Positioning method and system for detecting rotorcraft based on laser radar

Abstract

The invention provides a rotor aircraft positioning method and system based on laser radar detection, relates to the field of laser radar detection, and solves the technical problems of effective layering and accurate detection and positioning of a rotor aircraft under a low-altitude complex ground clutter background. The method comprises the following steps: layering according to the height of an airspace detected by a laser radar, and scanning the detected airspace by adopting a corresponding preset scanning rule according to a layering result to obtain echo signal data and flight time; calculating a radial wind speed value according to the echo signal data and the flight time; performing error compensation on the radial wind speed value and echo signal data to obtain high-precision radial wind speed data; after sub-data in the high-precision radial wind speed data are mapped, a detection airspace is reconstructed, and a three-dimensional grid wind speed field is obtained; and inputting the data of the three-dimensional grid wind speed field into a joint recognition network to obtain a positioning result of the rotorcraft. The method is used in the process of detecting the rotorcraft by the laser radar.

Description

Technical Field

[0001] This application relates to the field of lidar detection, and in particular to a positioning method and system based on lidar detection of rotorcraft. Background Technology

[0002] Existing technologies typically employ radio frequency radar (RF radar), single-pulse lidar (SPL) detection, or visual recognition methods for detecting and locating low-altitude rotorcraft. Specifically, RF radar determines location by receiving electromagnetic echoes reflected from the target, SPL calculates distance by measuring flight time, and visual recognition determines target position through image feature matching. However, these methods are suitable for ideal scenarios with open airspace and low background interference. They perform poorly in detecting and locating rotorcraft in complex low-altitude clutter environments. The complex low-altitude background makes existing detection methods susceptible to external environmental interference. Therefore, effectively stratifying and accurately detecting and locating rotorcraft in complex low-altitude background interference environments has become a pressing technical problem to be solved. Summary of the Invention

[0003] This application provides a positioning method and system for rotorcraft based on lidar detection, which solves the technical problem that existing technologies are unable to effectively stratify, accurately detect and locate rotorcraft in complex low-altitude clutter environments.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for locating a rotorcraft based on lidar detection is provided, comprising: dividing the airspace detected by the lidar into layers according to its altitude, and scanning the airspace according to the layering results using corresponding preset scanning rules to obtain echo signal data and flight time; calculating radial wind speed values ​​based on the echo signal data and flight time; performing error compensation between the radial wind speed values ​​and the echo signal data to obtain high-precision radial wind speed data; mapping the sub-data within the high-precision radial wind speed data to reconstruct the airspace to obtain a three-dimensional grid wind speed field; and inputting the data of the three-dimensional grid wind speed field into a joint identification network to obtain the location result of the rotorcraft.

[0005] Based on the above technical solutions, the positioning method for rotorcraft based on lidar detection provided in this application constructs a complete technical framework including height-layered scanning, radial wind speed calculation, error compensation, three-dimensional wind field reconstruction, and joint network positioning. It abandons the inefficient method of traditional lidar's uniform scanning across the entire domain and direct input of raw wind speed data for identification. By combining layered scanning with refined wind field processing and neural network technology, it achieves accurate positioning of rotorcraft, thereby improving the overall positioning accuracy, scanning efficiency, and environmental adaptability of lidar for rotorcraft detection. It solves the technical problems of large background interference, high false alarm rate, and waste of scanning resources in traditional detection methods.

[0006] In conjunction with the first aspect mentioned above, one possible implementation involves stratifying the airspace detected by the lidar based on its altitude, including the following steps: acquiring historical activity sample data of the rotorcraft; wherein the historical activity sample data includes several flight time periods and corresponding flight altitudes; performing statistical analysis on the historical activity sample data to obtain the activity frequency of the rotorcraft in different altitude ranges; and based on the activity frequency, using a clustering algorithm to divide the altitude range into a core layer and an extended layer.

[0007] The specific implementation method of height stratification is defined. Based on the activity frequency clustering of historical activity samples of rotorcraft, the core layer and the extension layer are divided. This makes the height stratification data-supported and targeted, avoiding indiscriminate stratification design. It can accurately match the actual activity patterns of rotorcraft, lay the foundation for the implementation of subsequent differentiated scanning rules, effectively improve the utilization rate of scanning resources, and reduce the invalid scanning time in non-core areas.

[0008] In conjunction with the first aspect above, in one possible implementation, the acquisition of high-precision radial wind speed data includes: correcting the radial wind speed value based on the signal-to-noise ratio and intensity in the echo signal data, and in conjunction with pre-calibrated system parameters, to obtain first radial wind speed data; wherein, the pre-calibrated system parameters are inherent parameters of the lidar, including laser wavelength, transmitting telescope aperture, and double-pulse transmission interval; inverting atmospheric parameters based on the attenuation degree and Doppler spectral broadening characteristics in the echo signal data, and performing atmospheric attenuation and turbulence effect compensation on the first radial wind speed data to obtain second radial wind speed data; and performing optimal estimation of the second radial wind speed data using a Kalman filter algorithm based on the noise floor energy of the echo signal data to obtain high-precision radial wind speed data.

[0009] The acquisition of high-precision radial wind speed data is refined. A three-step compensation and correction strategy, consisting of signal-to-noise ratio and intensity correction, atmospheric attenuation and turbulence effect compensation, and optimal estimation using Kalman filtering, is employed to eliminate the influence of echo signal noise, atmospheric environmental interference, and system errors on radial wind speed layer by layer. Compared with a single error correction method, this significantly improves the accuracy and stability of radial wind speed data, providing high-quality basic data for the subsequent reconstruction of the three-dimensional grid wind speed field and ensuring the accuracy of wind field reconstruction and aircraft positioning from the source.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, the method for obtaining the three-dimensional grid wind speed field includes: converting the high-precision radial wind speed data and the corresponding spherical coordinates into a three-dimensional Cartesian coordinate point cloud, and then mapping it to a preset three-dimensional regular grid to form an initial wind speed field; calculating the gradient structure tensor of each grid point in the initial wind speed field, and then calculating the coherence coefficient of the local wind field in the initial wind speed field based on the eigenvalues ​​of the gradient structure tensor.

[0011] Determine whether the coherence coefficient exceeds a preset threshold; if yes, it is determined to be a wind speed abrupt change region; if no, it is determined to be a wind speed level region; use the Kriging interpolation algorithm to reconstruct the data in the wind speed abrupt change region to obtain a first reconstructed wind speed field; use the inverse distance weighted interpolation algorithm to reconstruct the data in the wind speed level region to obtain a second reconstructed wind speed field; fuse the first reconstructed wind speed field and the second reconstructed wind speed field to obtain a three-dimensional grid wind speed field.

[0012] The specific reconstruction process of the 3D mesh wind speed field was clarified. Data standardization was achieved by converting spherical coordinates to Cartesian coordinates. Based on the eigenvalues ​​of the gradient structure tensor, wind speed abrupt / gradual regions were divided. Different reconstruction algorithms, such as Kriging interpolation and inverse distance weighted interpolation, were adopted for different regions. This solved the problems of data distortion in abrupt regions and computational redundancy in gradual regions caused by the single interpolation algorithm in traditional wind field reconstruction. It achieved a refined and high-fidelity reconstruction of the 3D wind field, which can clearly restore the abrupt characteristics of rotorcraft-induced wind and provide accurate wind field data support for joint identification networks.

[0013] In conjunction with the first aspect mentioned above, in one possible implementation, the data of the three-dimensional grid wind speed field is input into a joint recognition network, including: extracting grid data corresponding to wind speed change regions and wind speed level regions in the three-dimensional grid wind speed field to obtain sub-grid wind speed field data for the corresponding regions; associating the sub-grid wind speed field data with the corresponding height intervals of the core layer and the extended layer to obtain associated sub-data; and extracting multi-scale airflow change features of the associated sub-data in the wind speed change region through depthwise separable convolutional branches to obtain a change feature sequence.

[0014] Global wind field distribution features of seed data in areas with gentle wind speeds are extracted by point-by-point convolutional branches to obtain a global feature sequence. The mutation feature sequence and the global feature sequence are input into the feature fusion module of the joint recognition network, and feature fusion is performed by transposed convolution and skip connections to obtain a fused feature map. The fused feature map is input into the multi-task output head of the joint recognition network to obtain the rotorcraft existence determination result and the rotorcraft positioning result.

[0015] The process of inputting 3D mesh wind speed field data into the joint recognition network is constrained. By extracting sub-mesh data from different regions and associating them with height layers, differential features are extracted using a dual-branch approach of depthwise separable convolution and pointwise convolution. Feature fusion is then achieved through transposed convolution and skip connections. This enables the network to specifically capture the multi-scale abrupt changes in rotorcraft-induced wind and the global distribution features of background wind, avoiding the problem of effective features being submerged due to mixed feature input. This improves the targeting and effectiveness of network feature extraction, laying a feature foundation for subsequent accurate judgment and localization.

[0016] In conjunction with the first aspect above, in one possible implementation, the fused feature map is input into the multi-task output head of the joint recognition network, comprising: inputting the fused feature map into the shared feature encoding layer of the multi-task output head, compressing the feature dimension through x fully connected layers to obtain an encoded feature vector; inputting the encoded feature vector in parallel into the classification sub-head and regression sub-head of the multi-task output head; the classification sub-head performing binary classification processing on the encoded feature vector to obtain the existence probability value of the rotorcraft.

[0017] Determine whether the existence probability value is greater than a preset probability threshold; if yes, determine that a rotorcraft exists; if no, determine that a rotorcraft does not exist; the regression subhead performs three-dimensional coordinate regression processing on the encoded feature vector to obtain the initial three-dimensional Cartesian coordinates of the rotorcraft; the initial three-dimensional Cartesian coordinates are clipped and constrained by the preset coordinate range of the three-dimensional grid wind speed field to obtain the positioning coordinates; the determination result of the classification subhead is associated with the positioning coordinates of the regression subhead and output to obtain the positioning result.

[0018] The processing flow of the multi-task output head was refined. Feature dimension compression was achieved through a shared feature encoding layer. A parallel processing mode of classification subhead and regression subhead was adopted. The existence of the aircraft was determined by probability threshold and the positioning coordinates were constrained by the three-dimensional grid coordinate range. This integrated the processing of existence determination and precise positioning. It not only avoids invalid positioning calculations when there is no aircraft, but also eliminates invalid positioning results outside the detection airspace through coordinate constraints, thereby improving the effectiveness and accuracy of the positioning results. At the same time, the parallel processing mode improves the computational efficiency of the network.

[0019] In conjunction with the first aspect above, in one possible implementation, the radial wind speed value is obtained by: performing system delay correction processing on the flight time of the dual pulses to obtain a smoothed dual pulse flight time; wherein the dual pulses include a first pulse and a second pulse; when the ratio of the flight time difference between the first pulse and the second pulse to the transmission interval is within a preset first reasonable range, the spatial sampling point corresponding to the smoothed dual pulse flight time is determined as a valid spatial sampling point, and the flight time difference between the first pulse and the second pulse of the valid spatial sampling point is marked as a valid flight time difference. The initial radial wind speed value of the effective spatial sampling point is calculated based on the laser wind measurement formula; when the absolute value of the initial radial wind speed value is less than the preset upper limit value of the induced wind of the rotorcraft, the initial radial wind speed value is determined to be the radial wind speed value.

[0020] The acquisition of radial wind speed values ​​is precisely limited. The flight time is smoothed by system delay correction, and valid sampling points are selected by verifying the rationality of the flight time difference ratio. Then, the valid radial wind speed values ​​are selected by verifying the physical upper limit of the induced wind of the rotorcraft. The interference caused by system delay, invalid sampling points and abnormal wind speed values ​​is eliminated layer by layer, which ensures the validity and rationality of the radial wind speed values, avoids invalid data from entering the subsequent processing flow, improves the efficiency and quality of the overall data processing, and provides reliable basic data for the acquisition of high-precision radial wind speed data.

[0021] In conjunction with the first aspect mentioned above, in one possible implementation, the method for acquiring echo signal data and flight time includes: according to the scanning rules, traversing each spatial sampling point of the detection airspace by the lidar and emitting a double-pulse laser; collecting the echo generated by the double-pulse laser to obtain echo signal data; and recording the time interval between the emission of the double-pulse laser and the reception of the echo as the flight time of the double pulse.

[0022] The specific methods for acquiring echo signal data and flight time were clarified. Based on the emission, acquisition, and time recording process of dual-pulse laser, the acquisition process of echo signal data and flight time was standardized and repeatable, providing accurate and consistent raw data for subsequent radial wind speed calculation. At the same time, the use of dual-pulse laser improved the identification of echo signals and the measurement accuracy of flight time, reducing the problems of signal ambiguity and large time measurement errors caused by single-pulse laser.

[0023] In conjunction with the first aspect mentioned above, in one possible implementation, the scanning rule is as follows: a preset scanning method is performed on the azimuth and elevation angles; wherein the scanning method is as follows: the core layer is scanned with a first step length and a first frequency, and the extended layer is scanned with a second step length and a second frequency, and the first step length is less than the second step length and the first frequency is higher than the second frequency.

[0024] The scanning rules are specifically defined, and differentiated scanning methods are adopted for the core layer and the extended layer, with small step size and high frequency, and large step size and low frequency. In the core layer of the rotorcraft with high activity frequency, the scanning precision and data density are guaranteed, while in the extended layer with low activity frequency, the scanning coverage is guaranteed and the scanning time is reduced. This achieves a balance between scanning accuracy and scanning efficiency. Compared with uniform scanning of the whole area, it significantly saves scanning time and system resources, while improving the detection accuracy of the core area.

[0025] Secondly, this application provides a positioning system for detecting rotorcraft based on lidar, including: a layered scanning module, a wind speed distribution module, and a rotorcraft judgment module; the layered scanning module divides the airspace detected by lidar into layers according to the height, and scans the detected airspace according to the layered results using corresponding preset scanning rules to obtain echo signal data and flight time.

[0026] The wind speed distribution module calculates the radial wind speed value based on the echo signal data and flight time; performs error compensation between the radial wind speed value and the echo signal data to obtain high-precision radial wind speed data; and maps the sub-data within the high-precision radial wind speed data to reconstruct the detection airspace to obtain a three-dimensional grid wind speed field.

[0027] The rotorcraft determination module inputs the data of the three-dimensional grid wind speed field into the joint identification network to obtain the positioning result of the rotorcraft.

[0028] The method has been modularized and implemented. The functional boundaries of each module are clear and they work together. It can quickly and efficiently execute all the steps of the above positioning method. At the same time, the modular design facilitates the system's debugging, maintenance and upgrade, improves the system's practicality and scalability, and enables the technical effects of the above method to be stably realized in actual engineering.

[0029] This application provides a positioning method and system for rotorcraft based on lidar detection. By using a high-level hierarchical approach based on the historical activity patterns of rotorcraft and combining differentiated scanning rules for the core and extended layers, the system can significantly reduce the scanning time in the extended areas while ensuring the detection precision of the core area. This achieves a reasonable allocation of scanning resources and resolves the contradiction of low accuracy or low efficiency in traditional uniform scanning across the entire area.

[0030] From the standardized acquisition of echo signals and time of flight, to the multiple rounds of validity verification of radial wind speed values, and then to the three-step error compensation correction of high-precision radial wind speed data, the entire process of wind speed data quality control from raw acquisition to fine processing was achieved. Interference caused by the system, environment and noise was eliminated layer by layer, ensuring the high accuracy and high validity of wind speed data, and providing a high-quality data foundation for subsequent wind field reconstruction and positioning.

[0031] From data acquisition and processing to identification, each layer incorporates anti-interference and error correction mechanisms, effectively eliminating the effects of various interference factors such as system delay, atmospheric attenuation, turbulence, signal noise, and background wind. This enables lidar to accurately detect and locate rotorcraft even in complex atmospheric environments, enhancing the environmental adaptability and robustness of the technical solution.

[0032] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0033] Figure 1 A system architecture diagram of a positioning system based on lidar detection of rotorcraft provided in this application embodiment; Figure 2 A flowchart illustrating a positioning method for a rotorcraft based on lidar detection, provided in an embodiment of this application; Figure 3 A flowchart illustrating another positioning method for rotorcraft based on lidar detection, provided in an embodiment of this application; Detailed Implementation

[0034] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0035] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0036] like Figure 1 As shown, this application provides a positioning system for detecting rotorcraft based on lidar, including: a layered scanning module, a wind speed distribution module, and a rotorcraft judgment module; the layered scanning module divides the airspace detected by lidar into layers according to the height, and scans the detected airspace according to the layered results using corresponding preset scanning rules to obtain echo signal data and flight time.

[0037] The wind speed distribution module calculates the radial wind speed value based on the echo signal data and flight time; performs error compensation between the radial wind speed value and the echo signal data to obtain high-precision radial wind speed data; and maps the sub-data within the high-precision radial wind speed data to reconstruct the detection airspace to obtain a three-dimensional grid wind speed field.

[0038] The rotorcraft determination module inputs the data of the three-dimensional grid wind speed field into the joint identification network to obtain the positioning result of the rotorcraft.

[0039] To address the technical problem of effectively stratifying, accurately detecting, and locating rotorcraft in complex environments in existing technologies, this application provides a positioning method for rotorcraft based on lidar detection. The method includes: stratifying the airspace detected by the lidar according to its altitude, and scanning the detected airspace using corresponding preset scanning rules based on the stratification results to obtain echo signal data and flight time; calculating radial wind speed values ​​based on the echo signal data and flight time; performing error compensation between the radial wind speed values ​​and the echo signal data to obtain high-precision radial wind speed data; mapping the sub-data within the high-precision radial wind speed data to reconstruct the detected airspace to obtain a three-dimensional grid wind speed field; and inputting the data from the three-dimensional grid wind speed field into a joint identification network to obtain the positioning result of the rotorcraft.

[0040] like Figure 2 As shown in the embodiment of this application, a positioning method for a rotorcraft based on lidar detection includes: S201. The airspace detected by the lidar is divided into layers based on its altitude, and the detected airspace is scanned according to the corresponding preset scanning rules based on the layering results to obtain echo signal data and flight time.

[0041] The historical activity sample data includes: the rotorcraft during several flight periods and the corresponding flight altitudes.

[0042] In some implementations, the airspace detected by the lidar is layered according to its altitude, including the following steps: acquiring historical activity sample data of the rotorcraft; performing statistical analysis on the historical activity sample data to obtain the activity frequency of the rotorcraft in different altitude ranges; and using a clustering algorithm on the flight altitude based on the activity frequency to divide the altitude range into a core layer and an extended layer.

[0043] It should be noted that when using clustering algorithms to stratify flight altitudes, the historical activity sample data of the rotorcraft is first preprocessed: outlier data is screened and removed, including flight altitudes outside the range of lidar detection airspace, duplicate flight activities, and invalid data with incomplete flight time periods; at the same time, all flight altitudes are uniformly converted to absolute altitudes to eliminate altitude deviations caused by different reference planes. Then, the flight altitude is divided into continuous altitude intervals at 10-meter intervals, and the number of flight activities in each interval is counted and the activity frequency is calculated to form an altitude interval-activity frequency feature dataset.

[0044] It should be noted that when using clustering algorithms to stratify flight altitudes, if there are discrete high-activity altitude ranges in the scene, the DBSCAN density clustering algorithm can be used.

[0045] For example, 1200 historical flight activity samples of rotorcraft were collected, including a specific flight time period and the corresponding relative airport reference plane altitude. After cleaning the raw data, 1080 valid data points were obtained. Then, all flight altitudes were converted to absolute altitudes based on sea level to eliminate reference plane deviation. Next, the 500-meter flight altitude was divided into 50 consecutive intervals at 10-meter intervals. For flight activities spanning multiple intervals, the average altitude was assigned to the corresponding interval. For example, if a flight altitude was between 15 and 25 meters and the average altitude was 20 meters, it would be assigned to the 10-20m interval.

[0046] Statistics show that the cumulative number of activities within the 0-120 meter range was 972, accounting for 90% of the total valid activities; the cumulative number of activities within the 120-500 meter range was 108, accounting for 10% of the total valid activities. The average activity frequency within the 0-120 meter range was calculated to be 85%, and the average activity frequency within the 120-500 meter range was 12%. A K-means algorithm with a cluster size of 2 was selected, using the activity frequency of each range as the feature value for clustering. The algorithm automatically divided all altitude ranges from 0-120 meters into a high-activity-frequency group and all altitude ranges from 120-500 meters into a low-activity-frequency group. The altitude ranges from 0-120 meters corresponding to the high-activity-frequency group were merged into a core layer, and the altitude ranges from 120-500 meters corresponding to the low-activity-frequency group were merged into an extension layer. Based on airport security requirements, the activity frequency threshold of the core layer was set to ≥60%, ensuring coverage of the activity areas of the vast majority of rotorcraft.

[0047] In some implementations, a preset scanning method is performed on the azimuth and elevation angles; wherein the scanning method is: a first step length and a first frequency are performed on the core layer, and a second step length and a second frequency are performed on the extended layer, wherein the first step length is smaller than the second step length and the first frequency is higher than the second frequency.

[0048] It should be noted that the preset scanning method is dynamically preset based on the difference in activity frequencies between the core layer and the extended layer, the accuracy requirements of the detection scene, and the resource constraints of the lidar system. The specific values ​​of the step size and frequency can be adjusted according to the range of the detection airspace, the hardware performance of the lidar, and the prevention and control requirements. For example, in key prevention and control scenarios, the core layer step size can be further reduced and the frequency increased, while in open scenarios, the parameters can be appropriately relaxed.

[0049] For example, taking a low-altitude air defense scenario at an airport as an example, the preset scanning method is as follows: a high-density scan with a first step length of 0.5° and a first frequency of 10Hz is used for the core layer of 0-120 meters to ensure accurate capture of rotorcraft in the area; a low-density scan with a second step length of 2° and a second frequency of 2Hz is used for the extended layer of 120-500 meters.

[0050] In some implementations, the methods for acquiring echo signal data and flight time include: according to the scanning rules, the lidar traverses each spatial sampling point in the detection airspace and emits a double-pulse laser; the echo generated by the double-pulse laser is collected to obtain echo signal data; and the time interval between the emission of the double-pulse laser and the reception of the echo is recorded as the flight time of the double pulse.

[0051] It should be noted that the dual-pulse laser includes a first pulse and a second pulse, with a preset fixed emission interval. The echo signal data includes the temporal intensity sequence of the dual-pulse laser, noise floor energy, signal-to-noise ratio, and Doppler spectral characteristic data. The flight time needs to be corrected for system delay to eliminate system errors such as lidar circuit delay and optical path deviation, resulting in a smoothed dual-pulse flight time.

[0052] It should also be noted that the noise floor energy in the echo signal data is calculated by extracting the pure noise segment of the signal outside the effective echo period of the double pulse; the formula for calculating the noise floor energy is: N is the number of sampling points in the noise segment, and x[n] is the voltage quantization value of the nth sampling point; the formula for calculating the voltage quantization value is: , , The raw digital code output by the ADC. This is the input offset voltage of the ADC. This is the lowest effective bit voltage of the ADC. This represents the full-scale input range of the ADC. Simultaneously, the emission interval of the dual-pulse laser needs to be adapted to the maximum distance in the detection space to avoid time measurement errors caused by overlapping echo signals.

[0053] For example, assuming the application scenario is the control of low-altitude rotorcraft in a factory area, an LD-800 pulsed lidar is used. The detection airspace is the low-altitude range of 0-300 meters in the core area of ​​the factory, with a horizontal azimuth angle of 0°-360° and a pitch angle of 0°-30°. Using a layered differentiated scanning rule, the core control layer is determined to be 0-100 meters, and the extended control layer is 100-300 meters. All calculations are based on the lidar's preset parameters, which are: laser wavelength of 1550nm and dual-pulse emission interval. The laser pulse width is 2ns and the sampling rate is 8µs. The core layer scanning angle step size is 100MHz. The scanning angle step size of the extended layer is The ADC quantization bit depth is 12 bits. It has a voltage of 1V and a time recording accuracy of 1ns.

[0054] Step 1: Pre-calculate the basic parameters and sampling interval. seconds, ADC quantization step size The number of spatial sampling points in the core layer is: the number of azimuth sampling points is There are [number] points, and the number of pitch angle sampling points is [number]. The total number of sampling points in the core layer is [number missing]. indivual.

[0055] Step 2: The lidar divides the detection airspace into spatial sampling points in polar coordinates according to a layered scanning rule. The polar coordinates of each sampling point are represented as (r, θ, ...). Taking a specific spatial sampling point in the core layer as an example, calculate its polar coordinates: Let the sampling point be denoted as , and its azimuth angle be . The elevation angle is 30.2°, the pitch angle θ is 15.1°, and the radial distance r is 85 meters. This has been verified. Points within 0°~360°, θ~30°, and r~100 meters that meet the scanning rules are identified as valid spatial sampling points. After the lidar traverses to the above valid sampling points, it triggers a dual-pulse emission command. The first pulse emission time is... The second pulse emission time is After being collimated by the optical antenna, the laser beam travels along the polar coordinate direction of the sampling point (r=85 meters). The laser beam is emitted from a point with a radius of 30.2° and an angle of θ = 15.1°, with a divergence angle of 0.1 mrad, ensuring that the laser beam accurately covers the sampling point.

[0056] The third step involves acquiring dual-pulse echoes to obtain echo signal data. The echo signal is preprocessed: the APD photodetector at the receiving end captures the echo light signal and converts it into a weak analog current signal; this current signal is amplified into an analog voltage signal by a transimpedance amplifier. After conditioning, the voltage amplitude range is 0.1V-0.8V. The conditioned analog voltage Vcond1 = 0.235V for the first pulse echo and Vcond2 = 0.240V for the second pulse echo. The ADC continuously samples the conditioned analog voltage signal at a sampling rate of 100MHz and a sampling interval of 10ns, with a sampling duration of 10μs, and calculates the quantized voltage value at the nth sampling point.

[0057] Determine the sampling point number of the first pulse echo: the time it takes for the laser to travel to and from that sampling point. Sampling point number That is, the 57th sampling point corresponds to the peak value of the first pulse echo. Digital code Voltage quantization value The echo delay time of the second pulse at the same sampling point is consistent with that of the first pulse. Second pulse emission time Therefore, the peak echo sampling time is 8566.67 ns, and the sampling point number is... After conditioning, the voltage Vcond2 = 0.240V, digital code

[0058] Voltage quantization value .

[0059] The quantized values ​​of all sampling points of the first pulse echo, the quantized values ​​of all sampling points of the second pulse echo, and the polar coordinates of the sampling point are associated and stored to form the echo signal data of the spatial sampling point; the above calculation is repeated to complete the echo acquisition of all spatial sampling points.

[0060] Fourth, the flight time of a dual pulse is defined as the time interval between the transmission and reception of a single pulse. ,in This is the moment of peak echo reception. First pulse flight time. Seconds, second pulse flight time Second.

[0061] The lidar uses the calculated T1=567ns and T2=567ns as the double-pulse flight time of the spatial sampling point, and stores it in association with the echo signal data and the polar coordinates of the sampling point. At the same time, the rationality of the flight time is verified: based on the radial distance r=85 meters, the theoretical flight time is 2r / c≈566.67ns, and the actual recorded flight time is 567ns. Therefore, the time recording accuracy with an error of 0.33ns is less than or equal to 1ns, which meets the requirements.

[0062] Repeat steps two through four. The lidar sequentially traverses all spatial sampling points in the core layer and the extended layer according to the scanning rules. Each sampling point completes the dual-pulse emission, echo acquisition quantization calculation, and flight time calculation, and finally obtains the echo signal data and dual-pulse flight time of all spatial sampling points in the entire detection airspace.

[0063] S202. Calculate the radial wind speed value based on the echo signal data and flight time.

[0064] The dual pulses include a first pulse and a second pulse.

[0065] In some implementations, the radial wind speed value is obtained by: performing system delay correction processing on the flight time of the dual pulses to obtain a smoothed dual pulse flight time; when the ratio of the flight time difference between the first pulse and the second pulse to the transmission interval is within a preset first reasonable range, the spatial sampling point corresponding to the smoothed dual pulse flight time is determined as a valid spatial sampling point, and the flight time difference between the first pulse and the second pulse at the valid spatial sampling point is marked as a valid flight time difference. The initial radial wind speed value of the effective spatial sampling point is calculated based on the laser wind measurement formula; when the absolute value of the initial radial wind speed value is less than the preset upper limit value of the induced wind of the rotorcraft, the initial radial wind speed value is determined to be the radial wind speed value.

[0066] It should be noted that system delay correction is used to eliminate system errors such as lidar circuit delay and optical path deviation. Valid sampling point verification can eliminate invalid time differences caused by echo signal interference. The upper limit of rotor-induced wind physical value is preset based on the aerodynamic characteristics of the rotorcraft and is used to filter abnormal wind speed values ​​that exceed the reasonable range, ensuring that the wind speed data processed subsequently is valid and reasonable.

[0067] It should be noted that the preset first reasonable range is determined through a combination of three steps: theoretical boundary derivation, experimental data calibration, and engineering error compensation. The specific process is as follows: First, the formula for calculating the ratio is defined as follows: ,in The time difference of flight after smoothing the double pulse. The interval between the two pulses is defined as the emission interval. Then, based on the core formula for laser two-pulse wind measurement... ,in, Let be the radial wind speed, and c be the speed of light in air. The theoretical flight time difference for targetless motion is obtained by transformation, resulting in ratio=2. / c, combined with the physical upper limit of rotor-induced wind The absolute value is less than or equal to 35 m / s, and the boundary value of the ratio is derived as follows: .

[0068] Through extensive experiments across multiple scenarios and aircraft models, valid echo data were collected. The ratio samples were calculated, and their mean μ and standard deviation σ were statistically analyzed. The 99% confidence interval [μ-3σ, μ+3σ] was selected as the experimental calibration range. Finally, considering systematic errors such as lidar sampling error and atmospheric turbulence error, a 20% engineering redundancy was added to obtain the optimized range, i.e., the optimized lower limit = (μ-3σ)×(1-20%) and the optimized upper limit = (μ+3σ)×(1+20%). At the same time, this range can be dynamically fine-tuned according to key prevention and control scenarios such as airports and factory areas or open airspace scenarios, so as to ensure that the range can cover the vast majority of valid sampling points and effectively eliminate outliers caused by clutter and multipath interference.

[0069] For example, taking the scenario of low-altitude rotorcraft control in a factory area as an example, the original flight time of the dual pulse is... System fixed delay The interval between the two pulses is 10 ns. The value is 8µs, the moving average window size is 5 sampling points, and the preset first reasonable range is... The physical upper limit of the rotor-induced wind is 35 m / s.

[0070] The first step is system delay correction and smoothing for the dual-pulse time-of-flight. The system delay correction formula is as follows: The flight time after the first pulse correction is The flight time after the second pulse correction is Random noise is filtered out using a 5-point moving average method, the formula is as follows: Five consecutive sets of corrected flight times were selected. The first pulse's five consecutive corrected values ​​were: 556.9998 ns, 556.9999 ns, 557 ns, 557.0001 ns, and 557.0002 ns. The second pulse's five consecutive corrected values ​​were: 557.0014 ns, 557.0015 ns, 557.0016 ns, 557.0017 ns, and 557.0018 ns. The smoothed flight time of the first pulse was... The flight time of the second pulse after smoothing is .

[0071] The second step is to determine the effective spatial sampling points and the effective time-of-flight difference. (Time-of-flight difference) , The first reasonable range is preset to be ,therefore If the result meets the reasonable range requirement, the spatial sampling point corresponding to the smoothed double pulse flight time is determined to be a valid spatial sampling point, and the effective flight time difference of the valid spatial sampling point is marked as [value missing]. .

[0072] The third step is to calculate the initial radial wind speed value based on the laser wind measurement formula. The initial radial wind speed value of this effective spatial sampling point is... .

[0073] Step 4: Verification and final determination of the validity of the initial radial wind speed value. The absolute value of the initial radial wind speed value... Less than the physical upper limit of the induced wind of a rotorcraft Therefore, the initial radial wind speed value is determined as the radial wind speed value.

[0074] S203. Perform error compensation between the radial wind speed value and the echo signal data to obtain high-precision radial wind speed data.

[0075] The pre-calibrated system parameters are inherent parameters of the lidar, including laser wavelength, transmitting telescope aperture, and dual-pulse emission interval.

[0076] In some implementations, the acquisition of high-precision radial wind speed data includes: correcting the radial wind speed value based on the signal-to-noise ratio and intensity of the echo signal data, combined with pre-calibrated system parameters, to obtain first radial wind speed data; inverting atmospheric parameters based on the attenuation degree and Doppler spectral broadening characteristics of the echo signal data, and compensating the first radial wind speed data for atmospheric attenuation and turbulence effects to obtain second radial wind speed data; and using a Kalman filter algorithm to optimally estimate the second radial wind speed data based on the noise floor energy of the echo signal data to obtain high-precision radial wind speed data.

[0077] It should be noted that the formula for calculating the noise floor power is: ;in, This is a sequence of sampling points in the pure noise segment outside the effective echo period. This represents the number of sampling points in the noise segment. The signal-to-noise ratio (SNR) is the ratio of signal strength to the noise floor power; the formula for converting it to decibels is: .

[0078] It should also be noted that the correction of radial wind speed values ​​requires prior calibration using a systematic error model established in both laboratory and field experiments. This model establishes a correspondence between signal-to-noise ratio (SNR), signal strength, correction coefficient, and fixed compensation value. Different SNR and signal strength ranges correspond to different correction coefficients k and fixed error compensation values ​​b. , M represents the total number of sample groups collected within the same signal-to-noise ratio-signal strength range, and i represents the sample number. This is the raw radial wind speed value measured by lidar. This represents the actual wind speed value measured using a reference device. During calibration, the corresponding k and b are first matched based on the signal-to-noise ratio and intensity of the measured echo signal, and then substituted into the calibration formula. ,in, This represents the original effective radial wind speed value. This is the first radial wind speed data obtained after correction.

[0079] It should be noted that the atmospheric extinction coefficient α is derived from a modified lidar equation, and its calculation formula is: Where r is the detection distance, Where is the laser emission power, and A is the effective area of ​​the receiving telescope. Let be the atmospheric backscattering coefficient at a distance r. Let be the measured echo signal intensity at a distance r. The turbulence dissipation rate ε is derived by modifying the empirical relationship between Doppler spectral broadening and turbulence intensity. It is first obtained by performing spectral analysis on the echo signal using FFT transform. Substitute into the calculation formula to solve. This completes the inversion of atmospheric parameters. The calculation formula is: ;in, The standard deviation of the Doppler spectrum. is an empirical constant, and r is the detection distance.

[0080] Atmospheric attenuation and turbulence effect compensation are performed in two steps. The first step is atmospheric attenuation compensation, which is based on the atmospheric extinction coefficient α obtained by inversion, using the formula... Correcting the wind speed measurement deviation caused by signal strength reduction due to atmospheric attenuation, among which The first step is radial wind speed data, where r is the detection distance. The second step is turbulence effect compensation, based on the inverted turbulence dissipation rate ε, combined with the turbulence compensation coefficient calibrated by the system. Substitute into the formula By correcting for the wind speed fluctuations caused by turbulence-induced spectral broadening, the compensated second radial wind speed data is obtained. .

[0081] It should also be noted that, when performing optimal estimation, the noise basis power is used. Determine the process noise covariance Q required for filtering. / The covariance of observation noise R = / And establish the state equation and observation equation for wind speed; where the state equation is: The observation equation is ; Let A be the actual radial wind speed at time k, and let A be the state transition matrix. The actual radial wind speed value at the previous moment. Due to the uncertainty of short-term wind speed fluctuations, This is the second radial wind speed data. For the observation matrix, This is noise during the measurement process. The filtering process is implemented through recursive iteration. First, the current state is predicted based on the estimated value of the previous moment. Then, the predicted value is corrected by combining the observed value with the Kalman gain. After multiple iterations, a stable optimal estimate is obtained, which is the high-precision radial wind speed data.

[0082] For example, taking the control of low-altitude rotorcraft in a factory area as a scenario, an LD-800 pulse lidar is used, with a known original radial wind speed of 30 m / s and an echo signal strength of 55225 m. The noise floor power is 0.04m. The detection range is 85m, the laser emission power is 10W, and the receiving telescope area is 0.01. The atmospheric backscattering coefficient β(r) is Doppler spectrum broadening It is 0.3 m / s.

[0083] The first step is to calculate the first radial wind speed data. The specific process is as follows: First, calculate the signal-to-noise ratio. According to SNR=61.4dB, =55225m The calibration correction coefficient k = 0.995, and the fixed compensation value b = 0.1 m / s; the first radial wind speed data were obtained through calibration calculation. .

[0084] The second step is to calculate the second radial wind speed data. The specific process is as follows: First, retrieve the atmospheric extinction coefficient. ; Calculate atmospheric attenuation compensation Inversion turbulent dissipation rate The second radial wind speed data was obtained using the turbulence compensation coefficient. .

[0085] The third step is to calculate high-precision radial wind speed data. The specific process is as follows: First, calculate the noise covariance. , Then, based on the assumption of short-term stationarity of wind speed, a linear Kalman filter model is constructed, and the parameters are initialized: initial state estimate. Initial covariance = Inferring the current state based on the estimate from the previous time step. , By calculating the Kalman gain Determine the ratio of predicted values ​​to observed values The current optimal estimate of covariance is High-precision radial wind speed data is obtained by performing four rounds of iteration and convergence on the second radial wind speed values ​​at subsequent time points k=2, ...,5. .

[0086] S204. After mapping the sub-data within the high-precision radial wind speed data, the detection airspace is reconstructed to obtain a three-dimensional grid wind speed field.

[0087] The sub-data includes high-precision radial wind speed values, corresponding spatial sampling point coordinates, and inverted atmospheric parameters.

[0088] In some implementations, the acquisition of the three-dimensional mesh wind speed field includes: converting high-precision radial wind speed data and corresponding spherical coordinates into a three-dimensional Cartesian coordinate point cloud, mapping it to a preset three-dimensional regular mesh to form an initial wind speed field; calculating the gradient structure tensor of each grid point in the initial wind speed field, and calculating the coherence coefficient of the local wind field based on the eigenvalues ​​of the tensor; determining whether the coherence coefficient exceeds a preset threshold, if so, determining it as a wind speed abrupt change region, otherwise determining it as a wind speed level region; using the Kriging interpolation algorithm to reconstruct the data in the wind speed abrupt change region to obtain a first reconstructed wind speed field, and using the inverse distance weighted interpolation algorithm to reconstruct the data in the wind speed level region to obtain a second reconstructed wind speed field; and fusing the first reconstructed wind speed field and the second reconstructed wind speed field to finally obtain the three-dimensional mesh wind speed field.

[0089] It should be noted that the preset coherence coefficient threshold is obtained based on the statistical analysis of wind field measured data and is used to distinguish the fluctuation characteristics of background wind and induced wind.

[0090] Compared to traditional single interpolation algorithms, this hierarchical interpolation strategy can more accurately preserve the details of areas with abrupt changes in wind speed, while ensuring computational efficiency in flat areas, thereby improving the reconstruction accuracy of the three-dimensional wind field and the recognizability of target features.

[0091] S205. Input the data of the three-dimensional grid wind speed field into the joint identification network to obtain the positioning result of the rotorcraft.

[0092] The joint recognition network is a deep learning network that fuses two-branch features.

[0093] In some implementations, the input of 3D mesh wind speed field data into a joint recognition network includes: extracting grid data corresponding to wind speed abrupt change regions and wind speed flat regions in the 3D mesh wind speed field to obtain sub-mesh wind speed field data for the corresponding regions; associating the sub-mesh wind speed field data with the corresponding height intervals of the core layer and extended layer to obtain associated sub-data; extracting multi-scale airflow abrupt change features of the associated sub-data in the wind speed abrupt change regions through depthwise separable convolutional branches to obtain abrupt change feature sequence; extracting global wind field distribution features of the associated sub-data in the wind speed flat regions through pointwise convolutional branches to obtain a global feature sequence; inputting the abrupt change feature sequence and the global feature sequence into the feature fusion module of the joint recognition network, and performing feature fusion through transposed convolution and skip connections to obtain a fused feature map; inputting the fused feature map into the multi-task output head of the joint recognition network to obtain the rotorcraft existence determination result and the rotorcraft positioning result.

[0094] The specific implementation of the mutation feature sequence is as follows: First, the wind speed values, gradient tensors, and coherence coefficients of the sub-grids in the wind speed mutation region are input into a depthwise separable convolution branch. This branch consists of multiple parallel 3D depthwise separable convolution modules. Each module uses a 3D depthwise convolution kernel of different sizes. The small-sized convolution kernel captures small-scale airflow mutation features such as rotor blade disturbances, while the large-sized convolution kernel captures large-scale airflow mutation features such as induced wind vortices. After each depthwise convolution, the feature expression is enhanced by batch normalization and ReLU activation function, and then the feature information of the channel dimension is fused by 1×1×1 pointwise convolution. Then, the feature maps of different scales are aggregated by a multi-scale pyramid pooling layer. Finally, the features of each scale are spliced ​​according to the channel dimension to form a multi-scale airflow mutation feature that includes micro-turbulent fluctuations to macro-induced wind fields, thus obtaining the mutation feature sequence.

[0095] The global feature sequence is implemented as follows: First, the average wind speed, wind field gradient, and atmospheric temperature and humidity data of the sub-grid in the region with gentle wind speed are input into a pointwise convolution branch. This branch uses a 1×1×1 three-dimensional pointwise convolution as its core. While maintaining the spatial dimension, it fuses local features from different channels to capture the correlation information of the wind field in the channel dimension. Then, the feature mapping of the entire gentle region is spatially aggregated through a global average pooling layer to extract global statistical features, such as regional average wind speed, wind speed variance, and dominant wind direction. Then, these global statistical features and local channel features are further fused through stacked pointwise convolution layers to enhance the expression of the overall trend of the background wind field. Finally, the features obtained from multiple rounds of pointwise convolution and pooling are concatenated according to the channel dimension to form a global feature sequence containing global wind field average distribution, spatial gradient, and atmospheric background parameter information.

[0096] It should be noted that the mutation feature sequence is implemented based on a multi-scale feature extraction mechanism of depthwise separable convolution, and the global feature sequence is implemented based on a global feature aggregation mechanism of pointwise convolution.

[0097] In some implementations, the fused feature map is input into the multi-task output head of the joint recognition network, including: inputting the fused feature map into the shared feature encoding layer of the multi-task output head, compressing the feature dimension through x fully connected layers to obtain an encoded feature vector; inputting the encoded feature vector in parallel into the classification sub-head and regression sub-head of the multi-task output head; the classification sub-head performs binary classification on the encoded feature vector to obtain the probability value of the rotorcraft's existence; determining whether the probability value of existence is greater than a preset probability threshold; if yes, it is determined that a rotorcraft exists; if no, it is determined that a rotorcraft does not exist; the regression sub-head performs three-dimensional coordinate regression on the encoded feature vector to obtain the initial three-dimensional Cartesian coordinates of the rotorcraft; clipping the initial three-dimensional Cartesian coordinates through a preset coordinate range of the three-dimensional grid wind speed field to obtain the positioning coordinates; and associating the determination result of the classification sub-head with the positioning coordinates of the regression sub-head to obtain the positioning result.

[0098] It should be noted that the fused feature map is obtained by fusing multi-scale mutation features and global background features through transposed convolution and skip connections, and includes local fluctuation details of rotor-induced wind and the overall distribution trend of background wind field.

[0099] It should also be noted that the x-layer fully connected layer of the shared feature encoding layer can be dynamically adjusted according to the complexity of the fused feature map, and a balance needs to be struck between feature compression efficiency and information preservation; the preset probability threshold is set based on the balance between precision and recall of the model training data, and can be adjusted according to the risk level of the scenario; the coordinate clipping constraint is set based on the physical boundary of the probe spatial domain, and is used to filter out abnormal coordinates in the regression subhead output to ensure the rationality of the localization results.

[0100] For example, taking an airport scenario as an example, after fusing the feature map input to the multi-task output head, the feature dimension is compressed from 1024 to 256 through a shared encoding layer of two fully connected layers to obtain the encoded feature vector; the classification sub-head outputs an existence probability value of 0.92, which is greater than the preset threshold of 0.8, indicating the presence of a rotorcraft; the regression sub-head outputs initial three-dimensional Cartesian coordinates of (52.3m, 31.7m, 78.5m), which, after being clipped and constrained by the preset coordinate range (0-500m, 0-500m, 0-500m) of the three-dimensional grid wind speed field, finally results in positioning coordinates of (52m, 32m, 79m), with a deviation of less than 2 meters from the actual target position, achieving the associated output of accurate positioning and existence determination.

[0101] Based on the above technical solution, the positioning method for rotorcraft based on lidar detection provided in this application uses a clustering and hierarchical strategy based on activity frequency and a differentiated scanning strategy. High-density scanning resources are allocated in the core layer of the rotorcraft with high activity frequency to ensure detection accuracy, while low-density scanning is used in the extended layer with low activity frequency to improve overall efficiency. Compared with uniform scanning across the entire domain, this reduces scanning time.

[0102] Dual-pulse laser velocimetry combined with multi-step error compensation reduces wind speed measurement errors, providing high-quality basic data for 3D wind field reconstruction. The layered interpolation strategy for 3D wind field reconstruction employs Kriging interpolation to preserve the detailed features of rotor-induced winds in areas of abrupt wind speed changes, while inverse distance-weighted interpolation is used in flat areas to ensure computational efficiency. This allows the reconstructed 3D wind field to accurately represent both the local fluctuations of induced winds and the global distribution of background winds, providing reliable feature data for target identification.

[0103] In one possible implementation of the embodiments of this application, combined with Figure 2 ,like Figure 3 As shown, the above S204 can be specifically implemented through the following S301-S306, which are explained in detail below: S301. After converting the high-precision radial wind speed data and the corresponding spherical coordinates into a three-dimensional Cartesian coordinate point cloud, it is mapped onto a preset three-dimensional regular grid to form an initial wind speed field.

[0104] The spherical coordinates include the azimuth, elevation, and radial distance detected by the lidar.

[0105] In some implementations, during point cloud mapping, the high-precision radial wind speed value of each Cartesian coordinate point is assigned to its corresponding grid cell. If multiple points fall on the same grid, the average wind speed value is taken as the initial wind speed of that grid, ultimately forming an initial wind speed field covering the entire detection airspace.

[0106] It should be noted that the resolution of the three-dimensional regular mesh needs to be dynamically adjusted according to the angular resolution of the lidar and the detection range. For example, a resolution of 10 meters is used in the core layer of 0-120 meters to retain details, and a resolution of 20 meters is used in the extended layer of 120-500 meters to balance efficiency and accuracy. When transforming coordinates, the installation tilt angle and reference plane deviation of the lidar need to be compensated to avoid mapping errors caused by spatial reference offset.

[0107] For example, the spherical coordinates of a sampling point measured by lidar are: radial distance r = 100 meters, elevation angle θ = 30°, azimuth angle = 100 meters. 45°, converted to Cartesian coordinates is ≈61.2 meters =50 meters; the high-precision radial wind speed at this point is 7.38 m / s. After being mapped to a 3D regular grid with a resolution of 10 meters, it is assigned to grid cells of (60-70 meters, 60-70 meters, 40-50 meters). The initial wind speed value of the grid is formed by averaging the wind speeds of other sampling points within the grid.

[0108] S302. After calculating the gradient structure tensor of each grid point in the initial wind speed field, calculate the coherence coefficient of the local wind field in the initial wind speed field based on the eigenvalues ​​of the gradient structure tensor.

[0109] In some implementations, the partial derivatives of the three-dimensional wind speed in the X, Y, and Z directions are first calculated for each grid point of the initial wind speed field using the finite difference method to construct a 3×3 gradient structure tensor. Then, eigenvalue decomposition is performed on this tensor to obtain three eigenvalues ​​arranged in descending order. The largest eigenvalue The intensity reflects the direction of the most drastic change in the local wind field; the coherence coefficient is expressed by the formula... The calculation takes values ​​from 0 to 1. The closer the value is to 1, the stronger the directional coherence of the local wind field, corresponding to areas of sudden wind speed changes. The closer the value is to 0, the more uniform the wind field, corresponding to areas of gentle wind speed.

[0110] The calculation process of the gradient structure tensor is as follows: First, the initial wind speed field is preprocessed with Gaussian smoothing to filter out random noise introduced by discrete sampling and avoid instability in subsequent gradient calculations; then, for each grid point, the first-order partial derivatives of the wind speed in the three orthogonal directions of X, Y, and Z are calculated using the three-dimensional finite difference method to obtain the gradient vector of the local wind speed; based on this, a 3×3 gradient structure tensor is constructed, where each element of the tensor is the spatial average of the product of the partial derivatives in the corresponding direction in the local neighborhood, and finally, a gradient structure tensor that can reflect the intensity and direction of local wind field changes is obtained.

[0111] It should be noted that the gradient structure tensor is used to capture the intensity and direction of the local wind field in three-dimensional space, while the coherence coefficient quantifies the directional coherence of the local wind field.

[0112] It should also be noted that when calculating the gradient, the initial wind speed field is first smoothed using Gaussian to filter out noise caused by discrete sampling and avoid instability of the gradient structure tensor. The threshold for the coherence coefficient is based on a large amount of measured wind field data and is used to balance the identification accuracy and computational efficiency of abrupt change regions. It can be dynamically adjusted according to the needs of the scenario.

[0113] S303. Determine whether the coherence coefficient exceeds a preset threshold; if yes, it is determined to be a region of sudden wind speed change, corresponding to the high fluctuation characteristics of rotor-induced wind; if no, it is determined to be a region of gentle wind speed, corresponding to the uniform distribution characteristics of background wind; at the same time, the threshold can be dynamically adjusted according to the risk level of the detection scene.

[0114] It should be noted that the preset threshold is the core criterion for distinguishing between the wind speed change region caused by the induced wind of the rotorcraft and the wind speed level region dominated by the background wind. Its value is set based on the statistical characteristics of the measured wind field data.

[0115] It should also be noted that the preset threshold is based on a large amount of measured wind field data. Statistics show that the coherence coefficient of more than 90% of the rotor-induced wind areas is greater than the preset threshold of 0.7, while the coherence coefficient of the background wind area is mostly less than 0.7. The dynamic adjustment of the threshold can balance the recognition accuracy and the calculation efficiency. For example, in key prevention and control scenarios such as airports, the threshold can be lowered to 0.65 to improve sensitivity, and in open airspace scenarios, it can be raised to 0.75 to reduce the false judgment rate.

[0116] S304. The Kriging interpolation algorithm is used to reconstruct the data in the wind speed change region to obtain the first reconstructed wind speed field.

[0117] In some implementations, spatial correlation analysis is first performed on discrete sampling points in the wind speed change region to construct a semi-variogram model. Then, the kriging weight of each unknown grid point is calculated based on the semi-variogram, and the interpolation result of the grid point is obtained by weighted averaging of the wind speed values ​​of the surrounding sampling points. Finally, the interpolation results of all unknown grid points are integrated to form the first reconstructed wind speed field containing details of local fluctuations in rotor-induced wind.

[0118] It should be noted that the model selection of the semivariogram needs to be determined based on the spatial correlation characteristics of the wind field measured data. For example, the spherical model is more suitable for the short-distance strong correlation characteristics of rotor-induced wind. The computational complexity of Kriging interpolation is slightly higher than that of traditional interpolation algorithms, but by limiting the range of the abrupt change region, the computation time can be controlled while ensuring accuracy.

[0119] For example, taking the gradient structure tensor of a certain lattice point as an example, assume that the decomposition yields eigenvalues... Then the coherence coefficient The value is close to 1, indicating that the grid point is in a region of sudden wind speed change, and Kriging interpolation needs to be used for reconstruction.

[0120] For the wind speed change region within the core layer of 0-120 meters, the coherence coefficient is greater than 0.7. A spherical semi-variogram model is constructed based on 120 discrete sampling points. After calculating the Kriging weights of the unknown grid points, the first reconstructed wind speed field of the region is obtained by interpolation. The results clearly show the vortex structure of the rotor-induced wind and the local wind speed peak, with a deviation of less than 1 m / s from the measured data.

[0121] S305. The inverse distance weighted interpolation algorithm is used to reconstruct the data in the area with gentle wind speed to obtain the second reconstructed wind speed field.

[0122] In some implementations, a certain number of neighborhood sampling points are first selected for each unknown grid point. The weights of the sampling points are calculated based on the distance between them and the unknown grid point, with the closer the sampling points, the greater their weights. Then, the wind speed values ​​of the neighborhood sampling points are weighted and averaged to obtain the interpolation results for the unknown grid points. Finally, the interpolation results of all unknown grid points are integrated to form a second reconstructed wind speed field with a uniform background wind field distribution.

[0123] It should be noted that the power value is determined based on the measured wind field data, which avoids excessive smoothing caused by an excessively small power and prevents local fluctuations caused by an excessively large power. Since the area with gentle wind speed accounts for 80% to 90% of the detection airspace, the inverse distance weighted interpolation can improve the calculation efficiency of the overall wind field reconstruction and the accuracy meets the requirements for expressing the background wind field.

[0124] S306. The first and second reconstructed wind speed fields are merged to obtain a three-dimensional grid wind speed field.

[0125] In some implementations, the corresponding grid values ​​of the first and second reconstructed wind speed fields are directly spliced ​​together based on the grid point region markings. For the boundary grid points of the two regions, a weighted average method is used to smooth the transition. The weight is determined by the coherence coefficient, and the closer to the abrupt change region, the more the weight is biased towards the first reconstructed wind speed field. Finally, a three-dimensional regular grid wind speed field with uniform resolution covering the entire detection airspace is generated.

[0126] It should also be noted that boundary smoothing can eliminate wind field discontinuities caused by regional splicing and improve the physical rationality of the wind field.

[0127] For example, the first reconstructed wind speed field of the 0-120 meter core layer and the second reconstructed wind speed field of the 120-500 meter extension layer are spliced ​​together, and the boundary grid points are smoothed by weighted average. The resulting three-dimensional grid wind speed field clearly shows the vortex structure of the rotor-induced wind and the uniform distribution of the background wind, with an overall wind field error of less than 1.5 m / s.

[0128] Based on the above technical solution, by quantifying the gradient structure tensor and coherence coefficient, the region of sudden wind speed change caused by rotorcraft-induced wind and the region of gentle wind speed dominated by background wind can be automatically and accurately identified. This avoids the inadequacy of traditional single interpolation algorithms in adapting to different wind field characteristics and provides a reliable basis for subsequent accurate reconstruction of the region division.

[0129] Kriging interpolation is used in areas of abrupt wind speed changes to accurately preserve the local fluctuation details of rotor-induced wind, meeting the feature accuracy requirements for target recognition. In areas of moderate wind speed, inverse distance weighted interpolation is employed, significantly improving the reconstruction efficiency of the background wind field. This layered strategy enhances the overall wind field reconstruction efficiency while controlling wind speed errors in the core area, resolving the pain point of traditional single interpolation methods where accuracy and efficiency are mutually exclusive.

[0130] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0131] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A positioning method for a rotorcraft based on lidar detection, characterized in that, include: The airspace detected by the lidar is divided into layers based on its altitude, and the detected airspace is scanned according to the corresponding preset scanning rules based on the layering results to obtain echo signal data and flight time. Calculate the radial wind speed value based on the echo signal data and flight time; The radial wind speed value and the echo signal data are subjected to error compensation to obtain high-precision radial wind speed data. After mapping the sub-data within the high-precision radial wind speed data, the detection airspace is reconstructed to obtain a three-dimensional grid wind speed field; The data of the three-dimensional grid wind speed field is input into the joint identification network to obtain the positioning result of the rotorcraft.

2. The positioning method for a rotorcraft based on lidar detection according to claim 1, characterized in that, The process of stratifying the airspace based on the altitude detected by the lidar includes the following steps: Acquire historical activity sample data of the rotorcraft; wherein, the historical activity sample data includes: several flight time periods and corresponding flight altitudes; Statistical analysis was performed on the historical activity sample data to obtain the activity frequency of rotorcraft in different altitude ranges; Based on the activity frequency, a clustering algorithm is used to divide the flight altitude range into a core layer and an extension layer.

3. The positioning method for a rotorcraft based on lidar detection according to claim 1, characterized in that, The methods for acquiring the high-precision radial wind speed data include: Based on the signal-to-noise ratio and intensity in the echo signal data, and in conjunction with the pre-calibrated system parameters, the radial wind speed value is corrected to obtain the first radial wind speed data; wherein, the pre-calibrated system parameters are the inherent parameters of the lidar. Based on the attenuation level and Doppler spectrum broadening characteristics in the echo signal data, atmospheric parameters are inverted, and atmospheric attenuation and turbulence effect compensation are applied to the first radial wind speed data to obtain the second radial wind speed data. Based on the noise floor energy of the echo signal data, the second radial wind speed data is optimally estimated using the Kalman filter algorithm to obtain high-precision radial wind speed data.

4. The positioning method for a rotorcraft based on lidar detection according to claim 1, characterized in that, The method for obtaining the three-dimensional grid wind speed field includes: After converting the high-precision radial wind speed data and the corresponding spherical coordinates into a three-dimensional Cartesian coordinate point cloud, it is mapped onto a preset three-dimensional regular grid to form an initial wind speed field. After calculating the gradient structure tensor of each grid point in the initial wind speed field, the coherence coefficient of the local wind field in the initial wind speed field is calculated based on the eigenvalues ​​of the gradient structure tensor. Determine whether the coherence coefficient exceeds a preset threshold; if yes, it is determined to be a region of sudden wind speed change; if no, it is determined to be a region of gentle wind speed. The Kriging interpolation algorithm is used to reconstruct the data in the wind speed abrupt change region to obtain the first reconstructed wind speed field; The inverse distance weighted interpolation algorithm is used to reconstruct the data in the region with gentle wind speeds to obtain the second reconstructed wind speed field; The first and second reconstructed wind speed fields are merged to obtain a three-dimensional grid wind speed field.

5. The positioning method for a rotorcraft based on lidar detection according to claim 4, characterized in that, The step of inputting the data of the three-dimensional grid wind speed field into the joint identification network includes: The grid data corresponding to the regions of sudden wind speed change and the regions of gradual wind speed in the three-dimensional grid wind speed field are extracted to obtain the sub-grid wind speed field data of the corresponding regions. The subgrid wind speed field data is correlated with the corresponding height ranges of the core layer and the extended layer to obtain correlated subdata; Multi-scale airflow abrupt change features of correlated sub-data in wind speed abrupt change regions are extracted by depthwise separable convolutional branches to obtain abrupt change feature sequence; The global wind field distribution features of seed data in areas with gentle wind speeds are extracted by point-by-point convolutional branches, resulting in a global feature sequence. The mutation feature sequence and the global feature sequence are input into the feature fusion module of the joint recognition network, and the features are fused through transposed convolution and skip connections to obtain a fused feature map; The fused feature map is input into the multi-task output head of the joint recognition network to obtain the existence determination result and the positioning result of the rotorcraft.

6. The positioning method for a rotorcraft based on lidar detection according to claim 5, characterized in that, The fused feature map is input into the multi-task output head of the joint recognition network, including: The fused feature map is input into the shared feature encoding layer of the multi-task output head, and the feature dimension is compressed through x fully connected layers to obtain the encoded feature vector. The encoded feature vectors are input in parallel into the classification subhead and regression subhead of the multi-task output head; The classifier performs binary classification on the encoded feature vector to obtain the probability value of the existence of the rotorcraft. Determine whether the probability value of existence is greater than a preset probability threshold; if yes, determine that a rotorcraft exists; if no, determine that a rotorcraft does not exist. The regression subhead performs three-dimensional coordinate regression processing on the encoded feature vector to obtain the initial three-dimensional Cartesian coordinates of the rotorcraft. The initial three-dimensional Cartesian coordinates are clipped and constrained by the preset coordinate range of the three-dimensional grid wind speed field to obtain the positioning coordinates; The determination result of the classification subhead is correlated with the positioning coordinates of the regression subhead to obtain the positioning result.

7. The positioning method for a rotorcraft based on lidar detection according to claim 1, characterized in that, The radial wind speed value is obtained through the following methods: The flight time of the dual pulses is subjected to system delay correction processing to obtain a smoothed dual pulse flight time; wherein, the dual pulses include a first pulse and a second pulse; When the ratio of the time difference between the first pulse and the second pulse to the transmission interval is within a preset first reasonable range, the spatial sampling point corresponding to the smoothed double pulse flight time is determined as a valid spatial sampling point, and the time difference between the first pulse and the second pulse at the valid spatial sampling point is marked as the valid time difference. ; The initial radial wind speed value of the effective spatial sampling point is calculated based on the laser wind measurement formula. When the absolute value of the initial radial wind speed is less than the preset upper limit of the physical limit of the induced wind of the rotorcraft, the initial radial wind speed is determined to be the radial wind speed value.

8. The positioning method for a rotorcraft based on lidar detection according to claim 1, characterized in that, The methods for acquiring the echo signal data and time of flight include: According to the scanning rules, the lidar will traverse every spatial sampling point in the detection airspace and emit dual-pulse lasers; The echo generated by the dual-pulse laser is collected to obtain echo signal data; The time interval between the emission of the dual-pulse laser and the reception of the echo is recorded as the flight time of the dual pulse.

9. A positioning method for a rotorcraft based on lidar detection according to claim 1, characterized in that, The scanning rule is as follows: a preset scanning method is performed on the azimuth and elevation angles; wherein, the scanning method is as follows: the core layer is scanned with a first step length and a first frequency, and the extended layer is scanned with a second step length and a second frequency, and the first step length is smaller than the second step length and the first frequency is higher than the second frequency.

10. A positioning system based on lidar detection of rotorcraft, characterized in that, The system includes: a layered scanning module, a wind speed distribution module, and a rotorcraft identification module; The layered scanning module divides the airspace detected by the lidar into layers based on the height of the airspace, and scans the detected airspace according to the layered results using corresponding preset scanning rules to obtain echo signal data and flight time. The wind speed distribution module calculates the radial wind speed value based on the echo signal data and flight time; performs error compensation between the radial wind speed value and the echo signal data to obtain high-precision radial wind speed data; and maps the sub-data within the high-precision radial wind speed data to reconstruct the detection airspace to obtain a three-dimensional grid wind speed field. The rotorcraft determination module inputs the data of the three-dimensional grid wind speed field into the joint identification network to obtain the positioning result of the rotorcraft.

Images