Unmanned aerial vehicle laser radar calibration quality control system and method

Abstract

The invention belongs to the technical field of atmospheric environment monitoring, and particularly relates to an unmanned aerial vehicle laser radar calibration quality control system and method, and the system comprises an unmanned aerial vehicle flight platform, a laser radar module, a navigation positioning module, a control communication module, a data processing module and a ground station. The unmanned aerial vehicle flying platform comprises a shockproof holder and a protective pod; the laser radar module comprises a laser radar, a scanner module and a photoelectric detector module. According to the invention, one unmanned aerial vehicle is adopted to carry various different types of laser radars, the extinction coefficient is measured by combining optical beam splitting, signal processing and algorithm inversion technologies, and atmospheric particulate classification is effectively monitored through cooperation of a high-precision optical system and an algorithm; the extinction coefficient and the depolarization ratio work cooperatively, high-precision quantitative analysis and classification of pollutants can be achieved, the combined measurement function of the extinction coefficient and the depolarization ratio is achieved, and high-resolution three-dimensional monitoring of atmospheric particulates is achieved through multi-parameter cooperative inversion of the laser radar.

Description

Technical Field

[0001] The present invention belongs to the technical field of atmospheric environment monitoring, and in particular relates to an unmanned aerial vehicle (UAV) lidar calibration and quality control system and method. Background Art

[0002] UAV lidar calibration and quality control has dynamic and multi-dimensional advantages in the field of atmospheric environment governance. Compared with traditional fixed calibration equipment, the UAV platform greatly reduces the deployment cost, and improves the operation safety and efficiency in complex environments through an autonomous obstacle avoidance flight algorithm.

[0003] The inversion of the extinction coefficient of traditional ground lidar relies on static calibration and cannot adapt to the vibration and temperature drift interference during the dynamic flight of UAVs, resulting in a large measurement error in the depolarization ratio. When the UAV flies in complex airspaces such as high-voltage lines and building clusters, the electromagnetic noise and non-uniform distribution of aerosols will reduce the reliability of the extinction coefficient inversion. The lidar carried by the UAV is affected by the vibration of the motor during flight, resulting in an optical path deviation, and a dynamic compensation algorithm is required to correct the vibration noise. Summary of the Invention

[0004] The purpose of the present invention is to provide an unmanned aerial vehicle lidar calibration and quality control system and method for the problems proposed in the above background art.

[0005] To achieve the above object, the present invention adopts the following technical solutions: An unmanned aerial vehicle lidar calibration and quality control system includes a UAV flight platform, a lidar module, a navigation and positioning module, a control and communication module, and a data processing module;

[0006] The UAV flight platform includes a shockproof cloud platform and a protective pod;

[0007] The lidar module includes a lidar, a scanner module, and a photodetector module. The lidar emits 532nm and 1064nm pulsed lasers for measuring the extinction coefficient and depolarization ratio in atmospheric particulate matter monitoring. The scanner module controls the emission direction and coverage range of the laser beam and realizes multi-dimensional scanning through a rotating mirror. The photodetector module is used to capture the laser signal reflected by the target, convert it into an electrical signal and amplify it to accurately detect weak photon-level signals;

[0008] The navigation and positioning module is used to sense the UAV flight environment, calculate the real-time position of the UAV, and plan the dynamic path of the UAV;

[0009] The control and communication module is used to control the stable flight of the UAV to perform environmental calibration and quality control tasks;

[0010] The data processing module is used to collect the detection information of the lidar module and upload it to the environmental database.

[0011] A calibration quality control method for an unmanned aerial vehicle (UAV) lidar includes the following steps:

[0012] Simulate the actual operation scenario through a preset flight path, collect dynamic point cloud data in combination with the UAV mobile platform, and calibrate the IMU / GNSS parameters of the lidar and the environmental measurement values;

[0013] The lidar emits 532nm and 1064nm pulsed lasers to detect the particulate extinction coefficient and depolarization ratio parameters in the atmospheric environment;

[0014] The data processing unit obtains the measured values of the extinction coefficient and depolarization ratio and uploads them to the environmental database.

[0015] A calibration quality control system and method for a UAV lidar according to the present invention uses a single UAV equipped with multiple different types of lidars, combines optical spectroscopy, signal processing, and algorithm inversion technologies to measure the extinction coefficient, and through the cooperation of a high-precision optical system and algorithms, effectively monitors the classification of atmospheric particulate matter. The coordinated operation of the extinction coefficient and depolarization ratio can achieve high-precision quantitative analysis and classification of pollutants. The combined measurement function of the extinction coefficient and depolarization ratio, through the multi-parameter cooperative inversion of the lidar, realizes the high-resolution three-dimensional monitoring of atmospheric particulate matter. Description of the Drawings

[0016] Figure 1 is a schematic diagram of a calibration quality control system for a UAV lidar provided by the present invention;

[0017] Figure 2 is a flowchart of a calibration quality control method for a UAV lidar provided by the present invention. Detailed Embodiments

[0018] The following embodiments are for illustrative purposes only and are not intended to limit the scope of the present invention.

[0019] Refer to Figure 1 , a calibration quality control system for a UAV lidar includes: a UAV flight platform, a lidar module, a navigation and positioning module, an environmental data acquisition module, a control and communication module, a data processing module, and a ground station;

[0020] The UAV flight platform includes a shock-proof gimbal and an IP65 protection pod;

[0021] The lidar module includes a lidar, a scanner module, and a photodetector module. The lidar emits 532nm and 1064nm pulsed lasers for measuring the extinction coefficient and depolarization ratio in atmospheric particulate matter monitoring. The scanner module controls the emission direction and coverage range of the laser beam, and realizes multi-dimensional scanning through a rotating mirror. The photodetector module is used to capture the laser signal reflected by the target, convert it into an electrical signal and amplify it to accurately detect weak photon-level signals;

[0022] The navigation and positioning module is used to perceive the flight environment of the drone, calculate the real-time position of the drone, and plan the dynamic path of the drone;

[0023] The control and communication module is used to control the stable flight of the drone and execute the environmental calibration quality control task;

[0024] The data processing module is used to collect the detection information of the lidar module and perform verification and processing;

[0025] The ground station is used to receive the measurement results of the extinction coefficient and depolarization ratio and upload them to the environmental database.

[0026] Refer to Figure 2 , a method for calibrating and quality controlling a drone lidar, including: simulating an actual operation scenario through a preset flight path, collecting dynamic point cloud data in combination with a drone mobile platform, and calibrating the IMU / GNSS parameters and environmental measurement values of the lidar; the lidar emits 532nm and 1064nm pulsed lasers for measuring the extinction coefficient and depolarization ratio in atmospheric particulate matter monitoring; the data processing unit obtains the measurement values of the extinction coefficient and depolarization ratio and uploads them to the environmental database.

[0027] In one embodiment, the specific steps for detecting the extinction coefficient are as follows:

[0028] S1. Data acquisition: The lidar emits 532nm and 1064nm pulsed lasers, scans the atmosphere vertically and horizontally, receives the backscattered signal and records its intensity attenuation characteristics with distance, synchronously collects environmental parameters for subsequent data calibration, the environmental parameters include temperature, humidity and air pressure, uses an avalanche photodiode APD detector to receive the signal, and converts the optical signal into a digital signal through a high-speed analog-to-digital converter ADC to generate an original distance-intensity curve;

[0029] S2. Preprocessing signal correction: Perform moving average filtering on the original signal to eliminate high-frequency noise interference such as electronic noise and solar background light, use wavelet transform or Savitzky-Golay filter to separate the effective signal and noise components, eliminate the detector background noise through dark current calibration, and normalize the signal intensity based on a clean atmosphere or a known reflectivity target point;

[0030] S3. Based on the relationship between the backscattered signal intensity and the extinction coefficient, establish the lidar equation:

[0031]

[0032] Where, P t =ρ i ·μ t ·μ ais the atmospheric transmittance, μ a is the extinction coefficient;

[0033] S4. Data calibration: The extinction data is calibrated and verified using the cross-validation method.

[0034] Furthermore, in step S2, the method steps for the Savitzky-Golay filter to separate the effective signal and noise components are as follows:

[0035] S21. Use moving average filtering to suppress high-frequency random noise and retain the low-frequency trend characteristics of the signal. The filter realizes noise suppression and signal feature retention through polynomial fitting and weighted averaging within the sliding window, local polynomial least squares fitting, while eliminating high-frequency noise and maintaining low-frequency trend characteristics such as the peak shape and width of the signal;

[0036] Taking the current data point as the center, define a window containing an odd number of points. The data within the window is used for local polynomial fitting. The window slides point by point to cover the entire signal sequence to ensure dynamic processing ability. The data within the window is least squares fitted with an nth-order polynomial to generate a fitting curve. According to the fitting result, the weight coefficient of each data point in the window is calculated. The weight of the center point is the highest, and the weight decreases towards the edge to suppress high-frequency mutation noise. The original data within the window is convolved with the weight coefficient to output the smoothed value of the current point, realizing noise suppression. High-frequency noise is significantly weakened because of its strong randomness and being judged as "deviating from the polynomial trend" in the fitting; low-frequency signals are retained because they conform to the polynomial trend;

[0037] S22. Eliminate the detector background noise through dark current calibration and apply the distance square 1 / r 2 compensation correction to restore the true scattering intensity;

[0038] Under the condition of no incident light, the dark current output of the detector is measured by multiple samplings, and its statistical mean value is calculated as the background noise reference value. Usually, a Gaussian distribution model is used to fit the noise amplitude distribution. In real-time signal acquisition, the background noise mean value is directly subtracted from the original data to eliminate inherent interferences such as the detector thermal noise and circuit offset, and improve the signal-to-noise ratio. The Savitzky-Golay filter is combined to perform sliding window smoothing on the signal after background subtraction. A window length and a quadratic polynomial fitting are adopted to suppress high-frequency random noise, and the low-frequency trend of the scattered signal is retained by weighted average to avoid the peak broadening caused by traditional mean filtering. According to the Lambert scattering law, the scattered light intensity is inversely proportional to the square of the detection distance. The intensity of the original signal received by the detector is lower than the true value due to distance attenuation, and inverse operation compensation correction is required. According to the Lambert scattering law, the scattered light intensity is inversely proportional to the square of the detection distance. The intensity of the original signal received by the detector is lower than the true value due to distance attenuation, and inverse operation compensation correction is required. For the non-uniform distance sampling scenario, the interpolation method is used to reconstruct the continuous distance function to ensure that the compensation coefficient is updated synchronously with the signal.

[0039] S23. Analyze the signal frequency spectrum characteristics. The fixed noise shows a constant amplitude, and the random noise follows a Gaussian distribution. The noise frequency band is separated by Fourier transform, and a noise power spectral density PSD model is established.

[0040] Perform a fast Fourier transform FFT on the original signal to obtain the signal power spectrum distribution, identify the fixed noise spikes and random noise frequency bands, eliminate the discrete spike components through a notch filter or spectrum zeroing, and retain the high-frequency continuous distribution frequency band for subsequent PSD modeling. Determine the constant amplitude model through the statistical mean value of the discrete spike amplitudes, adopt a Gaussian white noise model, fit the linear attenuation characteristics of the high-frequency band power spectrum, and superimpose the fixed noise and random noise PSDs to construct the total noise power spectrum for the frequency domain optimization of the Savitzky-Golay filter, and dynamically adjust the window length according to the main frequency bandwidth of the noise PSD.

[0041] S24. Use a spatial domain filtering method to eliminate impulse noise, suppress outliers by averaging the pixel gray levels within a local window, weight and average the neighboring pixel values, retain the signal edge features, perform multi-scale decomposition on the signal, filter out high-frequency noise through a hard threshold, and convert the signal to the frequency domain by discrete Fourier transform DFT, truncate the high-frequency noise frequency band, and retain the main frequency components of the effective signal.

[0042] Define a sliding window in the spatial domain. By calculating the median or weighted mean of the pixel grayscale values within the window, replace the abnormal pixel values to avoid the edge blurring caused by traditional mean filtering. Utilize the polynomial fitting property of the Savitzky-Golay filter to perform low-order polynomial fitting on the pixel values within the window, giving higher weight to the central pixel, suppressing noise while maintaining the sharpness of the signal edge. Separate the signal into high-frequency and low-frequency components through wavelet transform or pyramid decomposition. The Savitzky-Golay filter adaptively adjusts the window length for different scales, sets a threshold for the high-frequency subband, sets the coefficients with amplitudes lower than the threshold to zero, directly eliminating high-frequency random noise and retaining the main frequency energy of the effective signal. Perform discrete Fourier transform DFT on the spatially filtered signal, identify the noise frequency band and the main frequency of the effective signal, truncate the high-frequency noise component through a frequency-domain mask, retain the main frequency of the effective signal, and then restore the time-domain signal through inverse DFT to achieve the frequency-domain separation of noise and effective signal.

[0043] S25, in combination with principal component analysis, PCA separates the mixed noise sources, extracts the effective signal components related to the extinction coefficient, and uses Monte Carlo simulation to verify the physical consistency of the denoised signal;

[0044] Perform covariance matrix decomposition on the noisy signal through principal component analysis PCA, project the mixed noise onto the low-variance principal components, and the effective signal is concentrated in the high-variance principal components. Truncate the low-variance noise principal components and reconstruct the signal to eliminate the interference of irrelevant noise. Based on the PCA-reconstructed signal, use the Savitzky-Golay filter to perform multi-scale smoothing on the time-domain signal. According to the physical slow-varying characteristic of the extinction coefficient, select a long window and low-order polynomial to suppress high-frequency noise and retain the low-frequency effective trend. Give higher weight to the central point through polynomial fitting to avoid signal peak broadening and ensure the calculation accuracy of the extinction coefficient. Generate a simulation signal based on the theoretical model of the extinction coefficient, inject Gaussian white noise and impulse noise, repeat the "PCA-SG" joint denoising process, and statistically calculate the root mean square error between the denoised signal and the theoretical value. Adjust the number of principal components retained by PCA and the SG window length through Monte Carlo random sampling to verify the robustness of the denoising result to the parameters.

[0045] S26, introduce a temperature and humidity environmental parameter correction model to reduce the influence of temperature drift on the noise of the photodetector, monitor the signal quality index in real time, and when the coefficient of variation CV < 15%, trigger the adjustment of the adaptive filtering parameters;

[0046] By collecting the temperature and humidity data of the photodetector in real time, a linear / nonlinear correction model for temperature drift-signal baseline offset is established to calibrate the baseline of the original signal. Within the sliding window of SG filtering, the weights of data points are dynamically adjusted according to the temperature and humidity parameters to suppress the interference of outliers caused by environmental fluctuations. The coefficient of variation of the signal is calculated in real time within the sliding window to quantify the noise level and signal stability. A CV threshold is set. When CV≥15% is detected, it is determined that the noise is enhanced, and the filtering parameter adjustment process is triggered. In a high-noise scenario, the window length is increased to enhance the smoothing ability to suppress high-frequency noise. In a low-noise scenario, the window length is shortened to retain the detailed features of the signal.

[0047] S27. Compare the data of Raman lidar and dual-wavelength lidar to verify the consistency of the extinction coefficient inversion results, with the relative error ≤±15%. Through synthetic signal simulation, Gaussian white noise with a known amplitude is added to quantitatively evaluate the improvement effect of the signal-to-noise ratio of the denoising algorithm.

[0048] Apply SG filtering to the original signals of Raman lidar and dual-wavelength lidar respectively to suppress high-frequency noise and retain the low-frequency slow-varying characteristics of the extinction coefficient. Calculate the extinction coefficient based on the filtered data, and statistically calculate the relative error between the two through the cross-validation method. Calculate the extinction coefficient based on the filtered data, and statistically calculate the relative error between the two through the cross-validation method

[0049] S28. The contribution rate of instrument noise usually accounts for 10%-30%, the algorithm model error accounts for 5%-15%, and the environmental interference accounts for 20%-40%. For environments with high humidity >80%, activate the hygroscopic expansion correction factor to reduce the nonlinear interference of aerosol agglomeration effects on the signal.

[0050] In an environment with high humidity (>80%), establish an aerosol particle size-humidity relationship model based on the hygroscopic expansion coefficient to correct the nonlinear interference of agglomeration effects on the light scattering signal. Calibrate the baseline of the original signal before SG filtering, adjust the signal amplitude through the humidity correction factor to suppress the signal distortion caused by humidity, use a long window and a third-order polynomial fitting to improve the fitting ability for the high-order nonlinear noise caused by agglomeration effects, ensure the fidelity of the signal trend, dynamically adjust the window length according to the humidity parameters, balance noise suppression and signal response speed, assign lower weights to data points during high-humidity periods to reduce the interference of expansion effect outliers, quantify the noise contribution through covariance analysis, optimize the SG parameters preferentially for high-humidity environmental interference, and superimpose a secondary residual correction algorithm after SG filtering to eliminate the high-order nonlinear errors remaining from agglomeration effects.

[0051] In one embodiment, the specific steps of the cross-validation method are as follows:

[0052] S41. Based on the relationship between particulate matter mass and β-ray attenuation, compare with the β-ray method to verify the reliability of the extinction coefficient inversion results.

[0053] S42. Use satellite remote sensing data to conduct large-scale vertical profile comparison and evaluate the system error distribution;

[0054] S43. Correct the influence of aerosol hygroscopicity on the extinction coefficient according to real-time temperature and humidity data;

[0055] S44. Adopt multi-wavelength fusion technology to conduct joint inversion of 532nm and 1064nm data, distinguish the contributions of aerosol and molecular scattering, and reduce the inversion uncertainty.

[0056] In one embodiment, the specific steps of the airspace filtering method are as follows:

[0057] Determine the noise-dominated type through histogram statistics and local variance calculation, and select an appropriate filtering algorithm. Salt-and-pepper noise shows isolated high / low gray-level mutations, and Gaussian noise shows uniform random fluctuations;

[0058] Define the kernel size and weight distribution according to the filtering target. For example, all weights in the mean kernel are 1, and the Gaussian kernel is weighted by distance. Align the kernel center with the target pixel, calculate the sum of the products of the pixel values in the neighborhood and the kernel weights, and replace the original pixel value with the result. The calculation formula is as follows:

[0059]

[0060] where η(i,j) is the kernel weight, f(x,y) is the original image pixel, and k is the kernel radius;

[0061] Perform edge compensation on the filtered image, mirror-fill the boundary pixels, and quantify the denoising effect through peak signal-to-noise ratio (PSNR) or structural similarity (SSIM) metrics, and dynamically optimize the kernel parameters.

[0062] For mixed noise, cascade filtering is adopted. Cascade filtering first uses median filtering to remove salt-and-pepper noise, then uses bilateral filtering to combine double weights of spatial proximity and gray-level similarity, and finally uses Gaussian filtering to remove Gaussian noise;

[0063] All weights in the mean filtering kernel are equal, and a 3*3 kernel weight matrix is adopted

[0064] The Gaussian filtering kernel weights are distributed according to the Gaussian filtering function while smoothing the noise while retaining the edges;

[0065] Bilateral filtering combines double weights of spatial proximity and gray-level similarity, and the formula is:

[0066]

[0067] where, δ d controls the spatial attenuation, δ rControl the tolerance of gray-scale difference.

[0068] Furthermore, the specific method for depolarization ratio detection is as follows:

[0069] The lidar emits linearly polarized laser with a wavelength of 532 nm. The linearly polarized laser has horizontal and vertical polarization directions, the pulse energy is in the millijoule level, and the repetition frequency is 10 - 100 Hz. Ensure that it penetrates the atmosphere and excites the backscattering signal. Control the laser divergence angle through a collimating lens, and the laser divergence angle ≤ 0.1 mrad to reduce the signal attenuation caused by beam diffusion. Use a polarization beam splitter to separate the backscattered light into vertical S and horizontal P polarization components, which are respectively received by avalanche photodiodes APDPMT to form two independent signals. Synchronously record the variation of the signal intensities of the two channels with height to generate the original polarization component - distance curve;

[0070] Perform moving average filtering on the original signal to eliminate high-frequency electronic noise and background light interference, eliminate the detector background noise through dark current calibration, and perform distance-squared 1 / r 2 compensation correction. Use a calibration target to measure the gain difference between the two polarization channels, and through the gain ratio G S / G P normalize the original signal to eliminate the system polarization sensitivity deviation;

[0071] The depolarization ratio δ is defined as the ratio of the vertical to the horizontal polarization signal intensities: where, I S and I P are the signal intensities of the two channels, and G S and G P are the gain correction factors;

[0072] For cloud or high-concentration aerosol regions, use the Monte Carlo method to simulate the multiple scattering effect, correct the measured depolarization ratio value, analyze the sensitivity of the cloud droplet size (effective radius 5 - 20 μm) to the depolarization ratio through a semi-analytical Monte Carlo model, and optimize the correction parameters;

[0073] Combine the extinction coefficient data to determine the particle type through threshold judgment;

[0074] Based on the Klett-Fernald algorithm, use the depolarization ratio to constrain the ratio relationship between the backscattering coefficient and the extinction coefficient. For example, the lidar ratio of dust aerosol is 40 - 80 sr. Match the molecular scattering signal through the characteristic height segment to verify the consistency of the inversion results;

[0075] Based on the vertical profile of the backscattering signal, calculate the distribution of the extinction coefficient through iterative calculation. The formula is:

[0076]

[0077] In the formula, λz is the extinction coefficient and Sz is the radar ratio.

[0078] Perform dynamic verification on the inversion accuracy of the depolarization ratio, compare it with Raman-Mie scattering lidar data, verify that the depolarization ratio error ≤ ±0.05, analyze the regional depolarization ratio distribution by combining satellite remote sensing data CALIPSO, and correct the local environmental deviation;

[0079] Real-time monitor the temperature and humidity data. When the humidity > 80%, activate the hygroscopic correction factor to reduce the interference of humidity on the determination of particle shape.

[0080] The specific method for particle determination is based on the depolarization ratio threshold for preliminary classification. When δ < 0.1, the particles are mainly spherical (such as cloud droplets, fog droplets or water-soluble aerosols), and the polarization characteristics of their scattered light are close to isotropic. Combining the vertical distribution of the extinction coefficient (the extinction coefficient in a clean atmosphere < 0.1 km -1 ) can further distinguish clouds from aerosols. When δ > 0.2, the particles show significant non-spherical characteristics (such as dust, ice crystals or volcanic ash), and the depolarization effect of the scattered light is significant. If the extinction coefficient > 0.1 km is simultaneously observed -1 and the vertical gradient > 0.05 km -2 , it is determined as a dust transport layer. When 0.1 ≤ δ ≤ 0.2, it is a mixture of spherical and non-spherical particles (such as the coexistence of dust and aerosols).

[0081] In one embodiment, the specific steps of the dynamic calibration method are as follows:

[0082] Calibration scene setup: Set a standard diffuse reflection target in an interference-free light environment. In this embodiment, a Jingyi 10% / 50% / 80% reflectivity combination board can be used to ensure that the target is perpendicular to the lidar optical axis and the distance is fixed. The perpendicular distance between the target and the lidar optical axis. Synchronously deploy a high-precision rangefinder and temperature and humidity sensors for real-time monitoring of environmental parameters. The accuracy of the high-precision rangefinder is ±1 mm;

[0083] Calibration equipment linkage: Fix the lidar and the rotary table, and synchronously adjust the scanning angle of the lidar through the controller. The scanning angle is horizontal ±30° and vertical ±15° to achieve multi-attitude data acquisition. When integrating multiple sensors, it is necessary to use an ArUco marker board for joint coordinate system alignment to ensure data spatial consistency;

[0084] Polarization channel gain calibration: Emit a 532 nm linearly polarized laser, separate the vertical S and horizontal P polarization signals through a beam splitter, record the original intensity values of the two channels, use a target with a known reflectivity. In this embodiment, a 50% diffuse reflection plate is used to calculate the gain ratio G S / G P = I S / I P, dynamically correct the channel sensitivity difference;

[0085] Geometric parameter dynamic compensation: Continuously change the attitude of the lidar through a rotary table, collect calibration data at different angles, match the measured point cloud with the theoretical model based on the Iterative Closest Point (ICP) algorithm, generate an angle error compensation table, with the horizontal deviation angle ≤ 0.1° and the vertical inclination angle ≤ 0.05°;

[0086] Reflectivity real-time calibration: For different reflectivity targets (10%, 50%, 80%), measure the backscattering signal intensity, fit the reflectivity-signal intensity curve, and dynamically correct the depolarization ratio error in low reflectivity and high reflectivity scenarios through linear interpolation;

[0087] Environmental interference compensation: Real-time collect temperature and humidity data. When the humidity > 80%, activate the hygroscopic correction factor to reduce the influence of humidity on the determination of particle shape. Use a sliding time window of 30 seconds to statistically analyze the signal noise level, and dynamically adjust the Savitzky-Golay window width of the filtering threshold from 3 to 7 points to suppress random noise;

[0088] Integrate a high-precision temperature sensor on the surface of the polarization beam splitter prism to monitor the influence of temperature change on the refractive index in real time. Drive the micro-displacement platform through a PID controller to dynamically compensate for the depolarization ratio deviation caused by temperature drift. Install a temperature sensor on the surface of the polarization beam splitter prism (PBS) and complete thermal coupling calibration. The thin-film type sensor is equipped with 2 components, namely a uniform metal thin film and an anisotropic composite thin film. The thin-film type sensor can be switched according to spherical and non-spherical particles. The uniform metal thin film has high refractive index and uniformity, can accurately capture the symmetric scattering light intensity distribution of spherical particles, and is suitable for aerosol or microsphere particle size analysis. The anisotropic composite thin film identifies the irregular scattering characteristics of non-spherical particles through the difference in directional conductivity to improve the dynamic calibration accuracy of the depolarization ratio

[0089] Multi-source data cross-validation: Synchronously measure the target distance with a high-precision rangefinder to calibrate the lidar ranging deviation. Combine the camera image data and verify the lidar point cloud spatial coordinate accuracy through the Perspective-n-Point (PnP) algorithm.

[0090] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A UAV laser radar calibration quality control system, characterized in that, It includes a drone flight platform, a lidar module, a navigation and positioning module, a control and communication module, a data processing module, and a ground station; The drone flight platform includes a shock-proof gimbal and a protective pod; The lidar module includes a lidar, a scanner module, and a photodetector module. The lidar emits 532nm and 1064nm pulsed lasers for measuring the extinction coefficient and depolarization ratio in atmospheric particulate matter monitoring. The scanner module controls the emission direction and coverage of the laser beam and realizes multi-dimensional scanning through a rotating mirror. The photodetector module is used to capture the laser signal reflected by the target, convert it into an electrical signal and amplify it, and accurately detect weak photon-level signals; The navigation and positioning module is used to sense the drone flight environment, calculate the real-time position of the drone, and plan the dynamic path of the drone; The control and communication module is used to control the stable flight of the drone and execute environmental calibration and quality control tasks; The data processing module is used to collect the detection information of the lidar module and perform verification processing; The ground station is used to receive the measurement results of the extinction coefficient and depolarization ratio and upload them to the environmental database.

2. A calibration quality control method for an unmanned aerial vehicle (UAV) laser radar, which is applied to the UAV laser radar calibration quality control system described in the foregoing claim 1, and is characterized in that, It includes the following steps: Simulate the actual operation scenario through a preset route, combine with the drone mobile platform to collect dynamic point cloud data, and calibrate the IMU / GNSS parameters of the lidar and environmental measurement values; The lidar emits 532nm and 1064nm pulsed lasers to detect the particulate matter extinction coefficient and depolarization ratio parameters in the atmospheric environment; The data processing unit obtains the measurement values of the extinction coefficient and depolarization ratio and uploads them to the environmental database.

3. The method for calibrating and quality controlling a drone lidar according to claim 2, wherein, The specific steps of the extinction coefficient detection are as follows: S1. The lidar emits 532nm and 1064nm pulsed lasers, vertically and horizontally scans the atmosphere, receives and records the attenuation characteristics of the backscattered signal intensity with distance, collects environmental parameters for subsequent data calibration, uses an APD detector to receive the signal, converts the optical signal into a digital signal through a high-speed ADC, and generates an original distance-intensity curve; S2. Perform moving average filtering on the original signal to eliminate high-frequency noise, use a Savitzky-Golay filter to separate the effective signal and noise components, eliminate the detector background noise through dark current calibration, and normalize the signal intensity based on the calibration points; S3. Based on the relationship between the backscattered signal intensity and the extinction coefficient, establish the lidar equation: where P t = P i ·μ t ·μ a is the atmospheric transmittance, and μ(r) is the extinction coefficient; S4. Use the cross-validation method to calibrate and verify the extinction data.

4. The method for calibrating and quality controlling an unmanned aerial vehicle laser radar according to claim 3, wherein The specific steps of the cross-validation method are as follows: Based on the relationship between particulate matter mass and β-ray attenuation, compare and verify the reliability of the extinction coefficient inversion results with the β-ray method; Use satellite remote sensing data for large-scale vertical profile comparison to evaluate the system error distribution; Correct the influence of aerosol hygroscopicity on the extinction coefficient according to real-time temperature and humidity data; Adopt multi-wavelength fusion technology to perform joint inversion of 532nm and 1064nm data to distinguish the contributions of aerosol and molecular scattering.

5. The method for calibrating and quality controlling a drone lidar according to claim 4, wherein The steps for the Savitzky-Golay filter to separate the effective signal and noise components are as follows: Adopt moving average filtering to suppress high-frequency random noise and retain the low-frequency trend characteristics of the signal; The detector background noise is eliminated by dark current calibration, and the distance squared 1 / r 2 compensation correction is applied to restore the true scattering intensity; Analyze the spectral characteristics of the signal, separate the noise frequency band through Fourier transform, and establish a noise power spectral density model; Adopt a spatial domain filtering method to eliminate impulse noise, suppress outliers by averaging the pixel grayscales within a local window, weight-average the neighboring pixel values, retain the signal edge features, perform multi-scale decomposition on the signal, filter out high-frequency noise through a hard threshold, and convert the signal to the frequency domain through discrete Fourier transform, truncate the high-frequency noise band, and retain the main frequency components of the effective signal; Combine principal component analysis to separate the mixed noise sources, extract the effective signal components related to the extinction coefficient, and use Monte Carlo simulation to verify the physical consistency of the signal after denoising; Introduce environmental parameter correction models to reduce the influence of temperature drift on the noise of the photodetector, monitor the signal quality indicators in real time, and trigger the adjustment of adaptive filtering parameters; Compare the data of Raman lidar and dual-wavelength lidar to verify the consistency of the extinction coefficient inversion results, and quantitatively evaluate the signal-to-noise ratio improvement effect of the denoising algorithm through synthetic signal simulation.

6. The method for calibrating and quality controlling an unmanned aerial vehicle laser radar according to claim 5, wherein The specific steps of the spatial domain filtering method are as follows: Determine the dominant type of noise through histogram statistics and local variance calculation, select an appropriate filtering algorithm, define the kernel size and weight distribution according to the filtering target, and align the kernel center with the target pixel; Calculate the sum of the products of the pixel values in the neighborhood and the kernel weights, replace the original pixel value with the obtained result, perform edge compensation on the filtered image, and quantify the denoising effect through the PSNR index.

7. The method for calibrating and quality controlling an unmanned aerial vehicle laser radar according to claim 6, wherein For mixed noise, cascade filtering is adopted. The cascade filtering first uses median filtering to remove salt-and-pepper noise, then uses bilateral filtering to combine the dual weights of spatial proximity and gray similarity, and finally uses Gaussian filtering to remove Gaussian noise.

8. The method for calibrating and quality controlling an unmanned aerial vehicle laser radar according to claim 7, wherein The specific method for depolarization ratio detection is as follows: The lidar emits linearly polarized laser with a wavelength of 532nm in the horizontal and vertical polarization directions, the pulse energy is in the millijoule level, and the repetition frequency is 10 - 100Hz. Ensure that it penetrates the atmosphere and excites the backscattering signal. Control the laser divergence angle through a collimating lens to reduce the signal attenuation caused by beam divergence. Use a polarization beam splitter to separate the backscattered light into vertical and horizontal polarization components, which are received by avalanche photodiodes respectively to form two independent signals. Synchronously record the variation of the signal intensities of the two channels with height to generate the original polarization component - distance curve; Perform moving average filtering or wavelet denoising on the original signal to eliminate high-frequency electronic noise and background light interference, eliminate the detector background noise through dark current calibration, and perform a distance squared 1 / r 2 compensation correction, measure the gain difference between the two polarization channels using a calibration target, and through the gain ratio G S / G P Normalize the original signal to eliminate the system polarization sensitivity deviation. The depolarization ratio δ is defined as the ratio of the vertical to the horizontal polarization signal intensity: where, I S and I P are the signal intensities of the two channels, G S and G P are the gain correction factors; For cloud or high-concentration aerosol regions, use the Monte Carlo method to simulate the multiple scattering effect, correct the measured depolarization ratio value, analyze the sensitivity of the cloud droplet size to the depolarization ratio through a semi-analytical Monte Carlo model, and optimize the correction parameters; Combine the extinction coefficient data, determine the particle type through threshold judgment, based on the Klett-Fernald algorithm, use the depolarization ratio to constrain the ratio relationship between the backscattering coefficient and the extinction coefficient, match the molecular scattering signal through the characteristic height segment, verify the consistency of the inversion results, dynamically calibrate the inversion accuracy of the depolarization ratio, compare with the Raman-Mie lidar data, verify that the depolarization ratio error ≤ ±0.05, combine satellite remote sensing data to analyze the regional depolarization ratio distribution, correct the local environmental deviation, and monitor the temperature and humidity data in real time. When the humidity > 80%, activate the hygroscopic correction factor to reduce the interference of humidity on particle shape determination.

9. The method for calibrating and quality controlling an unmanned aerial vehicle laser radar according to claim 8, wherein, The steps of the dynamic calibration method are as follows: Set a standard diffuse reflection target in a light-free environment, ensuring that the target is perpendicular to the optical axis of the lidar and at a fixed distance. Synchronously deploy a high-precision rangefinder and a temperature and humidity sensor. Fix the lidar to the rotary table, and the controller synchronously adjusts the scanning angle of the lidar. Integrate multiple sensors to align the joint coordinate system using an ArUco marker board. Emit 532 nm linearly polarized laser light, separate the S-vertical and P-horizontal polarization signals through a beam splitter, record the original intensity values of the two channels, use a target with a known reflectivity, and calculate the gain ratio G S / G P =I S / I P , and dynamically correct the sensitivity difference between channels; Continuously change the attitude of the lidar through a rotating table, collect calibration data at different angles, match the measured point cloud with the theoretical model based on the ICP algorithm, generate an angle error compensation table, measure the backscattering signal intensity for different reflectivity targets, fit the reflectivity-signal intensity curve, and dynamically correct the depolarization ratio error in low-reflectivity and high-reflectivity scenarios through linear interpolation method; Real-time collect temperature and humidity data, activate the hygroscopic correction factor when the humidity > 80%, reduce the influence of humidity on the determination of particle shape, use a sliding time window to statistically calculate the signal noise level, dynamically adjust the filtering threshold to suppress random noise, synchronously measure the target distance with a high-precision rangefinder, calibrate the lidar ranging deviation, and combine the camera image data to verify the lidar point cloud spatial coordinate accuracy through the PnP algorithm.

Images