grid space. The non-physical interference regions within the dynamic grid space are marked using the refraction distribution model and the grid state data field, and based on this, the non-physical interference regions and their corresponding multi-dimensional interference parameters are integrated into non-physical interference data for the target fire field.

[0114] By iteratively constructing a refraction distribution model using the equivalent sound velocity in the fire field and Fermat's principle, the propagation trajectory of sound waves in the non-uniform medium of the fire field is simulated. The physical characteristics of non-physical disturbances in the fire field, such as high-temperature heat flow, gas jets, non-uniform medium density, and airflow turbulence, namely temperature density, refractive index, and spatial range, are transformed into quantifiable grid state data, thus realizing accurate perception of the root causes of non-physical disturbances in the fire field.

[0115] S109. Acquire point cloud data from the two aid drones simultaneously, and extract entity interference data of the target fire site based on the point cloud data from the two drones.

[0116] In this embodiment, multiple entity clusters are clustered based on dual-machine point cloud data. For each cluster, spatial and geometric features are extracted, including spatial centroid coordinates, bounding box range, overall size, volume, height, distribution pattern, and occlusion range. Then, based on preset rules, it is determined whether they constitute entity interference. For example, entity clusters located in the UAV operation path, rescue channel, or core fire area, and whose volume exceeds the minimum interference threshold or whose height exceeds the safe flight altitude, forming occlusion or dangerous obstacles, are identified as valid entity interference targets. Entity clusters located in safe areas, with excessively small volumes, on flat terrain, or with low-lying harmless vegetation are identified as non-interference entities and are removed. For multiple spatially adjacent clusters belonging to the same continuous obstacle, such as continuous walls, large areas of trees, or continuously collapsed structures, a merging process is performed to obtain entity obstacle entities. Their spatial and geometric features are then integrated into the final entity interference data for the target fire area, providing accurate solid obstacle data for dual-aid UAV obstacle avoidance, flight path planning, and safe fire operation.

[0117] S110: Combine non-physical interference data and physical interference data to generate an anti-jamming flight strategy for dual-aid UAVs.

[0118] In this embodiment, core parameters of non-physical interference data are extracted, including spatial centroid coordinates, coverage area, average grid sound velocity, average grid temperature, average density, and refractive index, as well as key features of physical interference data, including spatial location, bounding box size, volume, occlusion range, and geometry. These core parameters and key features are then uniformly mapped to the local coordinate system of the dual-drone collaboration, forming a "global interference data map" that visually presents the spatial overlap and correlation of various types of interference. Based on the global interference data map, an anti-interference flight strategy for the dual-aid UAV is generated. For example, for areas with high average temperatures, such as ≥500℃, a detour path is planned based on the coverage area and spatial shape of the high-temperature zone to ensure a safe distance of ≥10m between the dual drones and the boundary of the high-temperature zone; simultaneously, the flight altitude is increased by 5-8m to avoid the near-ground high-temperature airflow accumulation layer, reducing the risk of the fuselage being overheated. For regions where the grid sound velocity fluctuation exceeds a preset threshold (e.g., greater than 5 m / s) or the temperature fluctuation exceeds a preset temperature fluctuation threshold (e.g., greater than 20 degrees Celsius), the physical parameters of these regions change abruptly due to rapid airflow disturbances, forming a jet stream region. The airflow direction can be predicted based on the refractive gradient components of the jet stream region. If the jet stream is vertically upward, the flight altitude should be lowered by 3-5 meters to avoid the core disturbance region; if the jet stream is horizontal, the flight speed should be adjusted to match the airflow speed to reduce the impact of the airflow on the fuselage. For regions where the total refractive index gradient exceeds a preset gradient threshold (e.g., 0.0003 m⁻¹), there is no significant dynamic airflow disturbance, but the temperature and density exhibit spatial gradient distribution, leading to changes in acoustic refraction patterns. In these regions, it is necessary to maintain the continuity of the flight path and avoid frequent changes in direction. Furthermore, when facing large physical obstacles (e.g., ≥50 m³), ​​a detour strategy can be adopted. The shortest detour path should be planned based on the obstacle's geometry (e.g., cuboid, irregular collapse), with a detour radius ≥ 1.5 times the maximum size of the obstacle.

[0119] Through the above steps, specific response strategies can be formulated for different types of interference. For non-physical interference, such as high-temperature heat flow areas and gas jet areas, strategies such as detouring, altitude adjustment, and reducing flight speed are adopted to avoid flight instability caused by fuselage overheating and airflow disturbance. For physical interference, such as building debris and collapsed objects, strategies such as precise obstacle avoidance, flight path replanning, and maintaining a safe distance between two drones are adopted to avoid collision damage. In summary, this application effectively solves the problems of flight instability, mission interruption, and drone crashes caused by various types of interference such as high temperature, dense smoke, turbulence, physical obstacles, and self-noise in emergency scenarios such as fire scenes, significantly improving the flight stability, anti-interference capability, and continuity of rescue missions of drones in complex emergency scenarios.

[0120] In one embodiment, the real-time status parameters include real-time ranging parameters and real-time pose parameters. The step of dynamically correcting the initial spatial vector baseline by combining the real-time status parameters, fire environment parameters, and environmental reference parameters to obtain a dynamic spatial vector baseline, and completing the dual-clock phase alignment verification of the dual-aid UAVs based on the dynamic spatial vector baseline, includes the following steps:

[0121] The real-time ranging parameters are scaled and compensated by combining fire scene environmental parameters and baseline environmental parameters.

[0122] The initial spatial vector baseline is dynamically corrected by combining the real-time ranging parameters and real-time pose parameters that have completed scaling compensation, and the dynamic spatial vector baseline is output based on the dynamic correction results.

[0123] Extract the baseline magnitude of the dynamic space vector baseline;

[0124] The acoustic desired time window is constructed by combining the baseline modulus and the pre-acquired reference acoustic propagation velocity;

[0125] Complete the bidirectional timestamp exchange between the primary and secondary aid drones, and calculate the link transmission delay parameters of the two aid drones based on the bidirectional timestamp exchange results.

[0126] Collect dual clock pulse signals from the dual-aid UAV, and construct the clock phase deviation sequence of the dual-aid UAV based on the dual clock pulse signals;

[0127] The clock deviation time series of the dual-aid UAVs was calculated by combining the link transmission delay parameters and the clock phase deviation sequence.

[0128] The clock synchronization and synchronization stability of the dual-aid UAVs were verified based on the clock deviation time series and the acoustic expectation time window, respectively. The dual clock phase alignment of the dual-aid UAVs was verified by combining the clock synchronization and synchronization stability verification results.

[0129] In this embodiment, fire scene environmental parameters refer to parameters such as real-time temperature, real-time humidity, real-time smoke concentration, and real-time air pressure of the target fire scene. Baseline environmental parameters refer to parameters such as temperature, humidity, air pressure, and smoke concentration under standard atmospheric conditions collected by multiple types of sensors pre-loaded on the dual-aid UAVs before entering the target fire scene. Real-time ranging parameters refer to the initial distance values ​​obtained by the primary and secondary aid UAVs through laser ranging, ultra-wideband (UWB) ranging, etc., when there is no environmental interference, serving as the original basis for ranging correction. Next, based on calibration data from laboratory fire scene simulation experiments, a pre-stored table of single-environment parameter ranging influence coefficients is retrieved. This table includes the influence coefficients of changes in temperature, air pressure, humidity, and smoke concentration on the ranging results. For example, for every 10°C increase in temperature, the laser ranging value deviation increases by 0.2%; for every 10% increase in smoke concentration, the UWB ranging value deviation decreases by 0.15%. The environmental deviation values ​​between the fire scene environmental parameters and the baseline environmental parameters are calculated, including temperature difference, humidity difference, air pressure difference, and smoke concentration difference. Based on the environmental deviation value, the corresponding single-parameter influence coefficient is matched in the single-environment parameter ranging influence coefficient table. If the parameter deviation exceeds the calibration range, interpolation is used to calculate and supplement the corresponding influence coefficient in the single-environment parameter ranging influence coefficient table. The influence coefficients of each single environmental parameter are weighted and fused to obtain the comprehensive environmental scaling compensation coefficient. The weighting weights are set according to the degree of influence of each parameter in the fire scene environment on ranging. For example, temperature and air pressure are the main influencing factors in the fire scene, each accounting for 35% of the weight; humidity and smoke concentration are secondary factors, each accounting for 15% of the weight. The product between the real-time ranging parameter and the comprehensive environmental scaling compensation coefficient is calculated to obtain the corrected ranging parameter. This parameter eliminates the interference of the fire scene environment medium on ranging, reflects the true physical distance of the dual-assistance UAV in the target fire scene, and provides the core distance basis for subsequent spatial vector baseline correction.

[0130] The initial spatial vector baseline is the three-dimensional spatial vector at the initial deployment of the dual-aid UAVs. Since the flight attitude and relative position of the dual-aid UAVs continuously change during their fire relief missions, it is necessary to double-correct the initial spatial vector baseline using both range-measuring parameters and real-time attitude parameters to obtain a dynamic spatial vector baseline that conforms to the dynamic changes of the fire. The baseline magnitude of the initial spatial vector baseline is updated based on the corrected range-measuring parameters, and the relative attitude relationship of the dual-aid UAVs is updated based on the real-time attitude parameters.

[0131] The baseline modulus of the dynamic spatial vector baseline, i.e., the Euclidean distance between the two support UAVs, is extracted. Multiplying this baseline modulus by the pre-stored reference acoustic propagation speed in the onboard system yields the theoretical reference time for sound wave propagation between the two UAVs. Then, considering the complexity of the environment, such as fire turbulence and smoke disturbance, upper and lower deviation thresholds for the desired acoustic time window are set. These thresholds are adjusted according to the degree of disturbance in the fire environment, typically ranging from ±3% to ±5% of the theoretical reference time; a midpoint can be used. This threshold reflects the allowable time fluctuation range for sound wave propagation in the fire. Next, combining the theoretical reference time and the set deviation thresholds, the upper and lower limits of the time window are calculated: upper limit = theoretical reference time × (1 + deviation threshold), lower limit = theoretical reference time × (1 - deviation threshold). Using the theoretical reference time as the center and the upper and lower limits as boundaries, the desired acoustic time window is constructed and output. This time window serves as the core time determination basis for subsequent clock synchronization and stability verification.

[0132] The primary support drone generates a data packet with its own timestamp T1 and sends it to the secondary support drone via a low-latency wireless link. Upon receiving the data packet, the secondary support drone immediately records the received timestamp T2 without delay and generates a response data packet containing T2 and its own current timestamp T3, which it then sends back to the primary support drone. Upon receiving the response data packet, the primary support drone immediately records the received timestamp T4, completing one complete bidirectional timestamp exchange. After each exchange, the data packets are verified; if packet loss or out-of-order delivery occurs, retransmission is immediately triggered to ensure the integrity of the timestamp data. Based on the bidirectional timestamp values, the one-way link transmission delay from the primary support drone to the secondary support drone and from the secondary support drone to the primary support drone are calculated. This calculation method eliminates the influence of the initial deviation of the local clocks of the two drones, retaining only the actual link transmission delay. The one-way link transmission delays obtained from multiple bidirectional exchanges are statistically analyzed, and their arithmetic mean is calculated to obtain the link transmission delay parameter.

[0133] Activate the clock pulse acquisition module of the dual-aid UAV to acquire the core clock pulse signal of the dual-aid UAV. Once the dual clock pulse signal is obtained, a hard synchronization triggering method can be used, with the synchronization trigger signal of the low-latency link of the dual-aid UAV as the reference, to ensure that the acquisition time of the dual-aid UAV is completely synchronized. The sampling period can be set to 1ms, consistent with the link transmission delay acquisition and clock parameter calculation period.

[0134] The acquired dual-machine clock pulse signals are preprocessed. An edge detection algorithm is used to accurately extract the rising edge feature points of each clock pulse, and the corresponding local clock timestamp is recorded. The amplitude and frequency of the pulse signals are standardized to eliminate pulse characteristic deviations caused by hardware differences between the two machines. The rising edge feature points of the dual-machine clock pulse signals at the same acquisition moment are compared, and the time difference between the rising edges of the two machines is calculated. This time difference is then converted into a single-moment clock phase deviation value using the formula: Δφ = Δt × f × 2π, where Δt is the time difference and f is the natural frequency of the dual-machine clock pulses. The single-moment clock phase deviation values ​​from consecutive acquisition moments are arranged chronologically to obtain the initial clock phase deviation sequence. If there are missing values ​​in the initial clock phase deviation sequence, linear interpolation can be used to fill in the missing values ​​based on the deviation values ​​of the preceding and following moments. Simultaneously, the 3σ criterion is used to remove outliers from the initial clock phase deviation sequence. These outliers are mostly caused by electromagnetic interference from the fire scene or momentary malfunctions of the acquisition module. The initial clock phase deviation sequence after outlier removal is smoothed by using a moving average method, averaging the phase deviation values ​​at 3 to 5 consecutive times to eliminate small fluctuations caused by random noise, thus obtaining a clock phase deviation sequence that can truly reflect the stable trend of the dual-machine clock phase.

[0135] Each phase deviation value in the clock phase deviation sequence is converted into its corresponding original time deviation value. The converted original time deviation values ​​are then checked for consistency, verifying whether the changes in time deviation values ​​between adjacent moments conform to the physical laws of clock drift, for example, a rate of change not exceeding 0.01 ns / ms. If a large jump occurs, it is determined to be an invalid value and replaced with the average of adjacent valid values ​​to ensure the physical rationality of the sequence. The original time deviation values ​​obtained after the above steps are arranged in chronological order to obtain the initial deviation time sequence.

[0136] Next, the link transmission delay parameter is used to correct the initial deviation time series. If the link transmission delay parameter for the signal sent from the primary support UAV to the secondary support UAV during timestamp exchange is τ1, and for the reverse it is τ2, then τ = (τ1 + τ2) / 2 is taken as a unified compensation value to obtain the link transmission delay compensation value, ensuring that the compensation direction is consistent with the signal transmission direction of the phase acquisition. The link transmission delay compensation value is then subtracted from each original time deviation value in the initial deviation time series to obtain the clock deviation time series.

[0137] Extract the upper and lower limits of the acoustic expected time window and use this range as the threshold range for clock synchronization. Iterate through each clock time deviation value in the clock deviation time series, checking whether each value falls within the threshold range. Calculate the proportion of deviation values ​​falling within the threshold range to the total number of deviation values ​​to obtain the synchronization compliance rate. If the compliance rate is greater than or equal to a preset threshold (e.g., 95%, which can be adjusted according to fire scene perception accuracy requirements), the clock synchronization verification is considered passed. If the compliance rate is less than the preset threshold, the synchronization verification is considered failed, and the time and magnitude of deviation values ​​exceeding the threshold are marked for subsequent analysis of the reasons for deviation exceeding the standard, such as whether environmental interference exists. Calculate all clock time deviation values ​​in the clock deviation time series for trend detection. Use linear fitting to analyze the changing trend of the clock time deviation values. If the slope of the fitted trend line is less than or equal to a preset slope threshold (e.g., 0.01 μs / ms), it indicates no significant trend drift, and the synchronization stability verification is considered passed. If the slope of the trend line is greater than the preset slope threshold, the synchronization stability verification is considered failed. If both the synchronization stability check and the clock synchronization check pass, the dual-clock phase alignment check of the dual-aid UAV is considered successful. Conversely, if either the synchronization stability check or the clock synchronization check fails, dual-clock alignment needs to be performed again until both checks pass.

[0138] In one embodiment, reference is made to Figure 2 The process of using a filtering algorithm to perform multi-stage filtering on the dual-tone audio signal from the fire scene to obtain a low-noise dual-tone audio signal includes the following steps:

[0139] S201. Real-time acquisition of dual motor status information of dual-assistance UAVs;

[0140] S202. Calculate the rotor harmonic frequency of the dual-assistance UAV based on the dual-motor status information;

[0141] S203. Configure the basic filtering parameters of the defect filter according to the rotor harmonic frequency to obtain the target defect filter. The basic filtering parameters include the filter order and attenuation depth parameters.

[0142] S204. Use the target defect filter to filter the dual-tone signal of the fire scene and obtain the initial dual-tone signal.

[0143] S205. Determine the harmonic frequency range based on the rotor harmonic frequency, and extract the rotor harmonic component and the residual noise component of the initial dual-tone signal of the fire scene based on the harmonic frequency range.

[0144] S206. Calculate the residual harmonic noise rate of the initial dual-tone signal by combining the residual noise component and the rotor harmonic component.

[0145] S207. If the residual harmonic noise rate is greater than the preset residual rate threshold, the basic filtering parameters of the defective filter are adjusted according to the residual harmonic noise rate, and the initial dual-tone signal is repeatedly filtered using the defective filter with the adjusted basic filtering parameters until the residual harmonic noise rate of the initial dual-tone signal is less than or equal to the preset residual rate threshold.

[0146] S208. If the residual harmonic noise rate is less than or equal to the preset residual rate threshold, the rotor eddy current noise of the initial dual-tone signal is filtered using a filtering algorithm to obtain a low-noise dual-tone signal.

[0147] In this embodiment, the flight control system and ESC module of the dual-aid UAVs are used to collect the core motor operation status information of the main aid UAV and the slave aid UAV in real time. The core information includes parameters directly related to rotor rotation, such as instantaneous motor speed, drive current, PWM duty cycle, and motor load factor. During the acquisition process, the timestamps are precisely aligned through the dual-machine low-latency communication link, and the sampling frequency is matched with the acquisition frequency of the dual audio signals in the fire scene. This ensures that the timing of the motor status information and the audio signals corresponds one-to-one, avoiding the failure of subsequent filter parameter adaptation due to timing deviation. From the collected dual-motor status information, the real-time rotational speed (in RPM) of the motors of the dual-aid UAV is extracted. The number of fixed blades of the UAV rotor is retrieved from the dual-aid UAV's production manual. First, based on the core formula for blade passing frequency: rotor fundamental frequency = motor real-time rotational speed × number of rotor blades / 60, the fundamental frequency of a single rotor is calculated. Then, based on the fundamental frequency, the first to Nth order higher harmonic frequencies are calculated. The Nth order higher harmonic frequency = fundamental frequency × n. The rotor noise is mainly concentrated at the fundamental frequency and harmonics. The order of higher harmonics is determined based on the audio acquisition frequency band of the fire scene. Finally, the fundamental frequencies and higher harmonic frequencies of all rotors of the dual-aid UAV are integrated to form the rotor harmonic frequencies of the dual-aid UAV, which serve as the core frequency basis for subsequent filtering.

[0148] Based on the rotor harmonic frequencies, configure the basic filtering parameters of the defect filter. If the rotor harmonic frequency is single and the corresponding frequency band is narrow (i.e., the total number of rotor harmonic frequencies is ≤3, and the interval ratio of any adjacent harmonic frequencies is ≥5%), configure a low order filter, such as 2nd-4th order, to ensure filtering accuracy while taking into account real-time performance. If the number of rotor harmonic frequencies is large (i.e., the total number of rotor harmonic frequencies is >3, or the interval ratio of any adjacent harmonic frequencies is <3%), and there is slight overlap between frequency bands, configure a high order filter, such as 6th-8th order, to improve the frequency selectivity of the filter and avoid filtering out effective acoustic signals from the fire scene that are adjacent to the harmonic frequencies. Next, the rotor noise intensity is determined from the motor status information. This is because the higher the motor speed and the larger the load factor, the stronger the rotor noise. If the noise intensity is high, i.e., the real-time motor speed is >1500 RPM or the motor load factor is >0.7, a large attenuation depth of 40-60dB is configured to strongly suppress harmonic noise. If the noise intensity is low, i.e., the real-time motor speed is ≤1500 RPM and the motor load factor is ≤0.7, a moderate attenuation depth of 20-30dB is configured to reduce the impact on the effective signal at the fire scene. The configured filter order and attenuation depth parameters are imported into the defect filter and the parameters are solidified to obtain the target defect filter that is adapted to the current dual-rotor operating status.

[0149] The dual-audio signals of the fire scene collected by the main aid drone and the secondary aid drone through the microphone array are synchronously input into the target defect filter for filtering to obtain the initial dual-audio signal. The dual-audio signal of the fire scene contains a mixed signal of rotor harmonic noise, rotor vortex noise and effective acoustic signal of the fire scene. The initial dual-audio signal refers to the signal that has been greatly removed from the rotor harmonic noise, leaving only a small amount of harmonic noise and rotor vortex noise.

[0150] Taking each frequency in the rotor harmonic frequency set as the center, a narrow bandwidth of 1%-2% of the center frequency is extended to both sides to form an independent frequency interval corresponding to each harmonic frequency. Integrating all independent intervals yields the harmonic frequency interval. The dual-tone signal from the fire scene is decomposed in the frequency domain, and the signal components falling within the harmonic frequency interval are precisely extracted; these components are the rotor harmonic components. The initial dual-tone signal obtained after filtering is then decomposed in the frequency domain with the same resolution and frequency band as the fire scene dual-tone signal, and the remaining signal components falling within the harmonic frequency interval are extracted; these components are the residual noise components that were not completely filtered out by the target defect filter.

[0151] The total energy of the rotor harmonic components and the total energy of the residual noise components are calculated using frequency domain signal power integration, ensuring consistent dimensions and standards for the energy calculation of both components. Then, the residual harmonic noise rate of the initial dual-tone signal is calculated using the formula: Residual Harmonic Noise Rate = (Total Energy of Residual Noise Components / Total Energy of Rotor Harmonic Components) × 100%. This value directly reflects the filtering effect of the defective filter on rotor harmonic noise; a higher value indicates a worse filtering effect and more residual harmonic noise. If the residual harmonic noise rate exceeds a preset residual rate threshold, the basic filtering parameters of the defective filter are adjusted accordingly. This involves increasing the filter order and attenuation depth parameters. Specifically, the filter order can be increased by one order, and the attenuation depth parameter can be increased by 5 dB. The adjusted parameters are then reconfigured to the defective filter, and the initial dual-tone signal is repeatedly filtered using the defective filter with the adjusted basic filtering parameters until the residual harmonic noise rate of the initial dual-tone signal is less than or equal to the preset residual rate threshold.

[0152] If the residual harmonic noise rate is less than or equal to the preset residual rate threshold, it indicates that the rotor harmonic noise has been effectively filtered out. No filter parameter adjustment is needed, and the signal can directly proceed to the subsequent rotor vortex noise filtering stage to obtain a low-noise dual-tone signal. The residual rate threshold can be pre-calibrated based on the accuracy requirements of the fire scene audio acquisition, and is generally around 5%.

[0153] In one embodiment, the process of filtering the rotor vortex noise of the initial dual-tone audio signal using a filtering algorithm to obtain a low-noise dual-tone audio signal includes the following steps:

[0154] The stationarity of the initial dual-tone signal is detected by using the short-time autocorrelation function, and the initial stationary signal segment of the initial dual-tone signal is extracted based on the stationarity detection result.

[0155] Calculate the dual power spectral density parameters of the initial stationary signal segment;

[0156] The initial dual-tone signal is decomposed into frequency-point signal components, and the dual signal-to-noise ratio parameters of the frequency-point signal components are calculated.

[0157] The frequency domain gain compensation coefficient of the frequency point signal component is determined based on the dual signal-to-noise ratio parameters;

[0158] The frequency filtering coefficients of the frequency signal components are calculated by combining the dual power spectral density parameters and dual signal-to-noise ratio parameters;

[0159] By combining the frequency domain gain compensation coefficient and the frequency point filtering coefficient, the rotor eddy current noise of the initial dual-tone signal is filtered to obtain a low-noise dual-tone signal.

[0160] In this embodiment, the initial dual-tone audio signal with a harmonic noise residual rate meeting the standard is processed by frame segmentation to obtain multiple initial dual-tone audio signal frames. Before frame segmentation, it is necessary to set an appropriate frame length and frame shift to ensure that the signal between frames is continuous without discontinuities. Then, the short-time autocorrelation function of each initial dual-tone audio signal frame under different delays is calculated. The autocorrelation value R(m) of the initial dual-tone audio signal frame is obtained. Combined with its autocorrelation value, the peak autocorrelation decay rate and the variance of the autocorrelation function of the initial dual-tone audio signal frame are calculated. The peak autocorrelation decay rate = (R(0) - R(N / 4)) / R(0), where R(0) is the zero-delay autocorrelation value, R(N / 4) is the quarter-delay autocorrelation value, and N is the total number of sampling points of the initial dual-tone audio signal frame. The variance of the autocorrelation function is calculated by substituting the autocorrelation value into the variance calculation formula. If the initial dual-tone audio signal frame simultaneously satisfies that the peak autocorrelation decay rate is less than or equal to a preset decay rate threshold, such as 0.3, and the variance of the autocorrelation function is less than or equal to a preset variance, such as 0.05R(0), then the initial dual-tone audio signal frame is considered to have a peak autocorrelation decay rate less than or equal to a preset decay rate threshold, such as 0.3, and the variance of the autocorrelation function is less than or equal to a preset variance, such as 0.05R(0). 2 If the signal is stationary, it is determined to be a stationary signal. This is because the short-time autocorrelation function of a stationary signal decays smoothly without abrupt changes, while the autocorrelation function of non-stationary signals, such as the sound of explosions in a fire or the howling sound of a gas leak, will show sharp rises, falls, or periodic peaks. Based on the determination results, all initial dual-tone audio signal frames that meet the stationarity requirements are marked and seamlessly spliced ​​together to form an initial dual-tone audio signal segment with a length sufficient for subsequent frequency domain analysis. This segment contains only random stationary noise such as rotor vortex noise and electronic background noise, and contains no effective non-stationary sound signals from the fire scene.

[0161] Next, the extracted initial stationary signal segment is preprocessed. First, the DC component in the signal is removed, and then a Hanning window is applied to suppress spectral leakage in subsequent frequency domain transformations. Then, the initial stationary signal segment in the time domain is converted into a frequency domain signal through a short-time Fourier transform. The power spectral density of the corresponding frequency domain signals of the primary and secondary support UAVs is calculated separately, and the two are integrated to obtain the dual power spectral density parameter. This parameter reflects the noise power distribution at each frequency point in the initial stationary signal segment and is the core noise benchmark for subsequent calculation of frequency point filtering coefficients.

[0162] The initial dual-tone signal is subjected to full-band frequency domain discretization, dividing the entire fire scene audio acquisition frequency band into multiple continuous discrete frequency points. Each frequency point corresponds to an independent signal component, i.e., the frequency point signal component. The frequency resolution in the full-band frequency domain discretization process is completely consistent with the calculation resolution of the dual power spectral density parameters. Using the dual power spectral density parameters as the noise floor power benchmark for each frequency point, the ratio of signal power to noise floor power of each frequency point signal component of the main support UAV and the slave support UAV is calculated. The signal-to-noise ratio of each frequency point signal component is obtained through logarithmic transformation. The signal-to-noise ratio results of each frequency point signal component of the main support UAV and the slave support UAV are integrated to form the dual signal-to-noise ratio parameter of the frequency point signal component. This parameter quantifies the ratio of the effective acoustic signal of the fire scene at each frequency point to other noise. Other noise refers to rotor harmonic noise and rotor vortex noise that are not completely filtered out.

[0163] The dual signal-to-noise ratio (SNR) parameters are graded and determined by setting SNR threshold ranges and configuring different frequency domain gain compensation coefficients accordingly. The compensation coefficients for the same frequency point of the two assistance drones are kept consistent. Specifically, if the dual SNR parameters are within the SNR threshold range, the corresponding frequency signal component is determined to be a medium confidence frequency point. The effective acoustic signal of the fire scene at this type of frequency point accounts for a similar proportion to other noise, and a unity gain coefficient of 1 can be configured to obtain its corresponding frequency domain gain compensation coefficient, thus maintaining the original signal characteristics. If the dual SNR parameters are greater than or equal to the upper limit of the SNR threshold range, the corresponding frequency signal component is determined to be a high confidence frequency point. The high confidence frequency point is dominated by the effective signal of the fire scene, and a positive gain coefficient of 1.1-1.3 times is configured as its frequency domain gain compensation coefficient. The higher the dual SNR parameter ratio, the smaller the gain coefficient, to avoid excessive signal gain leading to distortion. If the dual signal-to-noise ratio (SNR) parameters are less than or equal to the lower limit of the SNR threshold range, the corresponding frequency signal component is determined to be a low-confidence frequency. This type of frequency is dominated by noise, and an attenuation coefficient of 0.5-0.8 times is configured as its frequency domain gain compensation coefficient. The lower the SNR, the smaller the attenuation coefficient, thus enhancing noise suppression.

[0164] The mean power spectral density of the signal components at the same frequency point of the primary and secondary support UAVs is calculated based on the dual power spectral density parameters, and this mean is used as the noise floor power value of that signal component. Mean fusion is used to eliminate noise floor power deviations caused by hardware differences between the two UAVs, such as microphone sensitivity and sampling accuracy, ultimately obtaining a uniform and stable noise floor power value for each frequency point, providing a benchmark for subsequent signal power inference and filter coefficient calculation. Then, the signal power value of the signal component at the frequency point is inferred from the noise floor power value. First, the signal-to-noise ratio (SNR) data of the signal components at the same frequency point of the primary and secondary support UAVs is extracted from the dual SNR parameters, and their mean SNR is calculated to eliminate SNR fluctuations caused by timing deviations or environmental interference during the dual-UAV signal acquisition process. Then, according to the core definition of SNR, SNR = 10 × log...10 The signal power value of the signal component at a given frequency is derived from (signal power value / noise floor power value). Integrating the signal power value and the noise floor power value yields the frequency power data. Next, the frequency filtering coefficient is determined based on the average signal-to-noise ratio (SNR). A higher average SNR indicates a higher proportion of effective signal at that frequency, and the frequency filtering coefficient is closer to 1 to maximize the retention of effective signals in the fire scene. Conversely, a lower average SNR indicates a higher proportion of noise, and the frequency filtering coefficient is closer to 0 to maximize the suppression of other noise. Furthermore, since there is a strong correlation between the SNR and the signal power value and the noise floor power value, the frequency filtering coefficient can also be determined based on these values: Frequency filtering coefficient = Signal power value / (Signal power value + Noise floor power value).

[0165] By combining frequency domain gain compensation coefficients and frequency point filtering coefficients, rotor eddy current noise filtering of the initial dual-tone audio signal is completed, resulting in a low-noise dual-tone audio signal. Specifically, firstly, all multi-frequency signal components of the initial dual-tone audio signal are multiplied by the corresponding frequency domain gain compensation coefficient to complete signal gain or attenuation processing at each frequency point. Then, the frequency point signal components after gain or attenuation processing are multiplied by the corresponding frequency point filtering coefficient to complete precise filtering at each frequency point, further eliminating rotor eddy current noise and other random background noise. Next, all frequency point signal components that have undergone coefficient weighting are subjected to inverse frequency domain transformation to convert them back to time domain signals. The resulting time domain signals are then subjected to inter-frame stitching and smoothing processing to eliminate signal discontinuities and amplitude fluctuations caused by frequency domain processing. Finally, low-noise dual-tone audio signals corresponding to the main and slave support UAVs are output. This signal has effectively filtered out rotor harmonic noise and rotor eddy current noise, completely preserving the effective acoustic signals from the fire scene. Furthermore, the timestamps of the two aircraft signals are strictly aligned and the amplitude characteristics are matched, making it directly usable for subsequent dual-modal acoustic data collaborative fusion.

[0166] In one embodiment, the process of collaboratively fusing dual-modal acoustic data, including acoustic detection signals and low-noise dual-audio signals, to obtain the equivalent sound velocity of the target fire site includes the following steps:

[0167] Based on the pre-acquired acoustic wave characteristics of the transmitting end, acoustic wave distortion analysis is performed on the acoustic wave detection signal, and the arrival time of the first wave and the total phase offset of the acoustic wave detection signal are extracted based on the acoustic wave distortion analysis results.

[0168] The initial acoustic velocity was calculated by combining the arrival time of the first wave and the dynamic spatial vector baseline.

[0169] The initial acoustic wave velocity is corrected by using the total phase offset to obtain the detected acoustic wave velocity;

[0170] Cross-correlation algorithm is used to perform cross-correlation calculation on low-noise dual-tone signals point by point, and dual-tone cross-correlation curve is constructed based on the cross-correlation calculation results;

[0171] The peak values ​​of the cross-correlation curves of the two-tone audio frequencies are detected using a peak detection algorithm, and a candidate peak list is obtained.

[0172] Traverse the candidate peak list and calculate the theoretical propagation path length of all cross-correlation peaks based on the dynamic spatial vector baseline and the acoustic propagation principle;

[0173] The core effective peaks in the candidate peak list are selected based on the theoretical propagation path length;

[0174] The noise delay deviation between the two aid drones is determined based on the peak coordinate value corresponding to the core effective peak value in the dual-audio cross-correlation curve.

[0175] The passive noise velocity is calculated by combining the noise delay bias and the dynamic space vector baseline;

[0176] The equivalent sound velocity of the target fire site is obtained by weighted fusion of the velocity of the detected sound wave and the velocity of the passive noise.

[0177] In this embodiment, acoustic distortion analysis is performed on the acoustic detection signal based on pre-acquired acoustic wave characteristics at the transmitting end to obtain acoustic distortion parameters, including the acoustic energy attenuation coefficient and the acoustic phase shift. The acoustic distortion parameters are then amplitude-corrected to obtain a corrected acoustic signal. The complex acoustic signal is then phase-calibrated to obtain a calibrated acoustic signal. The arrival time of the first wave and the total phase shift are extracted based on both the corrected and calibrated acoustic signals.

[0178] The low-noise dual-audio signal comprises a low-noise master audio signal from the primary support drone and a low-noise slave audio signal from the secondary support drone. First, equal-length segments of the low-noise master and slave audio signals are extracted to obtain low-noise dual-audio segments, with a duration of 50-200ms. Then, bandpass filtering and DC component removal are performed on these segments to obtain standard dual-audio segments, which include a standard master audio segment and a standard slave audio segment. Next, a cross-correlation algorithm is used to perform point-by-point cross-correlation calculations on the standard master and slave audio segments. Specifically, the standard master audio segment is used as the reference signal, and the standard slave audio segment is used as the signal to be matched. Then, the cross-correlation values ​​of the standard master and slave audio segments at different delay times are calculated using the cross-correlation formula.

[0179] Cross-correlation formula: ,in, This refers to the delay time, the sound propagation delay time determined based on experience. The duration of the signal segment for the low-noise dual-audio frequency band. This refers to the standard main audio segment. This refers to the standard derived from audio segments. After normalizing the calculated cross-correlation values, the time delay is used as the reference. Using the horizontal axis as the x-axis and the normalized cross-correlation value as the y-axis, a dual-audio cross-correlation curve is constructed. The delay time corresponding to the peak value in the dual-audio cross-correlation curve is the optimal matching delay time between the standard master audio segment and the standard slave audio segment, reflecting the propagation time difference of fire noise between the two support drones. Fire noise refers to the sound generated by fire combustion, such as the strong convection formed by the air density difference between the high-temperature zone and the normal-temperature zone in the fire, and the continuous whistling sound generated by the airflow hitting the building debris and walls.

[0180] A peak threshold is set based on the maximum normalized cross-correlation value, for example, 0.5 times the maximum normalized cross-correlation value. Then, a peak detection algorithm is used. An adaptive peak detection algorithm can be employed to traverse the dual-tone cross-correlation curves, using a sliding fixed-time window to filter out all cross-correlation peaks within the dual-tone cross-correlation curves that satisfy local maxima, have amplitudes greater than the peak threshold, and peak widths conforming to acoustic propagation characteristics. These are then included in the candidate peak list. A local maxima refers to the cross-correlation value of the point to the left of the candidate peak increasing with time delay. Monotonically increasing, the cross-correlation value of the points to its right increases with time delay. The peak width is monotonically decreasing, while allowing slight fluctuations in the cross-correlation values ​​on both sides, with the fluctuation amplitude less than 5% of the peak value determination threshold, to avoid slight fluctuations affecting the selection of candidate peaks. The peak width adopts the half-width at half-maximum (WHM) index, a general quantitative index for acoustic signal analysis. Half-width at half-maximum is set as half of the maximum amplitude of the candidate peak in the region. Then, the first point on the left and right sides of the candidate peak in the region that is lower than the WHM threshold is found along the horizontal axis. The time difference between the two points is calculated to obtain the WHM of the candidate peak. If the WHM of the candidate peak falls within the reasonable time width range of the cross-correlation peak of the acoustic signal in the fire field, it indicates that its peak width conforms to the acoustic propagation characteristics. This reasonable time width range is determined by the propagation characteristics of the fire field medium, the acoustic detection parameters of the dual-machine, and the bandwidth of the noise signal. It is the inherent width characteristic of the cross-correlation peak of the real fire field noise and can generally be set to [0.5ms, 50ms].

[0181] The basic principle of acoustic propagation includes the uniform propagation of sound waves in a straight line in a fire environment. The propagation time difference = actual baseline length / sound wave velocity, where the actual baseline length refers to the baseline modulus of the dynamic space vector baseline. Each candidate peak in the candidate peak list is traversed, and the theoretical propagation path length is calculated based on its corresponding delay time. Then, the product of the delay time and the sound wave velocity is calculated to obtain the theoretical propagation path length for each candidate peak. The difference between each theoretical propagation path length and the actual baseline length of the dynamic space vector baseline is calculated to obtain the path difference. The absolute value of the path difference is then divided by the actual baseline length to obtain the relative path error value for each candidate peak. If the relative path error value is less than a preset relative value threshold (e.g., the relative value threshold can be set to 0.05), and the angle between the propagation direction of the candidate peak and the propagation direction of the baseline is ≤10°, then the candidate peak is considered a valid peak. If only one valid peak exists, it is directly designated as the core valid peak. If multiple valid peaks exist, the valid peak with the highest normalized cross-correlation value is retained as the core valid peak. The propagation direction of the candidate peak can be determined based on the dynamic spatial vector baseline. Specifically, according to the principles of plane geometry, the spatial trajectory of the point between the two aid drones, whose path length difference is equal to the path difference, is a hyperbola with the dynamic spatial vector baseline as its real axis. Since only the core fire area between and around the two aid drones needs to be considered, only the branch of the hyperbola closest to the dynamic spatial vector is taken. The direction from all points on this branch towards the midpoint of the dynamic spatial vector baseline is taken as the potential propagation direction of the candidate peak. The potential propagation direction corresponding to the point on this branch closest to the dynamic spatial vector baseline is selected as the propagation direction of the candidate peak. Then, the angle between the propagation direction of the candidate peak and the propagation direction of the baseline is calculated using the vector dot product formula.

[0182] The peak abscissa of the core effective peak value in the dual-tone cross-correlation curve is used as the noise delay deviation between the two aid UAVs. The initial acoustic velocity is calculated by dividing the actual length of the dynamic space vector baseline by the arrival time of the first wave. Then, the initial acoustic velocity is corrected using the total phase offset. The total phase offset is positively correlated with the average refractive index of the fire medium, while the average refractive index of the fire medium is negatively correlated with the acoustic velocity. The proportion of the total phase offset to the reference phase offset is calculated, and the product of the initial acoustic velocity and the offset proportion is calculated to obtain the detection acoustic velocity. This parameter reflects the influence of the fire medium on the propagation of the detection acoustic signal. The reference phase offset refers to the baseline value of the phase offset under homogeneous medium, calculated based on the standard acoustic velocity of 343 m / s. ,in, For reference phase offset, The actual length of the baseline. The wavelength of the sound wave corresponds to the standard sound velocity. The passive noise velocity is obtained by dividing the actual baseline length by the noise delay deviation. Then, the detected sound velocity and the passive noise velocity are weighted and fused to obtain the equivalent sound velocity of the target fire. The weights can be determined based on relevant experience; for example, a weight of 0.7 can be assigned to the detected sound velocity, and a weight of 0.3 to the passive noise velocity.

[0183] This is because the sound velocity detected mainly reflects the average sound velocity of the two aid drones along a specific path. After phase correction, it can accurately characterize the average refractive index and sound velocity distribution of the medium along the path, making it suitable as a benchmark anchor point for calculating the equivalent sound velocity in a fire. However, because it can only reflect the acoustic characteristics of the direct or reflected path between the two drones, its coverage is narrow and its resistance to local interference is weak. For example, local high-temperature areas and small-scale turbulence in a fire can cause local distortion of the sound path. Even after phase correction, the sound velocity may still fail to reflect the characteristics of the medium across the entire fire area due to the single path, leading to misjudgment of non-physical interference areas, such as missing high-temperature areas outside the path. On the other hand, the passive noise velocity mainly reflects the local sound velocity between the two aid drones and the area around them. It can capture local medium abrupt changes outside the path, such as high-temperature areas at the edges and dispersed turbulence. However, the disadvantage is that the passive noise velocity has lower accuracy. Therefore, the sound velocity detected and the passive noise velocity can be weighted and fused to retain their respective advantages and compensate for each other's shortcomings, providing accurate acoustic data support for subsequent analysis of the characteristics of the fire medium.

[0184] In one embodiment, performing acoustic distortion analysis on the acoustic detection signal based on pre-acquired acoustic wave characteristics at the transmitting end, and extracting the first arrival time and total phase offset of the acoustic detection signal based on the acoustic distortion analysis results includes the following steps:

[0185] The acoustic wave detection signal is converted from analog to digital using an analog-to-digital converter to obtain a digital acoustic wave signal;

[0186] The digital acoustic signal is demodulated using orthogonal demodulation technology to obtain a complex acoustic signal.

[0187] Based on the pre-acquired acoustic wave characteristics of the transmitting end, the acoustic wave distortion parameters of the complex signal are extracted, including the acoustic wave energy attenuation coefficient and the acoustic wave phase offset.

[0188] The amplitude of the digital acoustic signal is corrected by using the acoustic energy attenuation coefficient, resulting in a corrected acoustic signal.

[0189] An edge detection algorithm is used to detect the amplitude gradient of the acoustic wave correction signal, and the arrival time of the first wave of the acoustic wave correction signal is determined based on the amplitude gradient detection results.

[0190] Phase calibration of the complex acoustic signal is performed using the acoustic phase offset, and folded phase identification is performed on the phase-calibrated complex acoustic signal to obtain folded phase data.

[0191] Phase folding compensation is performed on the complex acoustic signal based on the folded phase data, and the total phase offset of the complex acoustic signal is extracted based on the phase folding compensation result.

[0192] In this embodiment, appropriate sampling parameters are set for the analog-to-digital converter (ADC) based on the frequency characteristics of the acoustic wave detection signal. These parameters include the sampling rate and the number of quantization bits. The sampling rate is greater than twice the highest frequency of the acoustic wave detection signal, and the number of quantization bits is greater than or equal to 16 bits to ensure signal reconstruction accuracy. Next, the ADC with the set sampling parameters is used to discretize and quantize the analog acoustic wave detection signal. Specifically, the amplitude of the acoustic wave detection signal is collected point-by-point at a preset sampling rate, converting the continuous amplitude physical quantity into a discrete digital quantity. Simultaneously, a corresponding timestamp is bound to each discrete data point, resulting in a digital acoustic wave signal. The digital acoustic wave signal is represented as a one-dimensional sequence containing discrete digital amplitudes and timestamps, fully preserving the amplitude and time characteristics of the acoustic wave detection signal, and can be directly input into subsequent digital signal processing stages.

[0193] Next, orthogonal demodulation technology is used to separate the in-phase and quadrature components from the acoustic digital signal and combine them into a complex acoustic signal. Specifically, based on the reference frequency of the original acoustic detection signal emitted by the transmitter, two digital carrier signals with the same frequency and a 90° phase difference are generated in the digital domain, namely a sine carrier (I-channel carrier) and a cosine carrier (Q-channel carrier). The timestamps of the digital carrier signals are perfectly aligned with the acoustic digital signal. The acoustic digital signal is multiplied point-by-point with the I-channel and Q-channel digital carriers respectively to obtain the intermediate signal after multiplication, realizing the signal mixing and demodulation. Then, the obtained intermediate signal is subjected to low-pass filtering to remove the components that were mixed during multiplication. The generated high-frequency noise retains the effective signal components, resulting in the in-phase component (I component) and quadrature component (Q component) of the acoustic digital signal. Both components are discrete digital sequences. Then, the in-phase component and the quadrature component are combined in complex form to obtain the acoustic complex signal. The expression for a single discrete signal point in the acoustic complex signal is S=I+jQ, where j is the imaginary unit, the complex modulus corresponds to the instantaneous amplitude of the acoustic complex signal, the complex argument corresponds to the instantaneous phase of the acoustic complex signal, and the complex number refers to a single discrete signal point in the acoustic complex signal.

[0194] Next, the pre-acquired acoustic wave characteristics of the transmitting end are retrieved. The transmitting end refers to the acoustic signal transmitting end of the main support UAV. The acoustic wave characteristics of the transmitting end refer to the reference acoustic wave characteristics of the original acoustic wave detection signal. These characteristics include the standard amplitude and standard phase of the guide frequency band in the original acoustic wave detection signal. Both the standard amplitude and standard phase are fixed calibration values ​​with no propagation distortion. The guide frequency band is a dedicated and fixed reference frequency band that is pre-embedded in the signal when the transmitting end generates the original acoustic wave detection signal. It is not the main detection frequency band used to directly detect the fire medium. Its core function is to provide a fixed reference without propagation distortion for extracting the propagation distortion parameters of the acoustic wave detection signal. It is the reference signal segment for the entire signal distortion detection and correction.

[0195] Extracting acoustic wave characteristics of the corresponding pilot band from the complex acoustic wave signal involves accurately matching the corresponding frequency band in the complex acoustic wave signal based on the reference frequency and time range of the pilot band, and extracting the actual received amplitude and actual received phase of all time-domain sampling points in that frequency band. The actual received amplitude refers to the average value of the complex modulus, and the actual received phase refers to the average value of the complex argument. The actual received amplitude at each time-domain sampling point is divided by the corresponding standard amplitude to obtain the single-point energy attenuation coefficient. The average of the single-point energy attenuation coefficients at all time-domain sampling points is calculated after outlier removal using the 3σ criterion, yielding the acoustic energy attenuation coefficient. This coefficient quantifies the degree of absorption and scattering of acoustic energy by the fire-field medium, with a value range of (0,1], where a value closer to 1 indicates smaller energy attenuation. The single-point phase offset is obtained by subtracting the corresponding standard phase from the actual received phase at each time-domain sampling point. Consistency checks are performed on the single-point phase offset, eliminating phase abrupt changes caused by multipath reflections. The average is then used to obtain the final acoustic phase offset. This parameter quantifies the global systematic shift in acoustic phase caused by the fire-field medium, expressed in radians. The acoustic energy attenuation coefficient and the acoustic phase offset are integrated to obtain the acoustic distortion parameter, which serves as the core basis for subsequent signal correction.

[0196] A point-by-point amplitude correction of the acoustic digital signal is performed using an acoustic energy attenuation coefficient. The amplitude of each discrete signal point in the acoustic digital signal is divided by the acoustic energy attenuation coefficient to obtain the corrected amplitude for each discrete signal point. This restores the true amplitude characteristics of the acoustic wave before attenuation by the fire medium, avoiding misjudgment as noise due to excessively low initial signal amplitude caused by energy attenuation. Next, the amplitude sequence of the acoustic digital signal is smoothed and denoised using a moving average method to eliminate amplitude spikes caused by instantaneous noise, ensuring the continuity of the signal amplitude and obtaining the corrected acoustic signal. This provides an accurate amplitude basis for subsequent detection of the initial arrival time.

[0197] According to the formula The amplitude gradient of the corrected acoustic signal is calculated to obtain an amplitude gradient sequence, where, refers to timestamp The signal amplitude, The signal sampling time interval, Correcting the timestamp in the acoustic wave signal The amplitude gradient is used to reflect the rate of change of the amplitude of the acoustic correction signal. A significant gradient abrupt change occurs when the first wave arrives. A stable signal period before the first wave arrives is selected. This stable period refers to a signal segment containing only baseline noise and no effective acoustic signal, which can be selected based on the transmission time of the original acoustic detection signal. The mean amplitude gradient of the acoustic correction signal within the stable period is calculated, and a detection threshold for gradient abrupt change is determined based on this mean amplitude gradient. For example, three times the mean amplitude gradient can be used as the detection threshold. Then, an edge detection algorithm, such as Canny edge detection, is used to detect gradient abrupt changes in the amplitude gradient sequence of the acoustic correction signal. The first gradient abrupt change point exceeding the detection threshold is selected, and the timestamp corresponding to this gradient abrupt change point is used as the candidate first wave arrival time. Next, amplitude feature detection is performed on subsequent periods after the candidate first wave arrival time. If the signal amplitude in subsequent periods shows a stable upward trend, this candidate first wave arrival time can be used as the first wave arrival time of the acoustic correction signal. If the signal amplitude does not show a stable upward trend in the subsequent period after the candidate first wave arrival time, the amplitude gradient sequence is traversed to find the next abrupt change point until a valid point that meets the characteristics is found, and its corresponding timestamp is taken as the first wave arrival time.

[0198] The instantaneous phase of the complex acoustic signal is extracted, and phase calibration is performed by subtracting the acoustic phase offset from the instantaneous phase, resulting in an acoustic calibration signal. This step eliminates the global systematic phase offset caused by the fire medium, retaining only the phase information related to local refractive index changes in the medium, thus preventing the global offset from masking local folding features. Based on the continuity of acoustic phase, the phase change of normally propagating acoustic waves is continuous. Therefore, the phase difference between adjacent sampling points in the acoustic calibration signal should be within the range of [-π, π]. If it exceeds this range, the phase at that sampling point is determined to have undergone cross-cycle folding. If the adjacent phase difference is less than -π, it is determined to be a downward fold, i.e., the phase jumps from π to -π, losing 2π phase increments. If the adjacent phase difference is greater than π, it is determined to be an upward fold, i.e., the phase jumps from -π to π, gaining an additional 2π phase increment. The timestamps, folding types, and adjacent phase differences corresponding to all folded phases are recorded to obtain folded phase data.

[0199] Based on folded phase data, the phase sequence of the acoustic calibration signal is compensated point-by-point in chronological order. If a sampling point is folded downwards, 2π is added to the phase of all subsequent sampling points starting from that point to compensate for the lost phase increment. If a sampling point is folded upwards, 2π is subtracted from the phase of all subsequent sampling points starting from that point to compensate for the excess phase increment. Furthermore, phase continuity is maintained during the phase folding compensation process; the compensation amount for each sampling point with a folded phase is inherited by all subsequent phase points, avoiding phase breaks caused by segmented compensation. After completing the folding compensation of the entire phase sequence of the acoustic calibration signal, the restored true phase sequence is obtained, which fully reflects the phase changes of the sound wave propagating from the transmitter to the receiver. The phase value corresponding to the last timestamp in the true phase sequence is extracted to obtain the total phase offset of the acoustic complex signal. This parameter is directly related to the average refractive index of the fire field medium; the higher the refractive index, the greater the total phase offset.

[0200] In one embodiment, reference is made to Figure 3 The simulation of the sound wave trajectory of the target fire scene is completed based on the equivalent sound velocity of the fire scene and Fermat's principle. The non-physical interference data of the target fire scene is then derived from the sound wave trajectory simulation results, including the following steps:

[0201] S301. Based on the dynamic operating space between the two aid drones and the equivalent sound speed at the fire site, a refraction distribution model is iteratively constructed using Fermat's principle.

[0202] S302. Calculate the grid sound velocity of all grid cells in the dynamic grid space by combining the refraction distribution model and the sound wave energy attenuation coefficient.

[0203] S303. Using the ideal gas sound velocity formula and the ideal gas law, the grating sound velocity is mapped to the grating temperature and grating density, respectively.

[0204] S304. A raster state data field for dynamic raster space is constructed by combining raster temperature and raster density;

[0205] S305. Calculate the refractive gradient components of all grid cells according to the refractive distribution model, and divide all grid cells into interference interface grids and interference internal grids according to the refractive gradient components.

[0206] S306. Use the region growing algorithm to perform connected component analysis on all grid cells, and use the surface reconstruction algorithm to fit all the interference interface grids into the initial interference space based on the connected component analysis results.

[0207] S307. Extract multidimensional interference parameters of the initial interference space based on the grid state data field and refraction distribution model;

[0208] S308. Mark the non-physical interference areas in the dynamic grid space according to the multi-dimensional interference parameters, and integrate the multi-dimensional interference parameters and non-physical interference areas into non-physical interference data of the target fire site.

[0209] In this embodiment, a dynamic grid space is first constructed based on the dynamic operational space between the two support drones. Then, the corresponding equivalent sound velocity of the fire field is assigned to all grid cells within the dynamic grid space. Next, the acoustic refraction coefficient of all grid cells is calculated. The acoustic refraction coefficient is mapped to each grid cell, matching a corresponding acoustic refraction coefficient to each grid cell, generating a continuous spatial distribution model of the refraction coefficient within the dynamic grid space, thus obtaining an initial refraction distribution model. Then, the initial refraction distribution model is iteratively optimized using Fermat's principle to obtain a refraction distribution model. This model contains the true refractive index distribution that accurately reflects the fire field medium, i.e., the final refraction coefficient of each grid cell.

[0210] The final refractive index of the grid cells within the refractive distribution model is extracted. Then, the initial grid sound velocity corresponding to each grid cell is calculated based on the final refractive index. The formula for calculating the initial grid sound velocity is as follows: ,in, It is a proportionality constant, calculated by multiplying the refractive index of air under standard atmospheric pressure by the standard speed of sound. The refractive index of air under standard atmospheric pressure is approximately 1.0003, and the standard speed of sound is 343 m / s. This refers to the final refractive index corresponding to each grid cell. Next, if the acoustic energy attenuation coefficient is less than a preset attenuation coefficient threshold, such as 0.5, it indicates that the target fire area may have a high-temperature zone with strong medium energy absorption. Therefore, it is necessary to correct the initial grid acoustic velocity using the acoustic energy attenuation coefficient. The initial grid acoustic velocity correction formula is... ,in, The initial grid sound velocity, The sound wave energy attenuation coefficient is used, and the corrected initial grid sound velocity is taken as the grid sound velocity. If the sound wave energy attenuation coefficient is greater than or equal to the preset attenuation coefficient threshold, such as 0.5, then the initial grid sound velocity is directly taken as the grid sound velocity.

[0211] Next, the grating sound velocity is substituted into the ideal gas equation to calculate the thermodynamic temperature of the region corresponding to each grating cell, i.e., the grating temperature. Then, the grating temperature corresponding to each grating cell is substituted into the ideal gas law to calculate the air density corresponding to each grating cell, i.e., the grating density. Ideal gas law: ,in, This represents the atmospheric pressure within the dynamic grid space; standard atmospheric pressure can be used. This refers to the universal gas constant. Mapping grid temperature and grid density to the corresponding grid cells forms a dynamic grid space containing temperature and density data, i.e., a grid state data field.

[0212] The refractive index gradient components of each grid cell (i,j,k) in the x, y, and z directions are calculated based on the refractive index distribution model. The refractive index gradient component in the x direction is as follows: The refractive index gradient component in the y-direction: The refractive index gradient component in the z-direction: Where (i,j,k) are the coordinate values ​​of each grid cell in the x, y, and z directions. , and The x, y, and z coordinates of adjacent grid cells represent the differences in their coordinates, where n represents the acoustic refractive index of the grid cell. Next, the total refractive index gradient of each grid cell (i,j,k) is calculated based on its refractive index gradient components in the x, y, and z directions. The total refractive index gradient reflects the severity of refractive index changes. It directly affects the refractive index of corresponding grid cells due to changes in the position and intensity of non-physical disturbances in the target fire area, such as thermal flow and turbulence, thus causing synchronous updates to the refractive gradient components. The average total refractive index gradient of all grid cells within the dynamic grid space is calculated. A gradient abrupt change threshold is determined based on this average, for example, 0.2 times the average total refractive index gradient. All grid cells are iterated through; if their total refractive index gradient is greater than or equal to the gradient abrupt change threshold, they are marked as interfering interface grids; if their total refractive index gradient is less than the threshold, they are marked as interfering internal grids. Isolated grids with no adjacent interfering interface grids are identified as noise grids and reclassified as interfering internal grids to ensure the spatial continuity of interfering interface grids.

[0213] The connection component analysis of all grid cells is performed using a region growing algorithm. The specific implementation steps are as follows:

[0214] 1) Seed point selection: Traverse all interfering interface grids and take the first grid cell that has not been assigned a region identifier and has at least 2 other interfering interface grids in its 6-neighborhood as the initial seed point; remove isolated interfering interface grids, that is, grid cells with less than 2 interfering interface grids in their 6-neighborhood, and re-divide them into interfering internal grids to avoid noise points interfering with region growth.

[0215] 2) Growth criteria setting: The three-dimensional 6-neighborhood connectivity criterion is adopted. 6-neighborhood connectivity refers to grid cells that are adjacent vertically, horizontally, and front-back. The growth judgment condition is: the grid cell to be judged is an interfering interface grid cell and the total refractive index gradient difference between it and the seed point is ≤ the preset gradient difference threshold. In this embodiment, the preset gradient difference threshold can be taken as 20% of the average total refractive index gradient of all interfering interface grid cells in the dynamic grid space. If the condition is met, the neighboring grid cell is included in the same connected domain and set as a new seed point.

[0216] 3) Growth Iteration and Termination: Starting from the initial seed point, iterate through all neighboring rasters that meet the growth criteria until no new raster cells can be included in the growth region, thus completing the construction of a connected region; repeat the seed point selection and growth iteration process until all interfering interface rasters are assigned a unique region identifier, thus completing the full-region connected region analysis.

[0217] 4) Connected component filtering: Remove tiny connected components with fewer than the preset minimum grid number threshold, and retain effective and continuous interference interface connected components. The minimum grid number threshold can be 27, corresponding to a 3×3×3 grid in three-dimensional space, and a minimum spatial volume of 0.125m³. It can be dynamically adjusted according to grid resolution and detection requirements.

[0218] For the three-dimensional discrete grid connected domain, triangulation is used as the core algorithm for surface reconstruction. This algorithm can fit discrete spatial points into a continuous, non-overlapping, and gapless three-dimensional triangular surface, which fits the irregular features of the fire interference interface. The centroid spatial coordinates of all interference interface grids in each interference interface connected domain are extracted. These discrete centroid spatial coordinates are used as the input point set for surface reconstruction. The point set is fitted into a continuous three-dimensional triangular surface using the Delaunay triangulation method. This surface is the spatial contour of the fire sound propagation interference interface. The closed or semi-closed three-dimensional space enclosed by the three-dimensional triangular surface is the initial interference space. This initial interference space is the core region containing non-physical interferences from the fire, such as the mixing of hot and cold gases and inhomogeneous media. Furthermore, the boundary of the initial interference space is essentially a gradient abrupt region of the physical parameters (temperature, density, refractive index) of the air medium, which can be updated synchronously with the movement of the dual-assistance UAV.

[0219] The spatial contour coordinates of the initial interference space are extracted to determine all grid cells covered by this space, including the interference interface grid and internal grids. Based on the spatial contour coordinates, the volume, surface area, centroid coordinates, and maximum dimensions (length, width, height) of the initial interference space, as well as its contour irregularity (standard deviation of surface curvature), are calculated. Next, the grid temperature, grid density, and final refractive index of all grid cells covered by the initial interference space are retrieved from the grid state data field. The volume, surface area, centroid coordinates, maximum dimensions (three axes), contour irregularity, mean grid temperature, mean grid density, and mean refractive index of the initial interference space are integrated to obtain its multidimensional interference parameters. Considering the engineering requirements of fire detection, judgment thresholds are set for the core parameters in the multidimensional interference parameter set, such as a mean grid temperature ≥ 300℃, a volume ≥ 10m³, and a mean refractive index greater than 1.0003. If an initial interference space simultaneously meets the above judgment thresholds, the space it covers is marked as a non-physical interference area of ​​the target fire. The non-physical interference area and its corresponding multi-dimensional interference parameters are integrated into non-physical interference data of the target fire site.

[0220] Through the above steps, non-physical interferences that are difficult to identify based on images and point clouds within the target fire area can be identified, including high-pressure gas jet interference and extremely high temperature interference, providing a data foundation for the dual-aid drones to safely complete their aid missions.

[0221] In one embodiment, the refraction distribution model is iteratively constructed based on the dynamic operating space between the two aid drones and the equivalent sound speed at the fire site, using Fermat's principle, and includes the following steps:

[0222] The dynamic operational space between the two aid drones is processed by rasterization to obtain a dynamic raster space;

[0223] Based on the equivalent sound velocity at the fire scene and using the ideal gas sound velocity formula, an initial refraction distribution model is simulated in the dynamic grid space.

[0224] Based on the initial refraction distribution model and using Fermat's principle, the simulation of sound wave propagation in a non-uniform medium was completed, and the sound wave bending trajectory data was obtained.

[0225] Calculate the predicted phase of the acoustic wave trajectory data, and combine the predicted phase of the acoustic wave with the total phase offset to calculate the trajectory phase residual;

[0226] The initial refraction distribution model is iteratively optimized based on the trajectory phase residual and using the fast travel method until the trajectory phase residual is less than the residual threshold, thus obtaining the refraction distribution model.

[0227] In this embodiment, the dynamic operating space refers to the three-dimensional space in which the dual-aid UAV operates in real time. It is the effective area for fire detection and is determined by the flight altitude and detection radius of the dual-aid UAV. Based on an engineering-adapted grid resolution, such as 0.5m × 0.5m × 0.5m, balancing accuracy and real-time performance, the dynamic operating space is divided into discrete dynamic grid spaces. Each grid cell within the dynamic grid space is assigned a unique three-dimensional spatial index (x, y, z), forming a spatial framework for model construction. The grid cells are dynamically updated according to the position of the dual-aid UAV, ensuring the real-time performance of the space. Next, the real-time pose parameters in the real-time status parameters are timestamped with the acoustic wave detection signal. Based on the timestamp alignment result, the corresponding slave position coordinates when the acoustic wave detection signal is received from the signal receiver of the assistance drone are determined. Based on the slave position coordinates and the pre-acquired installation pose parameters of the acoustic receiver on the slave drone body, including the three-dimensional coordinate offsets ΔX, ΔY, ΔZ and attitude angles, the slave position coordinates are added to the three-dimensional coordinate offsets of the acoustic receiver to obtain the receiving point coordinates corresponding to the acoustic wave detection signal. The same method is used to calculate the transmission point coordinates of the original acoustic wave detection signal emitted from the signal transmitter of the main assistance drone.

[0228] Next, the product of the first wave arrival time and the reference acoustic propagation speed is calculated to obtain the actual propagation distance of the original acoustic wave detection signal from the main support UAV to the secondary support UAV. A geometric model of the main support UAV-propagation path point-secondary support UAV is constructed using the receiver and transmitter coordinates as two vertices and the actual propagation distance as the acoustic wave propagation path length. Then, the actual length of the dynamic space vector baseline is divided by the reference acoustic propagation speed to obtain the theoretical propagation time of the acoustic wave. If the absolute value of the difference between the theoretical direct propagation time and the actual first wave arrival time accounts for less than or equal to 5% of the theoretical propagation time, it indicates a high degree of agreement between the actual propagation time and the theoretical direct propagation time. This suggests that the propagation path of the original acoustic wave detection signal is likely unobstructed. Therefore, the straight path between the receiver and transmitter coordinates is divided into medium points at 1cm intervals along the direction of the dynamic space vector baseline, and the coordinates of these medium points are used as the corresponding sampling points for detecting the acoustic wave velocity. If the absolute value of the difference between the theoretical direct propagation time of the sound wave and the actual arrival time of the first wave accounts for more than 5% of the theoretical propagation time of the sound wave, it indicates that there is an obstacle in the propagation path of the original sound wave detection signal. The sum of the path lengths of the main support UAV, the propagation path point, and the support UAV equals the actual propagation distance. Then, using the principle of triangulation, combined with the coordinates of the receiving point, the transmitting point, and the baseline propagation direction, the coordinates of the reflection point are calculated. These coordinates of the reflection point are the coordinates of the detection sampling point corresponding to the sound wave velocity. The product between the noise delay deviation and the reference acoustic propagation velocity is calculated to obtain the noise propagation distance difference. Then, a three-dimensional hyperbolic model is constructed with the coordinates of the receiving point and the transmitting point as the two foci and the noise propagation distance difference as the distance difference. The three-dimensional hyperbolic model refers to the trajectory of points in three-dimensional space where the distance difference between the two foci is a constant value. Then, the intersection point of this three-dimensional hyperbolic model and the dynamic operating space is calculated. The coordinates of this intersection point are the coordinates of the passive sampling point corresponding to the passive noise velocity. Calculate the Euclidean distance between the coordinates of the passive sampling point and the coordinates of the detection sampling point at the same time point. If the Euclidean distance is less than or equal to a preset distance threshold, such as 2cm, then calculate the average coordinates of the passive sampling point and the detection sampling point as the sampling point coordinates of the equivalent sound velocity in the fire field.

[0229] Based on the coordinates of the sampling point, the equivalent sound velocity of the fire field is mapped to each grid cell in the dynamic grid space, and a corresponding equivalent sound velocity of the fire field is matched for each grid cell. Then, the sound wave refraction coefficient is determined according to the ideal gas sound velocity formula. Ideal gas sound velocity formula: ,in, This refers to the specific heat ratio of a gas. It refers to the thermodynamic temperature of a gas. This refers to the universal gas constant. This refers to the molar mass of the gas. First, the sound velocity under ideal conditions is calculated using the ideal gas sound velocity formula. Then, the sound velocity under ideal conditions is divided by the equivalent sound velocity in the fire field corresponding to each grid cell to obtain the sound refraction coefficient. The sound refraction coefficient is then mapped to each grid cell, matching the corresponding sound refraction coefficient to each grid cell. A continuous spatial distribution model of the refraction coefficient is generated within the dynamic grid space, resulting in the initial refraction distribution model.

[0230] The primary support UAV is designated as the acoustic transmitter, and the secondary support UAV as the acoustic receiver. Their corresponding coordinates are the receiver's coordinates and the transmitter's coordinates. A dynamic spatial vector baseline is used to determine the spatial relationships between the transmitter and receiver, including relative distance and azimuth, providing clear start and end boundaries and spatial references for acoustic propagation simulation. Fermat's principle in acoustic propagation applies that when sound waves propagate in a non-uniform medium, they spontaneously choose the path with the shortest propagation time from transmitter to receiver, rather than a simple straight path. This is the core theoretical basis for simulating the curved trajectory of sound waves, and all propagation direction adjustments revolve around this principle. Simultaneously, basic simulation parameters are set, including the acoustic emission frequency and simulation step size. The acoustic emission frequency is the hardware calibration value of the primary support UAV's acoustic emission module, which can be directly retrieved through the UAV's flight control system and is a key parameter for calculating the wavelength of the sound wave. The simulation step size balances simulation accuracy and computational efficiency; half the side length of the grid cell can be selected as the simulation step size, ensuring a detailed depiction of the propagation trajectory while avoiding excessive computation due to a too-small step size. The simulation employs ray tracing, approximating the non-uniform dynamic grid space as a segmented homogeneous medium. Each grid cell contains a homogeneous medium, and sound waves propagate in straight lines within the grid cells, undergoing refraction at the grid boundaries. The refraction follows Snell's law. ,in, and The acoustic refraction coefficient corresponding to adjacent grid cells. and These are the angle of incidence of the sound wave and the angle of refraction of the sound wave, respectively.

[0231] The receiver coordinates are set as the sound wave receiver, and the possible fire noise source areas within the dynamic grid space are set as the emission source range. The centroid of each grid cell is used as the calculation node for sound wave propagation, i.e., the grid node. Sound wave rays are emitted from each node within the emission source range at different initial angles. Combining the refractive index of each grid in the initial refraction distribution model, the refraction angle of the sound wave at the boundary of adjacent grids is calculated using Snell's law. The sound wave propagation path is traced grid by grid until the sound wave ray reaches the dual UAV receiver. Invalid rays that do not reach the receiver or whose propagation path exceeds the dynamic grid space are discarded, retaining the valid sound wave rays from the emission source to the dual receivers. For each valid sound wave ray, the coordinates of all grid nodes it passes through during propagation, the length of each path segment, the spatial location of the refraction point, the curvature change of the trajectory, and the total propagation path length are recorded to form sound wave curvature trajectory data.

[0232] The phase of a sound wave propagating in a medium is positively correlated with the propagation path length and angular frequency. The formula for calculating the predicted phase of a sound wave along a single trajectory is: ,in, The total propagation path length of the curved trajectory of the sound wave. The wavelength of the sound wave. ,in, For the speed of sound under ideal conditions, The frequency of the sound wave is determined by the characteristics of the noise source at the fire scene. The initial phase of the sound wave emission can be set to 0, which does not affect the phase difference calculation. The total propagation path length of each curved sound wave trajectory is extracted. Combined with the natural angular frequency and wavelength of the sound wave, the predicted phase of each curved trajectory is calculated. The sum of all predicted phases is calculated to obtain the total predicted phase offset. The difference between the total predicted phase offset and the sum of all total phase offsets is calculated to obtain the trajectory phase residual. If the trajectory phase residual is less than or equal to a preset residual threshold, such as π / 10, it indicates that the initial refractive distribution model can reflect the medium characteristics of the target fire site, and no further correction is needed; the initial refractive distribution model can be directly used as the refractive distribution model. If the trajectory phase residual is greater than the preset residual threshold, such as π / 10, it indicates that there is significant spatial inhomogeneity in the current propagation path. For example, drastic changes in temperature or density caused by high-temperature thermonuclear zones, high-pressure gas surges, or turbulent regions can lead to sudden changes in sound velocity, curved sound wave propagation paths, and ultimately, large phase deviations. This indicates that the medium of the target fire site has significant inhomogeneity, and the initial refractive distribution model needs to be corrected. If the predicted acoustic phase is greater than the corresponding total phase offset, it indicates that the actual sound velocity in the medium is lower. This is because a lower sound velocity results in a longer propagation time, a larger cumulative phase offset, and a higher refractive index. Therefore, the acoustic refractive index of the corresponding grid cell needs to be increased. If the predicted acoustic phase is less than the corresponding total phase offset, it indicates that the actual sound velocity in the medium is greater than the equivalent sound velocity in the corresponding fire field. Therefore, the acoustic refractive index of the corresponding grid cell needs to be decreased. The correction magnitude is determined based on the difference between the predicted acoustic phase and the total phase offset. The larger the difference, the larger the refractive index correction magnitude. The correction step for the acoustic refractive index stops when the recalculated trajectory phase residual is greater than a preset residual threshold, and the initial refractive distribution model obtained from the final correction round is used as the refractive distribution model.

[0233] In one embodiment, acquiring dual-drone point cloud data synchronously collected by two aid drones, and extracting entity interference data of the target fire site based on the dual-drone point cloud data includes the following steps:

[0234] Preprocess dual-machine point cloud data;

[0235] The preprocessed dual-machine point cloud data is spatiotemporally aligned with non-physical interference data to obtain synchronized point cloud data.

[0236] Based on the dynamic spatial vector baseline, dual-view point cloud stitching of synchronized point cloud data is completed to obtain stitched point cloud data;

[0237] Based on the non-entity interference data, the entity attributes of the merged point cloud data are marked in different regions to obtain the point cloud entity attributes;

[0238] Cluster analysis is performed on the stitched point cloud data based on the entity attributes of the point cloud, and the entity interference data of the target fire site is determined based on the cluster analysis results.

[0239] In this embodiment, the dual-aid drone point cloud data collected by the two drones are timestamped to ensure that the data reflects the same obstacle observed by the two drones at the same time but from different perspectives. A unique perspective label is added to each frame of dual-aid drone point cloud data to distinguish the point clouds of the primary and secondary drones, facilitating subsequent cross-validation. Next, the time-tamped point cloud data undergoes spatial coordinate overlap verification. For each point P1 in the primary drone point cloud, a candidate point P2 with a distance ≤ 5cm is searched in the secondary drone point cloud. If the Euclidean distance between the two points is ≤ a preset distance threshold, such as 3cm, they are considered coherent points. Then, geometric projection consistency verification is performed. A perspective projection model from the primary drone to the secondary drone is established based on a dynamic spatial vector baseline. The theoretical projection point P2' of the primary drone point P1 from the secondary drone's perspective is calculated. If the Euclidean distance deviation between the candidate point P2 in the secondary drone and the theoretical projection point P2' is ≤ 4cm, and the echo intensity difference between the two points is ≤ 20%, then the point is considered a real entity point. The established artifact detection rules directly mark points that exist only in the single-machine point cloud, have no corresponding candidate points, or have a geometric projection deviation >4cm as multipath artifact points, which need to be removed. Next, a sliding window collaborative filtering algorithm is used, setting a 5-frame sliding window to continuously track the real entity points in each frame of the point cloud, retaining only points that appear stably for 3 consecutive frames or more, and removing transient noise points that appear only once. Then, morphological filtering is used for further optimization. The filtered point cloud is first subjected to 3×3×3 voxel erosion kernels, and then to 2×2×2 voxel dilation kernels to eliminate small isolated noise points in the point cloud, retaining continuous entity contour point clouds, thus obtaining a dual-machine entity point cloud.

[0240] Next, the point clouds of the two physical drones are aligned with the non-physical interference area using timestamps and spatial locations to obtain synchronized point cloud data. Then, the synchronized point cloud data is stitched together using two time points to obtain a stitched point cloud. Specifically, a common observation reference point is first selected. In the area covered by both assistance drones, at least three high-confidence synchronized point clouds with a confidence level ≥ 1.0 are selected as registration references. Then, an optimized iterative nearest-point algorithm is executed. The synchronized point cloud of the primary assistance drone is used as the target point cloud, and the synchronized point cloud of the secondary assistance drone is used as the source point cloud. Initial translation and rotation transformation matrices are calculated. Then, the Euclidean distance between corresponding points in the source and target point clouds is calculated, and outliers with a distance > 5 cm are removed. The transformation matrix is ​​updated, and this iterative operation is repeated until the average distance error is ≤ 2 cm, reaching the convergence condition. After registration, the point cloud of the secondary assistance drone is mapped to the coordinate system of the primary assistance drone using the optimized transformation matrix, achieving seamless stitching of the two drone point clouds to obtain the stitched point cloud data. Based on the fire scene environment and the safety requirements of UAV operations, core clustering parameters were set, including spatial neighborhood radius, minimum number of cluster points, and spatial connectivity criteria. The spatial neighborhood radius matches the point cloud resolution and minimum entity size. The minimum number of cluster points is used to remove tiny invalid entities smaller than the safety threshold, such as small stones and debris. The spatial connectivity criteria are based on proximity in space and similar normal vectors. Next, density-based spatial clustering analysis was performed based on the core clustering parameters. A density clustering method adapted to irregular entities and uneven point cloud density was adopted. The merged point cloud data was aggregated region by region. Spatially closely connected points with sufficient density and continuous distribution were clustered into independent clusters, each corresponding to an independent solid entity. Points that were spatially discrete, had too low density, or could not form effective clusters were judged as noise and directly removed.

[0241] For each cluster of entities, spatial and geometric features are extracted, including spatial centroid coordinates, bounding box range, overall size, volume, height, distribution pattern, and occlusion range. Then, based on preset rules, it is determined whether they constitute entity interference. For example, entity clusters located in UAV operation paths, rescue corridors, or core fire areas, and whose volume exceeds the minimum interference threshold or whose height exceeds the safe flight altitude, forming occlusion or dangerous obstacles, are identified as valid entity interference targets. Entity clusters located in safe areas, with excessively small volumes, on flat terrain, or with low-lying harmless vegetation are identified as non-interference entities and are removed. For multiple spatially adjacent clusters belonging to the same continuous obstacle, such as continuous walls, large areas of trees, or continuously collapsed structures, a merging process is performed to obtain entity obstacle entities. Their spatial and geometric features are then integrated into the final entity interference data for the target fire area, providing accurate solid obstacle data for dual-assistance UAV obstacle avoidance, flight path planning, and safe fire area operations.

[0242] This application also provides an anti-interference system for unmanned aerial vehicle (UAV) flights in emergency scenarios, including:

[0243] The memory is configured to store instructions; and

[0244] The processor is configured to retrieve instructions from memory and, when executing the instructions, to implement the anti-interference method for UAV flight in emergency scenarios according to any one of the above.

[0245] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it.

[0246] The memory can be an internal storage unit of a computer device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) provided on the computer device. Furthermore, the memory can be a combination of internal storage units and external storage devices of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0247] This application also provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described anti-interference method for UAV flight in emergency scenarios.

[0248] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0249] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0250] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0251] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0252] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0253] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0254] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0255] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0256] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for anti-interference of UAV flight in emergency scenarios, characterized in that, Applied to dual-aid drones, wherein the dual-aid drones comprise a primary aid drone and a secondary aid drone, the method includes the following steps: Before the dual-aid drones enter the target fire area, environmental baseline parameters and drone pose information of the dual-aid drones are collected. Based on the UAV pose information, complete the two-way ranging interaction between the main assistance UAV and the slave assistance UAV, and construct the initial spatial vector baseline of the two assistance UAVs based on the two-way ranging interaction results. When the dual-aid drones enter the target fire site, they collect real-time fire environment parameters and real-time status parameters of the dual-aid drones. The initial spatial vector baseline is dynamically corrected by combining real-time status parameters, fire environment parameters, and environmental reference parameters to obtain a dynamic spatial vector baseline. The dual clock phase alignment verification of the dual assistance UAVs is then completed based on the dynamic spatial vector baseline. If the dual clock phase alignment verification of the dual-aid drone passes, the dual-audio signal of the target fire site will be collected using the microphone array carried by the dual-aid drone. A multi-stage filtering algorithm was used to filter the dual-tone audio signal from the fire scene to obtain a low-noise dual-tone audio signal. Acquire acoustic detection signals between the main support drone and the slave support drone, and complete the dual-modal acoustic data collaborative fusion of acoustic detection signals and low-noise dual-audio signals to obtain the equivalent sound velocity of the target fire site. Based on the equivalent sound velocity of the fire field and using Fermat's principle, the sound wave trajectory of the target fire field is simulated. Based on the sound wave trajectory simulation results, the non-physical interference data of the target fire field is derived. The non-physical interference data includes the high-temperature heat flow zone, gas jet zone and corresponding spatial centroid coordinates, coverage area, grid sound velocity mean, grid temperature mean, density mean and refractive index within the target fire field. The system acquires point cloud data collected simultaneously by two aid drones, and extracts entity interference data of the target fire site based on the point cloud data. The entity interference data includes building debris, collapsed objects and their corresponding spatial location, outer bounding box size, volume, obstruction range and geometric shape within the target fire site. Combining non-physical interference data and physical interference data, an anti-jamming flight strategy for dual-aid UAVs is generated.

2. The method according to claim 1, characterized in that, The real-time status parameters include real-time ranging parameters and real-time pose parameters. The step of dynamically correcting the initial spatial vector baseline by combining the real-time status parameters, fire environment parameters, and environmental reference parameters to obtain a dynamic spatial vector baseline, and completing the dual-clock phase alignment verification of the dual-aid UAVs based on the dynamic spatial vector baseline, includes the following steps: The real-time ranging parameters are scaled and compensated by combining fire scene environmental parameters and baseline environmental parameters. The initial spatial vector baseline is dynamically corrected by combining the real-time ranging parameters and real-time pose parameters that have completed scaling compensation, and the dynamic spatial vector baseline is output based on the dynamic correction results. Extract the baseline magnitude of the dynamic space vector baseline; The acoustic desired time window is constructed by combining the baseline modulus and the pre-acquired reference acoustic propagation velocity; Complete the bidirectional timestamp exchange between the primary and secondary aid drones, and calculate the link transmission delay parameters of the two aid drones based on the bidirectional timestamp exchange results. Collect dual clock pulse signals from the dual-aid UAV, and construct the clock phase deviation sequence of the dual-aid UAV based on the dual clock pulse signals; The clock deviation time series of the dual-aid UAVs was calculated by combining the link transmission delay parameters and the clock phase deviation sequence. The clock synchronization and synchronization stability of the dual-aid UAVs were verified based on the clock deviation time series and the acoustic expectation time window, respectively. The dual clock phase alignment of the dual-aid UAVs was verified by combining the clock synchronization and synchronization stability verification results.

3. The method according to claim 1, characterized in that, The process of using a filtering algorithm to perform multi-stage filtering on the dual-tone audio signal from the fire scene to obtain a low-noise dual-tone audio signal includes the following steps: Real-time acquisition of dual motor status information of dual-assistance drones; The rotor harmonic frequencies of the dual-assistance UAVs were calculated based on the dual-motor status information. The target defect filter is obtained by configuring the basic filtering parameters of the defect filter based on the rotor harmonic frequency. The basic filtering parameters include the filter order and attenuation depth parameters. The target defect filter is used to filter the dual-tone signal of the fire scene to obtain the initial dual-tone signal; The harmonic frequency range is determined based on the rotor harmonic frequency, and the rotor harmonic component and the residual noise component of the initial dual-tone signal of the fire scene are extracted based on the harmonic frequency range. The residual harmonic noise rate of the initial dual-tone signal was calculated by combining the residual noise component and the rotor harmonic component. If the residual harmonic noise rate is greater than the preset residual rate threshold, the basic filtering parameters of the defective filter are adjusted according to the residual harmonic noise rate, and the initial dual-tone signal is repeatedly filtered using the defective filter with the adjusted basic filtering parameters until the residual harmonic noise rate of the initial dual-tone signal is less than or equal to the preset residual rate threshold. If the residual harmonic noise rate is less than or equal to the preset residual rate threshold, the rotor eddy current noise of the initial dual-tone signal is filtered using a filtering algorithm to obtain a low-noise dual-tone signal.

4. The method according to claim 3, characterized in that, The process of using a filtering algorithm to filter the rotor vortex noise of the initial dual-tone signal to obtain a low-noise dual-tone signal includes the following steps: The stationarity of the initial dual-tone signal is detected by using the short-time autocorrelation function, and the initial stationary signal segment of the initial dual-tone signal is extracted based on the stationarity detection result. Calculate the dual power spectral density parameters of the initial stationary signal segment; The initial dual-tone signal is decomposed into frequency-point signal components, and the dual signal-to-noise ratio parameters of the frequency-point signal components are calculated. The frequency domain gain compensation coefficient of the frequency point signal component is determined based on the dual signal-to-noise ratio parameters; The frequency filtering coefficients of the frequency signal components are calculated by combining the dual power spectral density parameters and dual signal-to-noise ratio parameters; By combining the frequency domain gain compensation coefficient and the frequency point filtering coefficient, the rotor eddy current noise of the initial dual-tone signal is filtered to obtain a low-noise dual-tone signal.

5. The method according to claim 1, characterized in that, The process of collaboratively fusing dual-modal acoustic data, including acoustic wave detection signals and low-noise dual-audio signals, to obtain the equivalent sound velocity of the target fire site includes the following steps: Based on the pre-acquired acoustic wave characteristics of the transmitting end, acoustic wave distortion analysis is performed on the acoustic wave detection signal, and the arrival time of the first wave and the total phase offset of the acoustic wave detection signal are extracted based on the acoustic wave distortion analysis results. The initial acoustic velocity was calculated by combining the arrival time of the first wave and the dynamic spatial vector baseline. The initial acoustic wave velocity is corrected by using the total phase offset to obtain the detected acoustic wave velocity; Cross-correlation algorithm is used to perform cross-correlation calculation on low-noise dual-tone signals point by point, and dual-tone cross-correlation curve is constructed based on the cross-correlation calculation results; The peak values ​​of the cross-correlation curves of the two-tone audio frequencies are detected using a peak detection algorithm, and a candidate peak list is obtained. Traverse the candidate peak list and calculate the theoretical propagation path length of all cross-correlation peaks based on the dynamic spatial vector baseline and the acoustic propagation principle; The core effective peaks in the candidate peak list are selected based on the theoretical propagation path length; The noise delay deviation between the two aid drones is determined based on the peak coordinate value corresponding to the core effective peak value in the dual-audio cross-correlation curve. The passive noise velocity is calculated by combining the noise delay bias and the dynamic space vector baseline; The equivalent sound velocity of the target fire site is obtained by weighted fusion of the velocity of the detected sound wave and the velocity of the passive noise.

6. The method according to claim 5, characterized in that, The step of performing acoustic distortion analysis on the acoustic detection signal based on the pre-acquired acoustic wave characteristics of the transmitting end, and extracting the first arrival time and total phase offset of the acoustic detection signal based on the acoustic distortion analysis results, includes the following steps: The acoustic wave detection signal is converted from analog to digital using an analog-to-digital converter to obtain a digital acoustic wave signal; The digital acoustic signal is demodulated using orthogonal demodulation technology to obtain a complex acoustic signal. Based on the pre-acquired acoustic wave characteristics of the transmitting end, the acoustic wave distortion parameters of the complex signal are extracted, including the acoustic wave energy attenuation coefficient and the acoustic wave phase offset. The amplitude of the digital acoustic signal is corrected by using the acoustic energy attenuation coefficient, resulting in a corrected acoustic signal. An edge detection algorithm is used to detect the amplitude gradient of the acoustic wave correction signal, and the arrival time of the first wave of the acoustic wave correction signal is determined based on the amplitude gradient detection results. Phase calibration of the complex acoustic signal is performed using the acoustic phase offset, and folded phase identification is performed on the phase-calibrated complex acoustic signal to obtain folded phase data. Phase folding compensation is performed on the complex acoustic signal based on the folded phase data, and the total phase offset of the complex acoustic signal is extracted based on the phase folding compensation result.

7. The method according to claim 6, characterized in that, The process of simulating the sound wave trajectory of the target fire based on the equivalent sound velocity of the fire and using Fermat's principle, and then retrieving the non-physical interference data of the target fire based on the sound wave trajectory simulation results, includes the following steps: Based on the dynamic operating space between the two aid drones and the equivalent sound speed at the fire site, a refraction distribution model was iteratively constructed using Fermat's principle. The grid sound velocity of all grid cells in the dynamic grid space is calculated by combining the refraction distribution model and the sound wave energy attenuation coefficient. The lattice sound velocity is mapped to lattice temperature and lattice density using the ideal gas sound velocity formula and the ideal gas law, respectively. A raster state data field for dynamic raster space is constructed by combining raster temperature and raster density; The refractive gradient components of all grid cells are calculated based on the refractive distribution model, and all grid cells are divided into interference interface grids and interference internal grids based on the refractive gradient components. A region growing algorithm is used to perform connected component analysis on all grid cells. Based on the connected component analysis results and a surface reconstruction algorithm, all interfering interface grids are fitted to the initial interference space. Multidimensional interference parameters of the initial interference space are extracted based on the grid state data field and refraction distribution model. Non-physical interference regions within the dynamic grid space are marked based on multi-dimensional interference parameters, and the multi-dimensional interference parameters and non-physical interference regions are integrated into non-physical interference data of the target fire site.

8. The method according to claim 7, characterized in that, The process of constructing a refraction distribution model based on the dynamic operational space between the two aid drones and the equivalent sound velocity at the fire scene, and using Fermat's principle iteratively, includes the following steps: The dynamic operational space between the two aid drones is processed by rasterization to obtain a dynamic raster space; Based on the equivalent sound velocity at the fire scene and using the ideal gas sound velocity formula, an initial refraction distribution model is simulated in the dynamic grid space. Based on the initial refraction distribution model and using Fermat's principle, the simulation of sound wave propagation in a non-uniform medium was completed, and the sound wave bending trajectory data was obtained. Calculate the predicted phase of the acoustic wave trajectory data, and combine the predicted phase of the acoustic wave with the total phase offset to calculate the trajectory phase residual; The initial refraction distribution model is iteratively optimized based on the trajectory phase residual and using the fast travel method until the trajectory phase residual is less than the residual threshold, thus obtaining the refraction distribution model.

9. The method according to claim 1, characterized in that, The process of acquiring point cloud data from two aid drones simultaneously and extracting entity interference data of the target fire site based on the point cloud data includes the following steps: Preprocess dual-machine point cloud data; The preprocessed dual-machine point cloud data is spatiotemporally aligned with non-physical interference data to obtain synchronized point cloud data. Based on the dynamic spatial vector baseline, dual-view point cloud stitching of synchronized point cloud data is completed to obtain stitched point cloud data; Based on the non-entity interference data, the entity attributes of the merged point cloud data are marked in different regions to obtain the point cloud entity attributes; Cluster analysis is performed on the stitched point cloud data based on the entity attributes of the point cloud, and the entity interference data of the target fire site is determined based on the cluster analysis results.

10. An anti-interference system for unmanned aerial vehicle (UAV) flight in emergency scenarios, characterized in that, include: The memory is configured to store instructions; The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the anti-interference method for UAV flight in emergency scenarios according to any one of claims 1 to 9.

Images