Unmanned aerial vehicle dynamic path planning method and system facing strong electromagnetic interference environment

Abstract

The invention provides an unmanned aerial vehicle dynamic path planning method and system for a strong electromagnetic interference environment, and relates to the field of unmanned aerial vehicle navigation and interference resistance. The method mainly solves the problem of robust positioning of an optical sensing system in a strong electromagnetic environment. The core of the method lies in a spectrum anti-noise technology of a visual SLAM module, and spectrum separation of visible light and near-infrared band electromagnetic noise is realized through electromagnetic shielding packaging and an adaptive imaging algorithm of a multispectral camera. According to the method, a time-frequency double-domain filtering algorithm is adopted, the phenomena of light intensity abrupt change and frequency spectrum aliasing generated by electromagnetic pulses are eliminated in the feature extraction stage, and the electromagnetic interference mismatching rate of ORB feature points is reduced to 2% or below. The optimized feature point cloud keeps electromagnetic immune characteristics in the dynamic updating process, centimeter-level positioning precision can be achieved in the field intensity environment of 30 dBm or above, and meanwhile real-time detection of dynamic obstacles is achieved through an optical flow acceleration constraint mechanism.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) navigation and anti-interference, and in particular to a dynamic path planning method and system for UAVs in environments with strong electromagnetic interference. Background Technology

[0002] In environments with strong electromagnetic interference (field strength ≥ 30 dBm), UAV navigation faces severe challenges:

[0003] Optical perception degradation: Electromagnetic pulses cause drastic changes in light intensity in camera imaging (such as a brightness change of more than 2000 lux / microsecond in an instant), which seriously interferes with the feature point extraction and matching process of traditional visual SLAM, with a mismatch rate of more than 15%.

[0004] Sensor signal distortion: Electromagnetic noise intrudes into electronic devices through radiative coupling, resulting in a significant increase in image sensor noise and a decrease in SLAM mapping and positioning accuracy to the decimeter or even meter level;

[0005] Dynamic obstacle detection failure: Electromagnetic interference disrupts the continuity of optical flow calculation, causing a dynamic obstacle detection delay of more than 50 milliseconds, threatening flight safety.

[0006] Existing technical solutions mostly rely on multi-sensor fusion (such as lidar, IMU, GPS, etc.) or hardware-level electromagnetic shielding, but they have the following limitations:

[0007] Multi-sensor systems are complex, expensive, and cannot completely solve the core interference problem of optical sensing;

[0008] Traditional visual SLAM lacks robust design against electromagnetic noise and is prone to failure in scenarios with spectral aliasing and sudden changes in light intensity. Summary of the Invention

[0009] Purpose of the invention: To propose a dynamic path planning method for UAVs in strong electromagnetic interference environments, and further to propose a system that can implement the method, achieving noise suppression across the entire link from optical imaging and feature extraction to motion estimation, and realizing high-reliability operation of the vision system in strong electromagnetic environments.

[0010] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0011] Firstly, a dynamic path planning method for UAVs in environments with strong electromagnetic interference is proposed, including the following steps:

[0012] Capture electromagnetic shielding images;

[0013] Noise suppression is applied to the electromagnetic shielding image;

[0014] ORB feature points with electromagnetic immunity properties are extracted from the noise-suppressed electromagnetic shielding image.

[0015] By fusing the ORB feature points and dynamic obstacles detected based on optical flow acceleration information, a navigation trajectory is generated;

[0016] Based on the generated navigation trajectory, obstacle avoidance is initiated when an obstacle avoidance threshold is detected.

[0017] In a further embodiment of the first aspect, noise suppression is performed on the electromagnetic shielding image to eliminate spectral aliasing caused by electromagnetic pulses, with its notch center frequency w n The relationship between the variable t and time satisfies:

[0018]

[0019] In the formula, I EM (t) represents the real-time electromagnetic field strength; m is the modulation coefficient; w0 represents the default filter frequency when there is no electromagnetic field strength and it is zero.

[0020] In a further embodiment of the first aspect, the extraction of ORB feature points with electromagnetic immune properties specifically includes:

[0021] For each feature point, kNN matching is used to retain the first two nearest neighbor initial feature pairs and a dynamic consistency check is performed to reduce the number of mismatched feature points N. false Total number of matching pairs N total The following relationship exists between them:

[0022]

[0023] In the formula, II(*) is an indicator function that quantifies the geometric consistency test result into countable values; H is a 3×3 homogeneous transformation matrix that describes the planar projection transformation relationship between two images. Let x represent the coordinates of the k-th feature point obtained by transforming it through a 3×3 homogeneous transformation matrix. k ′ represents the coordinates of the k-th feature point actually observed by the sensor, when x k ′≈H xk Only pairs of points are considered a correct match; σ EM (k) represents the estimated electromagnetic noise intensity in the region where the k-th feature point is located; σ0 is the baseline noise level, used to normalize the local noise effect;

[0024] Feature point pairs that fail the dynamic consistency test are removed, thereby extracting ORB feature points with electromagnetic immune properties.

[0025] In a further embodiment of the first aspect, an electromagnetic susceptibility weighting factor w is introduced. EMOptimize ORB feature point selection:

[0026]

[0027] In the formula, (x,y) represents the position of the feature point in the image coordinate system; F EM (*) represents the electromagnetic field intensity distribution function at location (x,y); max(F EM This represents the maximum electromagnetic field strength in the current frame image, used for normalization.

[0028] In a further embodiment of the first aspect, the navigation trajectory is generated by fusing the ORB feature points and dynamic obstacles detected based on optical flow acceleration information, specifically including:

[0029] Detecting dynamic obstacles using the second derivative of optical flow field:

[0030]

[0031] ||α optical ||>3σ a

[0032] In the formula, α optical σ is the normalized optical flow acceleration; v is the instantaneous velocity of the feature point; t is the time variable, representing the timestamp difference between consecutive frames; 3σ λ This is the obstacle detection threshold; anything exceeding this threshold is considered an obstacle avoidance detection signal.

[0033] Dynamic target separation is achieved by constructing a motion consistency equation:

[0034]

[0035]

[0036] In the formula, I represents the gray value at position (x,y) in the image at time t; I0 ​​represents the gray value of the reference frame at (x,y), used to compare the difference between the current frame and the static background;

[0037] Extracting the electromagnetic interference immunity weight σ EM 2D-3D matching points; α calculated by optical flow difference across consecutive frames. optical , filter ||α optical ||>3σ α Outliers; Kalman filtering is used to fuse ORB feature points with acceleration information to form a state vector containing position, velocity, and acceleration information;

[0038] Predict the next position and velocity based on the current state and the drone's motion model:

[0039]

[0040] In the formula, S t+1|t S is the predicted state vector for the next time step; F is the state transition matrix; S t B is the current state vector; B is the control input matrix; u t This is the control input vector at the current moment;

[0041] Update state estimation by combining sensor observations:

[0042]

[0043] In the formula, S t+1 This is the updated state estimate; K(*) is the Kalman gain, which adjusts the weights of the prediction and observation; z t is the observation vector at the current moment, representing the data acquired from the sensor; H is the observation matrix; The observation residual represents the difference between the prediction and the actual observation.

[0044] Based on the optimization results, a series of trajectory points are calculated, and a smooth path for the navigation trajectory is generated using Minimum-snap.

[0045] In a further embodiment of the first aspect, based on the smooth path of the navigation trajectory, when an obstacle avoidance threshold α is detected... threhold When the obstacle avoidance threshold α is reached, obstacle avoidance is activated. threhold Dynamically adjusted based on electromagnetic field strength:

[0046] α threhold =5 + 0.2 × (E) EM -30),(E EM ≥30dBm)

[0047] In the formula, E EM This represents the electromagnetic field strength measured in real time.

[0048] A second aspect of the present invention provides a dynamic path planning system for unmanned aerial vehicles (UAVs), which can automatically execute the UAV dynamic path planning method for strong electromagnetic interference environments disclosed in the first aspect. Specifically, the UAV dynamic path planning system includes:

[0049] Electromagnetic shielding multispectral imaging module, used to capture electromagnetic shielding images;

[0050] A time-frequency dual-domain filtering module is used to suppress noise in the electromagnetic shielding image;

[0051] The electromagnetic immune feature point cloud generation module is used to extract ORB feature points with electromagnetic immune properties in an electromagnetically shielded image that has been noise-suppressed.

[0052] The optical flow acceleration constraint module detects dynamic obstacles using optical flow acceleration information.

[0053] The trajectory generation module is used to fuse ORB feature points and dynamic obstacles to generate a navigation trajectory; based on the generated navigation trajectory, obstacle avoidance is initiated when an obstacle avoidance threshold is detected.

[0054] In a further embodiment of the second aspect, the operating wavelength of the electromagnetic shielding multispectral imaging module covers the visible light range of 400-700 nm and the near-infrared range of 700-1100 nm, and its shielding effectiveness satisfies:

[0055]

[0056] In the formula, f represents the current frequency of the electromagnetic wave; c The cutoff frequency is represented by SE(f); SE(f) is the frequency function of the shielding effectiveness.

[0057] In a further embodiment of the second aspect, the filtering function of the time-frequency dual-domain filtering module is:

[0058] H(t,w)=G(t)·W(w)

[0059]

[0060] In the formula, t is the time variable; w is the angular frequency variable; H(t,w) represents the filtering function of the time-frequency dual-domain filtering module; G(t) represents the time-domain window function; W(w) represents the frequency-domain window function; σ t w is the time-domain Gaussian kernel width. c ω is the cutoff angular frequency; N is the filter order.

[0061] Beneficial Effects: This invention proposes a dynamic path planning method and system for UAVs in environments with strong electromagnetic interference, focusing on solving the robust positioning problem of optical sensing systems under strong electromagnetic conditions. The core of this method lies in the spectral noise reduction technology of the visual SLAM module: through electromagnetic shielding encapsulation of a multispectral camera and an adaptive imaging algorithm, spectral separation of electromagnetic noise in the visible light (400-700nm) and near-infrared (700-1100nm) bands is achieved. Innovatively, a time-frequency dual-domain filtering algorithm is employed to eliminate light intensity abrupt changes (>2000 lux / μs) and spectral aliasing caused by electromagnetic pulses during the feature extraction stage, reducing the electromagnetic interference mismatch rate of ORB feature points to below 2%. The optimized feature point cloud maintains electromagnetic immunity during dynamic updates, achieving centimeter-level positioning accuracy (positioning error ≤5cm@50Hz) in environments with field strengths above 30dBm. Simultaneously, real-time detection of dynamic obstacles is achieved through an optical flow acceleration constraint mechanism. Attached Figure Description

[0062] Figure 1 This is a flowchart of the UAV dynamic path planning method in the embodiment.

[0063] Figure 2 This is a schematic diagram of electromagnetic interference generation. Detailed Implementation

[0064] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0065] This invention addresses the failure of traditional navigation systems in environments with strong electromagnetic interference (such as high-voltage power stations). Through an innovative anti-interference visual SLAM architecture design, it significantly improves the positioning accuracy, environmental perception, and dynamic obstacle avoidance reliability of UAVs under extreme electromagnetic conditions. This solution includes the following key technologies:

[0066] 1. Electromagnetic shielding and multispectral fusion imaging technology

[0067] The electromagnetic shielding design of this invention adopts a triple composite structure of "conductive mesh-ferrite layer-wave absorbing coating", which improves the shielding effectiveness against high-frequency electromagnetic waves (1GHz-10GHz) by more than 40% compared with traditional single-layer shielding.

[0068] The conductive mesh layer is made of high-density copper-nickel alloy with a mesh density of 200 meshes / inch and a surface impedance of ≤0.1Ω / sq, which can reflect more than 80% of the incident electromagnetic wave energy; the ferrite layer uses manganese-zinc ferrite material with high magnetic permeability (μ_r≥5000@1MHz) to absorb low- and mid-frequency electromagnetic interference through hysteresis loss; the absorbing coating is made of carbon nanotube and silicon-based composite material with a thickness of ≤0.2mm and an absorption rate of ≥90% for electromagnetic waves in the frequency band above 10GHz.

[0069] Multispectral fusion mechanism: By simultaneously acquiring visible light (400-700nm) and near-infrared (700-1100nm) dual-band images, and utilizing the strong interference characteristics of electromagnetic noise in the visible light band (such as abrupt changes in light intensity and spectral aliasing) and the weak interference characteristics in the near-infrared band (noise energy attenuation of more than 30%), a dual-channel noise feature database is constructed. Based on this database, the adaptive spectral separation algorithm achieves anti-interference imaging through the following steps:

[0070] Noise energy comparison: Calculate the local energy difference between visible light and near-infrared images of the same scene to identify the region dominated by electromagnetic noise;

[0071] Dynamic weighted fusion: pixel-level compensation is performed using near-infrared image data for high-noise areas in visible light images;

[0072] Light intensity abrupt change suppression: For abnormal regions where the light intensity change rate exceeds 2000 lux / microsecond, the near-infrared channel is activated for priority rendering to avoid imaging distortion.

[0073] In this embodiment, a hexagonal honeycomb grid structure is fabricated using a high-purity copper-nickel alloy (85% copper, 15% nickel) through a precision etching process. The grid density is 200 meshes per inch, and the line width does not exceed 20 micrometers. A 0.1-micrometer-thick gold layer is plated onto the grid surface to reduce the contact resistance to below 0.1 ohms per square meter. This layer covers the outer frame of the camera lens, ensuring a reflectivity of no less than 80% for electromagnetic waves in the 1-10 GHz frequency band.

[0074] The ferrite shielding layer is made by mixing manganese zinc ferrite powder (magnetic permeability ≥5000) and epoxy resin in a 6:4 ratio, and then molding it into a 0.5 mm thick ring-shaped cover to wrap the camera circuit board, achieving magnetic interference attenuation of more than 30 dB in the 1MHz-100MHz frequency band.

[0075] The camera housing is coated with a microwave-absorbing coating made of a mixture of multi-walled carbon nanotubes and silicon-based polymer (PDMS), with carbon nanotubes accounting for 3% and the coating thickness of 0.2 mm. The coating is gradient-distributed and achieves a reflection loss of over -20 dB in the 10 GHz band, effectively absorbing residual electromagnetic waves.

[0076] The camera module integrates visible light and near-infrared dual sensors: the visible light sensor uses a CMOS chip with 2.4-micron pixels, and the near-infrared sensor uses an InGaAs chip, with a response band covering 700-1100 nanometers. The parallelism error of the optical axis of the dual sensors is controlled within 0.01 degrees. The dual-channel image synchronous acquisition is achieved through FPGA, with a frame rate of 30 frames per second and an inter-channel delay of no more than 1 millisecond.

[0077] During the electromagnetic interference calibration phase, a 1-10 GHz step signal was emitted through a radiation source to collect noise distribution data for dual-band images. The noise determination rule was: when the energy of a local area (16×16 pixels) in the visible light image is more than 2.5 times that of the corresponding near-infrared area, that area is determined to be the electromagnetic noise-dominant region.

[0078] During the dynamic fusion process, visible light and near-infrared pixels are weighted for high-noise areas, with visible light having a weight of 0.3 and near-infrared a weight of 0.7. If a pixel brightness change rate exceeding 2000 lux / microsecond is detected, the output is switched to the near-infrared channel, and a bilateral filter with a standard deviation of 15 is activated to eliminate high-frequency noise.

[0079] 2. Time-frequency domain joint filtering enhancement for visual SLAM

[0080] To address instantaneous brightness fluctuations (such as lightning-shaped stripes) caused by electromagnetic pulses, a short-window Gaussian filter (window width ≤ 5ms) is used to smooth the image sequence.

[0081] Dynamic window adjustment: Based on electromagnetic field strength monitoring data, the filtering window is dynamically reduced (down to a minimum of 2ms) to retain more texture details under strong interference;

[0082] Edge protection mechanism: Canny edge detection is used to lock the image contour region, and high-intensity filtering is applied only to non-edge regions to avoid loss of feature points.

[0083] Eliminating periodic spikes in the spectrum caused by electromagnetic interference using an adaptive notch filter:

[0084] Spectrum anomaly detection: Real-time analysis of image FFT spectrum to identify abnormal frequency points with amplitudes exceeding 10 times the background noise;

[0085] Notch bandwidth optimization: The notch width is dynamically adjusted (typical value 10-50Hz) according to the energy distribution of abnormal frequency points, accurately filtering out interference while retaining effective signals.

[0086] Eliminating periodic spikes in the spectrum caused by electromagnetic interference using an adaptive notch filter:

[0087] Spectrum anomaly detection: Real-time analysis of image FFT spectrum to identify abnormal frequency points with amplitudes exceeding 10 times the background noise;

[0088] Notch bandwidth optimization: The notch width is dynamically adjusted (typical value 10-50Hz) according to the energy distribution of abnormal frequency points, accurately filtering out interference while retaining effective signals.

[0089] Joint filtering mechanism: The synergistic effect of time-domain and frequency-domain filtering is achieved through the following process:

[0090] Temporal preprocessing: Eliminating sudden brightness fluctuations;

[0091] Frequency domain secondary filtering: suppresses residual periodic noise;

[0092] Residual compensation: For areas with blurred textures caused by filtering, local histogram equalization is used to enhance contrast.

[0093] In this embodiment, the raw data collected by the magnetometer is subjected to wavelet transform based on the db4 wavelet basis to effectively filter out transient electromagnetic noise caused by high-voltage arc. Subsequently, the sensor data (WGS84 geographic coordinate system) is mapped to the local grid map coordinate system of the power station through coordinate transformation, so as to achieve accurate spatial alignment of multi-source data such as magnetometer and millimeter-wave radar with the location of power station equipment. Finally, a high-resolution (0.1m×0.1m) electromagnetic interference heat map is generated, and the real-time electromagnetic intensity distribution of different areas is dynamically marked.

[0094] The Gaussian filter window width is dynamically adjusted based on real-time electromagnetic field strength data: the window width decreases by 0.06 milliseconds for every 1 dB increase in field strength, with a minimum width of 3.2 milliseconds (at a field strength of 30 dB). Before filtering, image contours are extracted using Canny edge detection. Strong Gaussian smoothing (standard deviation 1.5) is applied to non-edge regions, while weak smoothing (standard deviation 0.5) is applied to edge regions.

[0095] Local contrast analysis is performed on the filtered image: if the average gradient of the region is less than 10 (8-bit grayscale range), contrast-limited histogram equalization (CLAHE) with a block size of 8×8 is used to enhance texture details.

[0096] Perform a 512×512-point Fourier transform on each frame of the image to calculate the mean and standard deviation of the spectral background noise. Mark abnormal frequency points whose amplitude exceeds 10 times the mean of the background noise. Construct a notch filter centered on the abnormal points with a bandwidth of the maximum value of 10 Hz or 20% of the abnormal frequency. Recover the effective signal through inverse transform.

[0097] The total time consumption of time-frequency joint filtering is controlled within 15 milliseconds, and FAST corner features are pre-extracted from the output image, retaining no less than 200 effective feature points per frame.

[0098] 3. Electromagnetic Immunoassay Feature Extraction and SLAM Optimization

[0099] Introducing an electromagnetic susceptibility weight factor into the traditional ORB algorithm to dynamically optimize feature point selection:

[0100] Noise intensity mapping: Based on data from the electromagnetic field strength monitoring module, a noise intensity heatmap of the image region is generated;

[0101] Weighting rules: For regions with noise intensity >25dBm, the feature point scoring weight is reduced by 70%, and feature points in low-noise regions (<15dBm) are selected first.

[0102] Redundant feature backup: For key feature points (such as edge intersections) in high-noise areas, descriptors are reconstructed using near-infrared image data to ensure SLAM continuity.

[0103] A point cloud reliability model is established by combining real-time electromagnetic field strength data:

[0104] Credibility assessment: Dynamic weights are calculated based on three indicators: noise intensity in the region where the feature point is located, historical matching success rate, and motion consistency.

[0105] Point cloud self-cleaning: When the electromagnetic noise in a certain area continues to exceed the standard (>30dBm for more than 500ms), 80% of the feature points in that area are automatically removed, and the near-infrared channel is triggered to supplement the acquisition.

[0106] Density equalization control: The feature point density in low-noise areas is increased to 150 per frame to ensure the accuracy of SLAM mapping.

[0107] In this embodiment, a noise heatmap is generated based on electromagnetic field strength data, and the noise intensity is divided into three levels: low (<15 dB), medium (15-25 dB), and high (>25 dB). During feature matching, the feature weight is set to 1.0 for the low noise region, 0.5 for the medium noise region, and 0.3 for the high noise region. Feature points with a weight ≥0.7 are preferentially matched.

[0108] For key feature points in high-noise regions (such as loop closure detection points), ORB descriptors from the near-infrared channel are extracted simultaneously as backups. When the noise intensity remains above 30 dB for more than 500 milliseconds, 80% of the old feature points are automatically discarded, and 100 new feature points from the near-infrared channel are added.

[0109] Feature point reliability is calculated by comprehensively considering noise intensity, historical matching rate, and motion consistency. Low-noise areas (<15 dB) are uniformly sampled using a 32×32 pixel grid, increasing the feature point density to 150 per frame to ensure the stability of the SLAM map.

[0110] 4. Global-local collaborative path planning

[0111] A heterogeneous decision-making architecture with global-local collaboration is constructed, integrating and improving the A* algorithm. The A* algorithm based on time-varying electromagnetic interference constraints has a cost function defined as:

[0112] f(n)=g(n)+h(n)+γ·EMI penalty (n)

[0113] in This involves quantifying the cumulative risk in the high-interference region (EMI ≥ 50 dBμV / m); the planned path is smoothed using third-order uniform B-spline interpolation to satisfy the curvature constraint k ≤ m. -1 To ensure the kinematic feasibility of the mobile platform.

[0114] Design a Markov decision process based on PPO (Proximal Policy Optimization): electromagnetic intensity (normalized [0,50]->[0,1]), obstacle distance (piecewise mapping d∈[0,5m]->[0,1]), and remaining charge (linear mapping [0,100%]->[0,1]).

[0115] 5. Real-time obstacle avoidance fusion mechanism based on visual SLAM

[0116] Distinguishing between the UAV's own motion and dynamic obstacles through second-order derivative analysis of the optical flow field:

[0117] Motion decomposition model: The optical flow field is decomposed into translational components (caused by the motion of the UAV itself) and rotational / deformation components (caused by the motion of obstacles);

[0118] Acceleration threshold determination: When the local optical flow acceleration exceeds 3 standard deviations of the background noise (typical threshold 0.15 rad / s²). 2 ), which is determined to be a dynamic obstacle;

[0119] Obstacle avoidance priority classification: The obstacle avoidance path weight is dynamically adjusted based on the acceleration amplitude, prioritizing high-threat obstacles (acceleration > 0.3 rad / s²). 2 This triggers an emergency avoidance maneuver.

[0120] Positioning error correction: The high-precision pose (error ≤ 5cm) output of visual SLAM is real-time calibrated to correct the cumulative error of optical flow calculation, avoiding misjudging stationary obstacles;

[0121] Motion prior constraints: Optical flow information provides initial values ​​for motion estimation for SLAM, shortening the number of pose calculation iterations and increasing the SLAM update frequency from 30Hz to 50Hz;

[0122] Closed-loop verification mechanism: The obstacle avoidance path planning results are fed back to the SLAM system, and the feasibility of the path is verified through relocation to avoid getting trapped in local optima.

[0123] In this embodiment, the optical flow of the UAV itself is calculated using the pose data output by visual SLAM, and the residual optical flow caused by external obstacles is separated. The acceleration of the residual optical flow is calculated, and if it exceeds three times the standard deviation of the historical background value, it is determined to be a dynamic obstacle.

[0124] The optical flow offset is calibrated using the ground truth SLAM pose value within 50 milliseconds, and the initial velocity of the optical flow estimate is input into the SLAM optimization process to reduce the number of pose calculation iterations. The obstacle avoidance path planning results are verified by reprojection error; if the error exceeds 2 pixels, local map reconstruction is triggered.

[0125] Verification results: Under strong interference of 5GHz / 40dB, the PSNR of imaging quality is improved by 16dB; the SLAM positioning error in dynamic electromagnetic environment is reduced to 2.1cm; the obstacle avoidance response delay for obstacles of 30cm is ≤15ms, and the safe distance margin is ≥1.5m (UAV speed 10m / s).

[0126] The following embodiments further disclose details of a dynamic path planning method for unmanned aerial vehicles (UAVs) in environments with strong electromagnetic interference.

[0127] First, an electromagnetic shielding image is captured and noise-suppressed to eliminate spectral aliasing caused by electromagnetic pulses, with its notch filter center frequency w. n The relationship between the variable t and time satisfies:

[0128]

[0129] In the formula, I EM (t) represents the real-time electromagnetic field intensity; m represents the modulation coefficient.

[0130] Next, ORB feature points with electromagnetic immunity properties are extracted from the noise-suppressed electromagnetic shielding image. For each feature point, kNN matching is used to retain the initial two nearest neighbor feature pairs N. total A consistency check is then performed, using the following formula:

[0131]

[0132] In the formula, II(*) is an indicator function that quantifies the geometric consistency test result into countable values; H is a 3×3 homogeneous transformation matrix that describes the planar projection transformation relationship between two images. Let x represent the coordinates of the k-th feature point obtained by transforming it through a 3×3 homogeneous transformation matrix. k ′ represents the coordinates of the k-th feature point actually observed by the sensor, when x k ′≈H xk Only pairs of points are considered a correct match; σ EM (k) represents the estimated electromagnetic noise intensity in the region where the k-th feature point is located; σ0 is the baseline noise level, used to normalize the local noise effect;

[0133] Feature point pairs that fail the dynamic consistency test are removed, thereby extracting ORB feature points with electromagnetic immune properties.

[0134] In this process, an electromagnetic susceptibility weighting factor w is introduced. EM Optimize ORB feature point selection:

[0135]

[0136] Subsequently, dynamic obstacles are detected using the second derivative of the optical flow field, as shown in the following formula:

[0137]

[0138] ||α optical ||>3σ α

[0139] In the formula, α optical σ is the normalized optical flow acceleration; v is the instantaneous velocity of the feature point; t is the time variable, representing the timestamp difference between consecutive frames; 3σ α This is the obstacle detection threshold; anything exceeding this threshold is considered an obstacle avoidance detection signal.

[0140] Dynamic target separation is achieved by constructing a motion consistency equation:

[0141]

[0142] Next, ORB feature points and dynamically detected obstacles based on optical flow acceleration information are fused to generate a navigation trajectory. Electromagnetic interference immunity weights σ are then extracted. EM 2D-3D matching points; α calculated by optical flow difference across consecutive frames. optical , filter ||α optical ||>3σ α Outliers; Kalman filtering is used to fuse ORB feature points with acceleration information to form a state vector containing position, velocity, and acceleration information;

[0143] Predict the next moment's position and velocity based on the current state and the drone's motion model:

[0144]

[0145] In the formula, S t+1|t S is the predicted state vector for the next time step; F is the state transition matrix; S t B is the current state vector; B is the control input matrix; u t This is the control input vector at the current moment;

[0146] Update state estimation by combining sensor observations:

[0147]

[0148] In the formula, S t+1 This is the updated state estimate; K(*) is the Kalman gain, which adjusts the weights of the prediction and observation; z t Z represents the observation vector at the current moment, which is the data acquired from the sensor; H is the observation matrix; z t -HS t+1|tThe observation residual represents the difference between the prediction and the actual observation.

[0149] Based on the optimization results, a series of trajectory points are calculated, and a smooth path for the navigation trajectory is generated using Minimum-snap.

[0150] Finally, based on the generated navigation trajectory, obstacle avoidance is initiated when an obstacle avoidance threshold is detected. The obstacle avoidance threshold α is... threhold Dynamically adjusted based on electromagnetic field strength:

[0151] α threhold =5 + 0.2 × (E) EM -30),(E EM ≥30dBm)

[0152] Furthermore, to achieve the above process, this embodiment also discloses a UAV dynamic path planning system. This system comprises an electromagnetic shielding multispectral imaging module, a time-frequency dual-domain filtering module, an electromagnetic immune feature point cloud generation module, an optical flow acceleration constraint module, and a trajectory generation module. The electromagnetic shielding multispectral imaging module is used to capture electromagnetic shielding images. The time-frequency dual-domain filtering module is used to suppress noise in the electromagnetic shielding images. The electromagnetic immune feature point cloud generation module is used to extract ORB feature points with electromagnetic immune characteristics within the noise-suppressed electromagnetic shielding images. The optical flow acceleration constraint module detects dynamic obstacles using optical flow acceleration information. The trajectory generation module is used to fuse ORB feature points and dynamic obstacles to generate a navigation trajectory; based on the generated navigation trajectory, obstacle avoidance is initiated when an obstacle avoidance threshold is met.

[0153] Preferably, the electromagnetic shielding multispectral imaging module operates in the visible light 400-700nm and near-infrared 700-1100nm bands, and its shielding effectiveness meets the following requirements:

[0154]

[0155] In the formula, f represents the current frequency of the electromagnetic wave; c The cutoff frequency is represented by SE(f); SE(f) is the frequency function of the shielding effectiveness.

[0156] The filtering function of the time-frequency dual-domain filtering module is:

[0157] H(t,w)=G(t)·W(w)

[0158]

[0159] In the formula, t is the time variable; w is the angular frequency variable; σ t w is the time-domain Gaussian kernel width. c ω is the cutoff angular frequency; N is the filter order.

[0160] Where σ t =0.1ms is the Gaussian kernel width in the time domain. Where ω is the cutoff angular frequency, and N = 5 is the order.

[0161] In this embodiment, the electromagnetic immune feature point cloud generation module adopts an improved ORB feature extraction algorithm, and the mismatch rate satisfies:

[0162]

[0163] Where N false N represents the number of mismatched feature points. total This represents the total number of matched pairs.

[0164] The optical flow acceleration constraint module detects dynamic obstacles using the second derivative of the optical flow field.

[0165]

[0166] Where α optical σ is the normalized optical flow acceleration; v is the instantaneous velocity of the feature point; t is the time variable, representing the timestamp difference between consecutive frames; 3σ α This is the obstacle detection threshold; anything exceeding this value is considered an obstacle avoidance detection signal.

[0167] In this embodiment, the electromagnetic shielding multispectral imaging module adopts a layered shielding structure, including:

[0168] Outer conductive mesh layer (surface impedance ≤ 0.1Ω / sq);

[0169] Intermediate ferrite layer (permeability μ_r≥5000@1MHz);

[0170] Inner layer nano-absorbing coating (reflection loss ≥20dB@1-10GHz).

[0171] In this embodiment, the time-frequency dual-domain filtering module employs adaptive notch filtering in the frequency domain to eliminate spectral aliasing caused by electromagnetic pulses. Its notch center frequency is w. n satisfy:

[0172]

[0173] Among them I EM (t) represents the real-time electromagnetic field strength, and m = 0.02 is the modulation coefficient.

[0174] In this embodiment, the optical flow acceleration constraint module achieves dynamic target separation by constructing a motion consistency equation:

[0175]

[0176] In the formula, I represents the gray value of position (x,y) in the image at time t; I0 ​​represents the gray value of the reference frame at (x,y), used to compare the difference between the current frame and the static background.

[0177] The method flows disclosed in the above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. The computer program is written to a storage medium and runs on a corresponding system, thereby enabling the system to automatically execute the method flows disclosed in the above embodiments, which will not be elaborated further here. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0178] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A dynamic path planning method for unmanned aerial vehicles (UAVs) in environments with strong electromagnetic interference, characterized in that, Includes the following steps: Capture electromagnetic shielding images; Noise suppression is applied to the electromagnetic shielding image; ORB feature points with electromagnetic immunity properties are extracted from the noise-suppressed electromagnetic shielding image. By fusing the ORB feature points and dynamic obstacles detected based on optical flow acceleration information, a navigation trajectory is generated; Based on the generated navigation trajectory, obstacle avoidance is initiated when an obstacle avoidance threshold is detected.

2. The UAV dynamic path planning method for strong electromagnetic interference environments according to claim 1, characterized in that, The electromagnetic shielding image is subjected to noise suppression to eliminate spectral aliasing caused by electromagnetic pulses, with its notch filter center frequency w. n The relationship between the variable t and time satisfies: In the formula, I EM (t) represents the real-time electromagnetic field strength; m is the modulation coefficient; w0 represents the default filter frequency when there is no electromagnetic field strength and it is zero.

3. The UAV dynamic path planning method for strong electromagnetic interference environments according to claim 1 or 2, characterized in that, The extraction of ORB feature points with electromagnetic immune properties specifically includes: For each feature point, kNN matching is used to retain the first two nearest neighbor initial feature pairs and a dynamic consistency check is performed to reduce the number of mismatched feature points N. false Total number of matching pairs N total The following relationship exists between them: In the formula, II(*) is an indicator function that quantifies the geometric consistency test result into countable values; H is a 3×3 homogeneous transformation matrix that describes the planar projection transformation relationship between two images. Let x represent the coordinates of the k-th feature point obtained by transforming it through a 3×3 homogeneous transformation matrix. k ′ represents the coordinates of the k-th feature point actually observed by the sensor, when x k ′≈H xk Only pairs of points are considered a correct match; σ EM (k) represents the estimated electromagnetic noise intensity in the region where the k-th feature point is located; σ0 is the baseline noise level, used to normalize the local noise effect; Feature point pairs that fail the dynamic consistency test are removed, thereby extracting ORB feature points with electromagnetic immune properties.

4. The UAV dynamic path planning method for strong electromagnetic interference environments according to claim 3, characterized in that, Introducing an electromagnetic susceptibility weighting factor w EM Optimize ORB feature point selection: In the formula, (x,y) represents the position of the feature point in the image coordinate system; F EM (*) represents the electromagnetic field intensity distribution function at location (x,y); max(F EM This represents the maximum electromagnetic field strength in the current frame image, used for normalization.

5. The UAV dynamic path planning method for strong electromagnetic interference environments according to claim 1, characterized in that, By fusing the ORB feature points and dynamic obstacles detected based on optical flow acceleration information, a navigation trajectory is generated, specifically including: Detecting dynamic obstacles using the second derivative of optical flow field: ||a optical ||>3s α In the formula, α optical σ is the normalized optical flow acceleration; v is the instantaneous velocity of the feature point; t is the time variable, representing the timestamp difference between consecutive frames; 3σ α This is the obstacle detection threshold; anything exceeding this threshold is considered an obstacle avoidance detection signal. Dynamic target separation is achieved by constructing a motion consistency equation: In the formula, I represents the gray value at position (x,y) in the image at time t; I0 ​​represents the gray value of the reference frame at (x,y), used to compare the difference between the current frame and the static background; Extracting the electromagnetic interference immunity weight σ EM 2D-3D matching points; α calculated by optical flow difference across consecutive frames. optical , filter ||α optical ||>3σ α Outliers; Kalman filtering is used to fuse ORB feature points with acceleration information to form a state vector containing position, velocity, and acceleration information; Predict the next moment's position and velocity based on the current state and the drone's motion model: In the formula, S t+1|t S is the predicted state vector for the next time step; F is the state transition matrix; S t B is the current state vector; B is the control input matrix; u t This is the control input vector at the current moment; Update state estimation by combining sensor observations: In the formula, S t+1 This is the updated state estimate; K(*) is the Kalman gain, which adjusts the weights of the prediction and observation; z t is the observation vector at the current moment, representing the data acquired from the sensor; H is the observation matrix; The observation residual represents the difference between the prediction and the actual observation. Based on the optimization results, a series of trajectory points are calculated, and a smooth path for the navigation trajectory is generated using Minimum-snap.

6. The UAV dynamic path planning method for strong electromagnetic interference environments according to claim 5, characterized in that, Based on the smooth path of the navigation trajectory, when an obstacle avoidance threshold α is detected... threhold When the obstacle avoidance threshold α is reached, obstacle avoidance is activated. threhold Dynamically adjusted based on electromagnetic field strength: α threhold =5+0.2×(E EM -30),(E EM ≥30dBm) In the formula, E EM This represents the electromagnetic field strength measured in real time.

7. A dynamic path planning system for unmanned aerial vehicles (UAVs), used to execute the dynamic path planning method for UAVs in environments with strong electromagnetic interference as described in any one of claims 1 to 6, characterized in that, include: Electromagnetic shielding multispectral imaging module, used to capture electromagnetic shielding images; A time-frequency dual-domain filtering module is used to suppress noise in the electromagnetic shielding image; The electromagnetic immune feature point cloud generation module is used to extract ORB feature points with electromagnetic immune properties in an electromagnetically shielded image that has been noise-suppressed. The optical flow acceleration constraint module detects dynamic obstacles using optical flow acceleration information. The trajectory generation module is used to fuse ORB feature points and dynamic obstacles to generate a navigation trajectory; Based on the generated navigation trajectory, obstacle avoidance is initiated when an obstacle avoidance threshold is detected.

8. The UAV dynamic path planning system according to claim 7, characterized in that, The electromagnetic shielding multispectral imaging module operates in the visible light 400-700nm and near-infrared 700-1100nm bands, and its shielding effectiveness meets the following requirements: In the formula, f represents the current frequency of the electromagnetic wave; c The cutoff frequency is represented by SE(f); SE(f) is the frequency function of the shielding effectiveness.

9. The UAV dynamic path planning system according to claim 7, characterized in that, The filtering function of the time-frequency dual-domain filtering module is: H(t,w)=G(t)·W(w) In the formula, t is the time variable; w is the angular frequency variable; H(t,w) represents the filtering function of the time-frequency dual-domain filtering module; G(t) represents the time-domain window function; W(w) represents the frequency-domain window function; σ t w is the time-domain Gaussian kernel width. c ω is the cutoff angular frequency; N is the filter order.

Images