Crown shelter area ground point cloud supplementing method and device based on acoustic wave sensor

Abstract

The application relates to a tree crown shelter area ground point cloud supplementing method and device based on a sound wave sensor, and belongs to the technical field of ground three-dimensional measurement and sonar signal processing. The tree crown shelter area ground point cloud supplementing method based on the sound wave sensor comprises the following steps: identifying a ground point cloud missing area caused by tree crown shelter based on an original laser point cloud, then controlling a UAV to fly along a planned flight route according to the ground point cloud missing area, emitting coded sound wave pulses through a sound wave sensor array and recording an original echo sequence; identifying a ground echo through the original echo sequence and outputting time delay information; generating a ground three-dimensional point cloud based on the pose information of the UAV and the time delay information of the ground echo; and performing registration, discrete point elimination and fusion operations on the ground three-dimensional point cloud and the original laser point cloud to fill the ground point cloud missing area. The application can realize high-precision supplementing of the ground point cloud in the tree crown shelter area.

Description

Technical Field

[0001] This invention relates to the field of ground three-dimensional measurement and sonar signal processing technology, and in particular to a method and device for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors. Background Technology

[0002] With the rapid development of airborne LiDAR, terrestrial 3D laser scanning, and oblique photogrammetry technologies, the efficiency of acquiring high-precision terrestrial 3D point cloud data has been significantly improved. This point cloud data has wide and important applications in many fields, including digital elevation model construction, forest biomass estimation, power transmission line inspection, and geological disaster monitoring. However, in forest areas, especially those with dense canopies, obtaining complete surface point clouds has always been a challenge for those skilled in the art. The canopy layer strongly absorbs and scatters light waves, preventing a large amount of light energy from penetrating to the ground. Actual engineering data shows that in natural secondary forests with a canopy closure exceeding 0.7, the missing rate of ground point clouds acquired by airborne LiDAR is often as high as 40% to 60%, forming large areas of data holes.

[0003] To address the problem of missing ground point clouds caused by canopy shading, existing technologies mainly seek breakthroughs in the following directions: The first type of method uses multi-angle scanning or multi-platform data fusion strategies. However, the effect is still limited in areas with high canopy closure, and ground scanning equipment is difficult to move in dense forests. The second type of method uses ground-penetrating radar (GPR), but GPR equipment is bulky and cannot be effectively deployed in areas with complex forest terrain, and the data processing process is cumbersome. The third type of method is to interpolate or complete the existing sparse point cloud. However, this type of method is essentially mathematical or statistical inference rather than actual measurement. When the missing area is large or the terrain is complex, the deviation between the interpolation result and the actual terrain can reach decimeters or even meters.

[0004] In summary, existing methods have drawbacks in addressing the problem of missing ground point clouds due to tree canopy occlusion, including insufficient penetration, low operational efficiency, bulky equipment, complex data interpretation, or poor accuracy in remediation. Summary of the Invention

[0005] In view of this, it is necessary to provide a method and device for ground point cloud supplementation in tree canopy occlusion areas based on acoustic wave sensors, so as to solve the problems of large-area loss of ground point cloud caused by tree canopy occlusion, limited penetration ability of optical and laser means, and insufficient accuracy and efficiency of existing supplementation methods.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides a method for supplementing ground point cloud measurements in tree canopy occlusion areas based on an acoustic sensor, comprising: Based on the original laser point cloud, the missing area of ​​the ground point cloud caused by tree canopy occlusion is identified to determine the range of holes to be measured, and the flight path of the UAV is planned based on the range of holes to be measured, the flight path covering the range of holes to be measured. The drone is controlled to fly along the flight path, and coded acoustic pulses are emitted through the acoustic sensor array mounted on the drone, while multi-channel echo signals are collected simultaneously to record the original echo sequence. The original echo sequence is subjected to pulse compression and peak detection to extract the features of the reflection interface. Based on the features, the ground echo is identified and the time delay information of the ground echo is output. A ground 3D point cloud is generated based on the pose information of the UAV and the ground echo delay information. The ground 3D point cloud is registered with the original laser point cloud, discrete points are removed, and the points are fused to fill in the missing areas of the ground point cloud and output a complete ground point cloud of the canopy occlusion area.

[0007] In one possible implementation, the identification of missing ground point cloud regions caused by canopy occlusion based on the original laser point cloud includes: The original laser point cloud is projected onto a horizontal plane and divided into grids. The number of point clouds in each grid and the elevation difference between the highest and lowest points are counted. If the number of point clouds in any grid is less than the preset number and the elevation difference is greater than the preset difference threshold, it is determined to be an occluded or missing area. Morphological dilation is performed on the grid cells identified as occluded or missing regions to obtain continuous polygons of holes to be measured, which are used as missing regions of the ground point cloud.

[0008] In one possible implementation, the spacing between the flight paths satisfies the azimuth resolution requirement of the synthetic aperture of the acoustic sensor array, as shown in the following formula: Δd≤λ·R / D Where λ is the wavelength of the sound wave, R is the flight height of the UAV above the top of the tree canopy, and D is the effective aperture of the array; The flight speed v of the UAV satisfies the pulse repetition period constraint, as shown in the following formula: v≤Δd / Tprf Where Tprf is the time interval between adjacent pulse transmissions.

[0009] In one possible implementation, the encoded acoustic pulse uses a linear frequency modulated signal, with the time-domain expression as follows:

[0010] Where A is the transmitted amplitude and T is the pulse width. fc is the center frequency, B is the bandwidth, and the acoustic sensor array uses time division multiple access to transmit sequentially. After each sensor transmits, all sensors receive synchronously to form echo data from multiple transmit and receive channels.

[0011] In one possible implementation, the pulse compression is achieved through matched filtering, and each channel performs matched filtering independently to obtain the corresponding compressed range profile, wherein the compressed range profile is obtained by calculating the received echo and a copy of the transmitted signal; The peak detection employs a constant false alarm rate (CFAR) detector and uses a unit average CFAR algorithm on the range image amplitude sequence to detect all peak positions that exceed the adaptive threshold. Each peak corresponds to a two-way delay of a reflective interface.

[0012] In one possible implementation, generating a ground 3D point cloud based on the UAV's pose information and the ground echo delay information includes: Based on the range of the holes to be measured, a ground candidate grid is constructed to form a three-dimensional candidate space; For each transmitting sensor and each receiving sensor at each UAV trackpoint, calculate the theoretical two-way propagation delay from each transmitting sensor and each receiving sensor to each candidate point in the three-dimensional candidate space. Extract the energy value corresponding to the theoretical two-way propagation delay from the compressed range image corresponding to each transmit and receive channel at each track point, and coherently accumulate the energy values ​​extracted from all valid track points and all transmit and receive channels to obtain the accumulated energy of each candidate point. The accumulated energy is compared with a detection threshold. When the accumulated energy is greater than or equal to the detection threshold, the corresponding candidate point is determined to be a ground point, and the three-dimensional coordinates of the candidate point are output to generate a ground three-dimensional point cloud.

[0013] In one possible implementation, the registration, discrete point removal, and fusion operations between the ground 3D point cloud and the original laser point cloud include: The ground 3D point cloud and the original laser point cloud are aligned using coordinate system one, and fine registration is performed using the iterative nearest point algorithm. The coordinate system one uses global coordinate transformation parameters provided by the differential global positioning system and the inertial measurement unit. Each point in the ground 3D point cloud has global coordinates. Calculate the average distance from each point in the ground 3D point cloud to a preset number of nearest neighbor points. If the average distance is greater than a preset multiple of the standard deviation of the global average distance, it is determined to be a discrete point and removed. A fusion region is defined at the boundary of the missing region in the original laser point cloud. For points within the fusion region, the final elevation of the points within the fusion region is obtained by a weighted average of the elevation of the acoustic point cloud and the elevation of the original point cloud, as shown in the following formula:

[0014] The relationship between the weight w and the boundary distance d of the fusion region is w=exp(-d² / (2σ²)), z_final is the final elevation, z_acoustic is the acoustic point cloud elevation, and z_lidar is the original point cloud elevation.

[0015] Secondly, the present invention also provides a ground point cloud supplementation device for tree canopy shading areas based on an acoustic sensor, comprising: The occlusion recognition and flight path planning module is used to identify the missing area of ​​the ground point cloud caused by tree canopy occlusion based on the original laser point cloud, so as to determine the range of holes to be measured, and to plan the flight path of the UAV based on the range of holes to be measured, wherein the flight path covers the range of holes to be measured. The encoding transmission and echo acquisition module is used to control the UAV to fly along the flight path, transmit encoded acoustic pulses through an acoustic sensor array mounted on the UAV, and simultaneously acquire multi-channel echo signals to record the original echo sequence. The echo preprocessing and ground identification module is used to perform pulse compression and peak detection on the original echo sequence to extract the features of the reflection interface, identify the ground echo based on the features, and output the time delay information of the ground echo. A synthetic aperture point cloud reconstruction module is used to generate a ground three-dimensional point cloud based on the pose information of the UAV and the ground echo delay information. The point cloud registration and fusion module is used to register the ground 3D point cloud with the original laser point cloud, remove discrete points, and fuse them to fill in the missing areas of the ground point cloud and output a complete ground point cloud of the canopy occlusion area.

[0016] Thirdly, the present invention also provides an electronic device, including a signal identifier, a memory, and a processor, wherein the signal identifier is used to identify areas of missing ground point cloud caused by canopy occlusion based on the original laser point cloud; The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the above-described method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors.

[0017] Fourthly, the present invention also provides a ground point cloud supplementation system for tree canopy occlusion areas based on acoustic sensors. The ground point cloud supplementation system for tree canopy occlusion areas based on acoustic sensors includes electronic equipment, signal recognition equipment, and result output equipment as described above. The signal recognition equipment and the result output equipment are respectively connected to the electronic equipment. The signal recognition equipment is used to identify the missing area of ​​ground point cloud caused by tree canopy occlusion based on the original laser point cloud. The result output equipment is used to output and display the complete ground point cloud of the tree canopy occlusion area.

[0018] The beneficial effects of this invention are as follows: The ground point cloud supplementation method for tree canopy occlusion areas provided by this invention identifies the missing areas of ground point cloud caused by tree canopy occlusion based on the original laser point cloud, determines the range of holes to be supplemented, and plans the flight path of the UAV based on the range of holes; furthermore, by emitting coded sound pulses and simultaneously acquiring multi-channel echo signals through an array of sound wave sensors mounted on the UAV, the ground under the tree canopy is directly detected by utilizing the penetrating ability of centimeter-level sound wave wavelengths through leaf gaps, fundamentally overcoming the physical limitation that light waves cannot penetrate dense tree canopies; and by performing pulse compression and peak detection on the original echo sequence to extract the reflection boundary... Based on the features of the surface, ground echoes are identified and time delay information is output. A broadband linear frequency modulated signal, after pulse compression, achieves centimeter-level distance resolution, meeting the requirements of high-precision terrain mapping. Based on the UAV's pose information and ground echo time delay information, a bistatic synthetic aperture back projection algorithm is used to generate a 3D ground point cloud. A large-aperture sonar system is equivalently constructed by combining a multi-channel sensor array with the UAV's motion trajectory, and energy coherent accumulation effectively suppresses random scattering noise from the tree canopy. Finally, the 3D ground point cloud is registered with the original laser point cloud, discrete point removal is performed, and fusion operations are conducted to fill in the missing areas of the ground point cloud, outputting a complete ground point cloud of the canopy-covered area. This successfully solves the technical challenges of large-area missing ground point clouds due to canopy occlusion, limited penetration capabilities of optical and laser methods, and insufficient accuracy and efficiency of existing supplementary measurement methods. It achieves high-precision and high-efficiency supplementary measurement of ground point clouds in canopy-covered areas, significantly improving the integrity and reliability of digital terrain modeling. Attached Figure Description

[0019] Figure 1 A flowchart illustrating an embodiment of the ground point cloud supplementation method for tree canopy occlusion areas based on acoustic sensors provided by the present invention. Figure 2 A schematic diagram of the overall process of an embodiment of the ground point cloud supplementation method based on acoustic wave sensor in tree canopy shading area provided by the present invention; Figure 3 A flowchart of the echo preprocessing and ground echo identification algorithm of an embodiment of the ground point cloud supplementation method for tree canopy occlusion area based on acoustic wave sensor provided by the present invention; Figure 4 A schematic diagram illustrating the principle of three-dimensional point cloud reconstruction based on synthetic aperture back projection, according to an embodiment of the ground point cloud supplementation method for tree canopy occlusion area based on acoustic sensor provided by the present invention. Figure 5 A functional block diagram of a ground point cloud supplementation device for tree canopy shading area based on an acoustic sensor provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0022] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0023] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0024] Before demonstrating the embodiments, the following terms will be explained.

[0025] LiDAR (Light Detection and Ranging) is a technology that measures the distance to a target by emitting a laser beam and receiving the reflected signal.

[0026] IMU: Inertial Measurement Unit (IMU) is used to measure the acceleration and angular velocity of an object.

[0027] DGPS: Differential Global Positioning System, which improves GPS positioning accuracy through differential technology.

[0028] LFM: Linear Frequency Modulation (LFM) is a modulation method in which the frequency of a signal changes linearly over time.

[0029] CFAR: Constant False Alarm Rate, an adaptive threshold detection method.

[0030] ICP: Iterative Closest Point (ICP), a point cloud registration algorithm.

[0031] RBF: Radial Basis Function (RBF), a type of basis function used for interpolation.

[0032] TDMA: Time Division Multiple Access, a technology that enables multiple access through time division.

[0033] This invention provides a method and apparatus for supplementing ground point cloud measurements in tree canopy shading areas based on acoustic sensors, which will be described below.

[0034] Figure 1 This is a schematic flowchart of an embodiment of the ground point cloud supplementation method for tree canopy occlusion areas provided by the present invention, as shown below. Figure 1 As shown, the method for re-measuring ground point clouds in tree canopy occlusion areas based on acoustic sensors includes: S101. Based on the original laser point cloud, identify the missing area of ​​the ground point cloud caused by tree canopy occlusion, determine the range of holes to be measured, and plan the flight path of the UAV based on the range of holes to be measured.

[0035] It should be noted that this embodiment first describes the application scenarios involved in the ground point cloud supplementation method based on acoustic sensors for canopy occlusion areas provided by this invention. In application scenarios such as digital terrain modeling, forest resource surveys, power transmission line corridor inspections, and geological disaster monitoring, large-scale missing ground point clouds caused by canopy occlusion have long been a technical bottleneck plaguing the industry. Taking a subtropical evergreen broad-leaved forest area in Yunnan as an example, the canopy closure in this area is about 0.8, with camphor trees and *Symplocos edulis* as the main tree species, with an average tree height of 18m and a gentle slope (slope of about 15°) under the forest canopy. Using a UAV equipped with an airborne LiDAR for scanning, the obtained point cloud data shows that the ground point cloud missing rate reaches 45%, forming dozens of holes with diameters ranging from 5 to 15m, which seriously affects the accuracy of digital elevation model construction. The overall process based on the above scenario can be further referred to Figure 2 As shown.

[0036] In some embodiments, system deployment and calibration are required before applying the above method. A multi-channel acoustic sensor array is installed on the bottom of the UAV, with each sensor in the array capable of independently transmitting and receiving acoustic waves. An inertial measurement unit (IMU) and a differential global positioning system (DGPS) are installed at the center of the array to acquire real-time pose information of the UAV during flight. System calibration is performed in an open, unobstructed area to obtain the precise position coordinates of each sensor relative to the array center and the initial channel gain. The acoustic sensor array has a circular layout, with N≥4 array elements, and the center frequency of each sensor is 20~40kHz, enabling integrated transmission and reception.

[0037] The acoustic sensor array employs an 8-element uniform circular array layout with a ring diameter of 0.3m. Each sensor is a miniature capacitive ultrasonic transducer with a center frequency of 25kHz, a bandwidth of 12kHz, and a maximum sound pressure level of 120dB. All sensors are connected to an onboard embedded processor via a multi-channel synchronous acquisition card, enabling nanosecond-level synchronous triggering of all channels.

[0038] The IMU employs a fiber optic gyroscope or MEMS inertial measurement unit, achieving a heading angle accuracy better than 0.1° and a pitch / roll angle accuracy better than 0.05°. The DGPS uses RTK positioning mode, achieving a horizontal positioning accuracy better than 2cm and an elevation accuracy better than 3cm. IMU and DGPS data are tightly coupled and fused through an extended Kalman filter to output the UAV's centimeter-level pose sequence (position p_i and attitude rotation matrix R_i), with an update frequency of no less than 100Hz.

[0039] The system calibration includes: hovering the UAV at a known height H0 (e.g., 5m) on an open, unobstructed flat ground, with each sensor sequentially emitting calibration pulses and receiving ground echoes. The actual installation height hi = c·τ_i / 2 (where c is the speed of sound, taken as 340m / s) of sensor i is calculated based on the echo delay τ_i. Then, the installation deviation of each sensor relative to the array center is inferred. At the same time, the receiving gain of each channel is recorded for subsequent echo amplitude normalization. For example, a self-designed 8-element acoustic sensor array is installed on the bottom of the drone. The sensors use capacitive ultrasonic transducers with a center frequency of 25kHz, a bandwidth of 12kHz, and a maximum sound pressure level of 120dB@30cm. The eight sensors are evenly distributed on an aluminum alloy ring bracket with a diameter of 0.3m, and each sensor can transmit and receive independently. An inertial measurement unit (IMU) and a differential GPS receiver are installed at the center of the array. The IMU has a heading angle accuracy of 0.1°, a horizontal accuracy of 2cm, and an elevation accuracy of 3cm in DGPS RTK mode.

[0040] All sensors are connected to the onboard embedded processor (32GB RAM) via a 16-channel synchronous acquisition card (200kHz sampling rate, 24-bit resolution). The system weighs approximately 480g and consumes approximately 25W, and can be directly powered by the drone's power supply.

[0041] After completing ground assembly, a wide, flat, hardened surface was selected for calibration. The UAV was hovered 5.00m above the surface (precise measurement), and each sensor was sequentially controlled to emit calibration pulses (LFM signal, fc=25kHz, B=10kHz, T=5ms). The echo delay of each sensor was recorded: theoretically, the delay should be 5.00×2 / 340≈29.41ms. The measured echo delay of sensor A was 29.38ms, and the calculated installation height was 29.38×340 / 2≈4.995m; for sensor B, it was 29.45ms, and the height was 5.007m. From this, it was deduced that sensor A was offset downwards by 0.005m relative to the array center, and sensor B was offset upwards by 0.007m. The installation deviation correction coefficients for each sensor were recorded. Simultaneously, the received amplitude of each channel was measured, normalized, and stored as the initial gain matrix.

[0042] In this embodiment, the missing ground point cloud area caused by tree canopy occlusion is identified based on the original laser point cloud. This process involves projecting the original laser point cloud onto a horizontal plane and dividing it into grids. The number of point clouds in each grid and the elevation difference between the highest and lowest points are counted. If the number of point clouds in a grid is lower than a threshold and the elevation difference is greater than a threshold, it indicates that there is a tree canopy layer but no ground points, and it is determined to be an occluded missing area. Morphological dilation is performed on the missing grid to obtain continuous polygons of holes to be measured.

[0043] In some embodiments, step S101 includes: projecting the original laser point cloud onto a horizontal plane and dividing it into grids, counting the number of point clouds in each grid and the elevation difference between the highest and lowest points; if the number of point clouds in any grid is less than a preset number and the elevation difference is greater than a preset difference threshold, it is determined to be an occluded missing area; performing a morphological dilation operation on the grid determined to be an occluded missing area to obtain continuous polygons of holes to be measured, which are used as missing areas of the ground point cloud.

[0044] Specifically, the raw point cloud data acquired by the airborne LiDAR was imported into processing software to identify missing areas in the ground point cloud. First, the point cloud was projected onto the UTM horizontal plane and divided into 2m × 2m grids. The number of points in each grid was counted: the average number of points per grid in this area was 120 (including tree canopy and ground), but some grids had fewer than 20 points. Further calculation of the elevation difference between the highest and lowest points in these sparse grids was performed: the elevation of the tree canopy top was approximately 210m, and the ground elevation was approximately 192m, a difference of approximately 18m, far exceeding the threshold of 5m. Therefore, these grids were identified as areas with missing ground data due to tree canopy occlusion. The identified missing grids were morphologically dilated (structuring element size 5×5 grid) and connected to form continuous hole polygons. A total of 8 independent holes were identified in this area, with a total area of ​​approximately 4500m². The largest hole was located near the ridge and had a diameter of approximately 15m.

[0045] In some embodiments, the spacing between flight paths satisfies the azimuth resolution requirement of the synthetic aperture of the acoustic sensor array, as shown in the following formula: Δd≤λ·R / D Where λ is the wavelength of the sound wave, R is the flight height of the UAV above the top of the tree canopy, and D is the effective aperture of the array; The flight speed v of the UAV satisfies the pulse repetition period constraint, as shown in the following formula: v≤Δd / Tprf Where Tprf is the time interval between adjacent pulse transmissions.

[0046] For example, λ = c / fc ≈ 1.36 cm @ 25 kHz, R is the flight height of the UAV above the treetop, taken as 3-5 m, and D is the effective aperture of the array, taken as the diameter of the ring, 0.3 m. The calculated Δd ≤ 0.23 m, and considering engineering margin, the flight path spacing is taken as 0.2 m. The flight speed v satisfies the pulse repetition period constraint: v ≤ Δd / Tprf, where Tprf is the time interval between adjacent pulse transmissions, taken as 0.1 s, then v ≤ 2 m / s. Considering both operational efficiency and data quality, the flight speed is taken as 1.5 m / s. Combining the above aperture parameters, the flight path spacing Δd ≤ λ·R / D = 0.0136 × 4 / 0.3 ≈ 0.18 m, taken as 0.2 m. The flight height is set to 3 m above the treetop (since the treetop elevation is approximately 210 m, the UAV flight height is set to 213 m). The flight speed was set at 1.5 m / s, and the pulse repetition interval was 0.2 s (one complete 8-channel TDMA cycle), which met the azimuth sampling requirements. A serpentine flight path was used to cover the area with the largest hole, with a flight path length of approximately 300 m and a flight time of approximately 200 s.

[0047] S102. Control the UAV to fly along the flight path, transmit coded acoustic pulses through the acoustic sensor array mounted on the UAV, and simultaneously collect multi-channel echo signals to record the original echo sequence.

[0048] It should be noted that the encoded acoustic pulse uses a linear frequency modulated signal, and its time-domain expression is:

[0049] Where A is the transmitted amplitude and T is the pulse width. fc is the center frequency, B is the bandwidth, and the acoustic sensor array uses time division multiple access to transmit sequentially. After each sensor transmits, all sensors receive synchronously to form echo data from multiple transmit and receive channels.

[0050] Furthermore, the LFM signal parameters are set as follows: center frequency fc = 25kHz, bandwidth B = 10kHz, and pulse duration T = 5ms. The bandwidth B determines the system's distance resolution Δd = c / (2B) = 340 / (2 × 10000) = 0.017m = 1.7cm, meeting the centimeter-level ground ranging requirement. The product of duration and bandwidth, B·T = 50 ≥ 1, ensuring that the main-sidelobe ratio after pulse compression is better than 30dB.

[0051] The transmission strategy employs Time Division Multiple Access (TDMA): the eight sensors in the array transmit sequentially, with each sensor waiting for a maximum detection time window (corresponding to a maximum detection distance Rmax = 30m, time window Twindow = 2Rmax / c ≈ 0.176s) after transmission before the next sensor begins transmitting. A transmission interval of 0.2s is used to ensure no crosstalk. During each transmission event, all eight sensors receive simultaneously, generating 8 × 8 = 64 transmit-receive channels for echo data.

[0052] The receiver's sampling rate is fs = 200kHz, satisfying the Nyquist sampling theorem (fs > 2fc). Each channel has a recording time of 0.2s, corresponding to a detection depth of 34m. The echo data is stored in real-time to a solid-state drive by an onboard embedded processor.

[0053] For example, the UAV flies automatically along a planned route. The onboard system employs a time-division multiple access (TDMA) transmission strategy: sensors A through 8 sequentially transmit LFM pulses with parameters fc=25kHz, B=10kHz, and T=5ms. After each sensor transmits, all eight sensors simultaneously activate their receivers, with a sampling rate of 200kHz and a recording length of 0.2s. After a complete 8-channel transmission round is completed, there is a 0.1s delay (corresponding to the UAV moving forward 0.15m, meeting the route spacing requirements) before proceeding to the next transmission round.

[0054] The entire flight mission completed approximately 1,000 transmission rounds, with each round generating echo data from 64 channels, totaling approximately 1000 × 64 × 40000 × 2 bytes ≈ 5.12 GB. The data was stored in real time to the onboard solid-state drive.

[0055] After data acquisition, a preliminary analysis was conducted on the echoes from one of the transmitter-receiver combinations (sensor A transmitting, sensor 5 receiving) within a typical hole area (approximately 12m in diameter) of the original point cloud. The original echoes showed a strong direct wave at approximately 5ms (corresponding to a distance of 0.85m), followed by multiple undulations within the range of 12-15ms (corresponding to 2.0-2.6m), corresponding to reflections from the tree canopy. At approximately 29ms (corresponding to 4.93m), there was a medium-intensity echo, which, based on the UAV's flight altitude, was presumably a ground reflection signal. However, its amplitude was much smaller than that of the leaf echoes, indicating significant attenuation of the ground reflection signal.

[0056] S103. Perform pulse compression and peak detection on the original echo sequence to extract the features of the reflection interface, identify the ground echo based on the features, and output the time delay information of the ground echo.

[0057] It should be noted that pulse compression and peak detection are performed on the collected multi-channel echo data to extract all significant reflection interfaces; classification features are constructed, and a pre-trained random forest classifier is used to distinguish between ground echoes and vegetation echoes; direct wave and multipath interference are suppressed, and the output only contains effective time delay information of ground reflections. For details, please refer to [link / reference needed]. Figure 3 As shown.

[0058] In one embodiment, the pulse compression is achieved through matched filtering, and each channel performs matched filtering independently to obtain the corresponding compressed range image. The compressed range image is obtained by calculating the received echo and the transmitted signal copy. The peak detection adopts a constant false alarm rate detector, and the range image amplitude sequence is processed by a unit average constant false alarm rate algorithm to detect all peak positions that exceed the adaptive threshold. Each peak corresponds to a two-way delay of a reflection interface.

[0059] Specifically, pulse compression is achieved through matched filtering: the received echo r(t) is correlated with the transmitted signal replica s(t) to obtain the compressed range profile y(τ).

[0060] Its equivalent range resolution Δd = c / (2B). Matched filtering is performed independently on each of the 64 channels to obtain the corresponding high-resolution range image.

[0061] Furthermore, peak detection employs a constant false alarm rate (CFAR) detector. Specifically, for the range image amplitude sequence, a cell-averaged CFAR algorithm is used, with a reference window length of 2×20 sampling points (corresponding to a range window of approximately 0.34m), a guard window length of 2×5 sampling points (to avoid target energy leakage affecting noise estimation), and a false alarm probability Pfa set to 10. -4 The detector outputs the locations of all peaks exceeding the adaptive threshold. Each peak corresponds to a two-way time delay τk at a reflecting interface, and the distance dk = c·τk / 2 is then calculated.

[0062] Furthermore, a random forest classifier is used for ground echo identification. The input feature vector x_k of the classifier consists of the following five features: (1) Echo amplitude normalization value: the ratio of the amplitude of the peak value to the maximum amplitude value of the channel; (2) Pulse width: the width (number of sampling points) corresponding to the peak value at -3dB. Leaf echoes are usually wider (more than 5 sampling points) due to their loose structure, while ground echoes are narrower (usually 2 to 3 sampling points); (3) Skewness: the ratio of the third central moment of the distance image waveform to the cube of the standard deviation, reflecting the symmetry of the waveform; (4) Kurtosis: the ratio of the fourth central moment of the distance image waveform to the fourth power of the standard deviation, reflecting the sharpness of the waveform. Ground echoes have higher kurtosis; (5) Time difference with the upper echo: the time interval between the current peak value and the previous peak value. Ground echoes are usually the last significant echo in the receiving direction.

[0063] The training process of the random forest classifier is as follows: Data is collected using the same acoustic acquisition system in unobstructed open areas (such as grassland, bare soil, and paved roads) and areas with known vegetation. Each significant peak is manually labeled with its category (ground / non-ground). A total of 50,000 labeled samples are collected, with a positive-to-negative sample ratio of 1:1. The hyperparameters of the random forest are set as follows: 100 decision trees, a maximum depth of 15, a minimum number of leaf node samples of 5, and a feature subset size of 3 (i.e., randomly selecting 3 feature candidates when splitting each tree). After training, the classifier achieves an accuracy of over 92% in recognizing ground echoes.

[0064] Direct wave and multipath suppression are achieved by setting a fixed time threshold based on the relative geometric relationship between the UAV body and the sensor array, and eliminating echoes (corresponding to direct waves or fuselage reflections) with a distance less than 0.5m. For spurious peaks introduced by multipath effects, the consistency of different transmit-receive channels is used for discrimination: the time delay of a real ground target in multiple receive channels satisfies the bistatic elliptic positioning equation, while multipath signals do not possess this consistency. This can be effectively suppressed through coherent accumulation in subsequent synthetic aperture processing.

[0065] For example, pulse compression is first performed: the echo of each channel is matched and filtered against a copy of the transmitted signal. Taking the echoes transmitted and received by sensor A as an example, after matching filtering, the ground echo (delayed by 29ms) that was originally submerged in noise in the range image waveform is compressed into a spike, and the peak signal-to-noise ratio is improved from about 6dB to about 28dB. The range resolution is determined by the pulse length, and the measured distance corresponding to the full width at half maximum (FWHM) is about 0.018m, which is very close to the theoretical value of 1.7cm.

[0066] Next, CFAR peak detection was performed. A reference window length of 40 sampling points (corresponding to a distance of approximately 0.68m) was set, a protection window length of 10 sampling points, and a false alarm probability of 10%. -4Twelve significant peaks were detected in the distance images of channels 1-5 of the sensor, corresponding to distances of 0.85m (direct wave), 2.07m, 2.23m, 2.41m, 2.58m, 2.76m, 2.92m, 3.11m, 3.28m, 3.57m, 4.15m, and 4.93m (ground).

[0067] Feature vectors are extracted from these 12 peaks: taking the ground peak (4.93m) as an example, the amplitude normalization value is 0.62, the pulse width is 2.4 sampling points, the skewness is 0.15, the kurtosis is 3.8, and the time difference with the upper echo (4.15m) is 0.46ms. The input is a pre-trained random forest classifier.

[0068] The training process for the random forest is as follows: In an unobstructed open area of ​​a eucalyptus plantation (canopy closure 0.75) in Guangxi Zhuang Autonomous Region, acoustic data were collected using the same system, and measured manually with RTK. 15,680 ground echo samples and 14,720 non-ground (vegetation) echo samples were labeled. After extracting the above five features, a random forest model was trained with the following parameters: number of trees 100, maximum depth 15, minimum number of leaf node samples 5, and feature subset size 3. After training, the model achieved a classification accuracy of 93.7%, a recall of 91.2%, and a precision of 94.5% on the independent test set.

[0069] The 12 peak values ​​detected in the above steps were input into the classifier one by one, and classification labels were output. Results: The direct wave (0.85m) was correctly identified as "non-ground"; all leaf echoes (2.07-4.15m) were identified as "non-ground"; the ground echo (4.93m) was correctly identified as "ground". The above process was repeated for all transmit-receive channels in the aperture area, and approximately 8600 effective surface echoes were extracted.

[0070] Simultaneously, direct wave / multipath suppression was performed: all detection points with a distance less than 0.5m were eliminated. Statistical analysis shows that this operation removed approximately 3% of false positive peaks (mainly due to fuselage scattering and electromagnetic crosstalk), with no significant negative impact on signal quality.

[0071] S104. Generate a ground three-dimensional point cloud based on the pose information of the UAV and the ground echo delay information.

[0072] It should be noted that in this embodiment, echo energy is accumulated on the ground candidate grid in the missing area based on the UAV's pose information and ground echo delay information. A ground 3D point cloud is then generated through energy threshold detection. The specific process can be found in [reference needed]. Figure 4 As shown.

[0073] In some embodiments, step S104 includes: constructing a ground candidate grid based on the range of the hole to be measured, forming a three-dimensional candidate space; calculating the theoretical two-way propagation delay from each transmitting sensor and each receiving sensor at each UAV track point to each candidate point in the three-dimensional candidate space; extracting the energy value corresponding to the theoretical two-way propagation delay from the compressed range image corresponding to each transmitting and receiving channel at each track point, coherently accumulating the energy values ​​extracted from all valid track points and all transmitting and receiving channels to obtain the accumulated energy of each candidate point; comparing the accumulated energy with a detection threshold, and when the accumulated energy is greater than or equal to the detection threshold, determining the corresponding candidate point as a ground point, and outputting the three-dimensional coordinates of the candidate point to generate a ground three-dimensional point cloud.

[0074] Specifically, the process of constructing the ground candidate grid is as follows: based on the range of the hole polygon determined in the above steps, the grid is divided in the horizontal plane at intervals of Δxy=0.02m; the initial elevation value of each grid point is taken as the median elevation of the known ground points around the hole, and the search range is expanded by ±0.5m in the Z direction at Δz=0.01m to form a three-dimensional candidate space.

[0075] The process of calculating the bistatic delay involves calculating the theoretical two-way propagation delay for the m-th transmitting sensor (position T_m) and the n-th receiving sensor (position R_n) at the i-th UAV waypoint, and for a point Q in the three-dimensional candidate space:

[0076] The energy accumulation process involves traversing all valid transmit-receive channels (a total of M×N channels, where M is the number of track points and N is the number of sensors; the actual number of channels can reach several thousand), extracting the energy value at the corresponding time delay position from the pulse-compressed range image, and performing coherent accumulation.

[0077] Where y_{i,m,n}(τ) represents the pulse-compressed range profile corresponding to the transmitting sensor m and receiving sensor n at the i-th track point, M is the number of effective track points, N_t is the number of transmitting sensors, and N_r is the number of receiving sensors. After coherent accumulation, the energy I(Q) of the real ground reflection point is enhanced, while random scattering and noise are suppressed due to phase inconsistency.

[0078] The threshold detection and point cloud generation process involves estimating a global threshold λ = 3·σ_n based on the noise level, where σ_n is the standard deviation of the distance image amplitude in target-free areas (e.g., depth > 25m). When I(Q) ≥ λ, a ground reflector is determined to exist at candidate point Q, and the 3D point coordinates Q are output. After detecting all grid points, the ground point cloud of the hole region is obtained.

[0079] In some embodiments, due to the large computational load (up to millions of grid points), a fast back projection algorithm can be used to accelerate the process: the beam weight of each pixel is pre-calculated using the beam directionality of acoustic imaging; a GPU parallel computing architecture is employed, mapping the energy accumulation task of each candidate grid onto an independent CUDA thread block for real-time processing. On an airborne computing platform, the processing time for a 1km² area with a 2cm resolution grid can be controlled within 10 minutes.

[0080] For example, in the area of ​​the hole to be measured (range: horizontal X direction 712300~712350m, Y direction 2765200~2765250m, UTM48N coordinate system), a candidate horizontal grid is constructed with Δxy=0.02m, totaling 2500×2500=6.25×10 6 The initial elevation is taken as the median elevation of known ground points around the hole, which is 192.3m. This is then extended ±0.3m in the Z direction with a increment of Δz = 0.01m (due to relatively small terrain undulations). The actual number of 3D candidate points is approximately 6.25 × 10⁻⁶. 6 ×61≈3.8×10 8 indivual.

[0081] The UAV pose data is obtained using 1,000 waypoints (each waypoint corresponds to one launch round): each waypoint contains position (X, Y, Z) and attitude (roll, pitch, heading), obtained through tight coupling fusion of IMU / DGPS, with a sampling rate of 100Hz. After post-processing and smoothing, the position accuracy is better than 0.02m and the attitude accuracy is better than 0.05°.

[0082] For the i-th track point (i=1~1000), the m-th transmitting sensor (m=1~8), and the n-th receiving sensor (n=1~8), calculate the two-way delay τ_{m,n}(Q) of the 3D candidate point Q, and then extract the energy at that delay from the corresponding pulse compression range image y_{i,m,n}. For each Q, accumulate the energy I(Q) of all channels.

[0083] Due to the enormous computational complexity (approximately 3.8 × 10⁻⁶), 8 (Approximately 2.4 × 10¹³ computations for 1 candidate point × 1000 track points × 64 channels) is achieved using GPU acceleration: the candidate grid is divided into 64 × 64 blocks, and each thread processes the energy accumulation of one candidate point. The total computation time after optimization is approximately 8 minutes.

[0084] Estimating the noise floor σ_n: Select the range image amplitude of an area without ground targets (e.g., depth > 25m), and calculate the standard deviation σ_n ≈ 0.12 (normalized amplitude). Set the detection threshold λ = 3 × σ_n = 0.36. Statistically count candidate points with I(Q) ≥ 0.36, resulting in a total of 127,800 point clouds, with an average point density of approximately 1278 points / m² (127,800 points / 100m² hole area ≈ 1278 points / m²).

[0085] S105. The ground 3D point cloud is registered with the original laser point cloud, discrete points are removed and fused to fill in the missing areas of the ground point cloud and output a complete ground point cloud of the canopy occlusion area.

[0086] It should be noted that in this embodiment, the registration, discrete point removal, and fusion operations between the ground 3D point cloud and the original laser point cloud specifically include: the registration, discrete point removal, and fusion operations between the ground 3D point cloud and the original laser point cloud include: The ground 3D point cloud and the original laser point cloud are aligned using coordinate system one and finely registered using an iterative nearest-neighbor algorithm. The coordinate system one uses global coordinate transformation parameters provided by the differential global positioning system and inertial measurement unit. Each point in the ground 3D point cloud has global coordinates. The average distance from each point in the ground 3D point cloud to a preset number of nearest neighbors is calculated. If the average distance is greater than a preset multiple of the standard deviation of the global average distance, it is determined to be a discrete point and is removed. A fusion region is defined at the boundary of the missing region in the original laser point cloud. For points within the fusion region, the final elevation of the points within the fusion region is obtained by a weighted average of the elevation of the acoustic point cloud and the elevation of the original point cloud.

[0087] Specifically, the coordinate system adopts the global coordinate transformation parameters provided by DGPS / IMU in the above steps, and each point in the acoustic point cloud has accurate UTM coordinates. The coarse registration with the original LiDAR point cloud adopts the Iterative Closest Point (ICP) algorithm, and after fine registration, the average deviation of the overlapping area of ​​the two point clouds is less than 3 cm.

[0088] For example, the generated acoustic point cloud (127,800 points) was registered with the original airborne LiDAR point cloud. Since both had been DGPS positioned, the initial coordinate system deviation was less than 5 cm. ICP fine registration was performed: the overlapping boundary region of the two point clouds (1 m wide) was selected, and after 10 iterations, the average point-to-point distance converged to 2.6 cm.

[0089] Furthermore, outlier removal employs statistical filtering: for each point in the acoustic point cloud, the average distance to its k=20 nearest neighbors is calculated. If this distance is greater than twice the standard deviation of the global average distance, it is identified as an outlier and deleted.

[0090] For example, we set k=20 nearest neighbors and calculate the average neighborhood distance of each point. The global average distance is 0.09m and the standard deviation is 0.04m. Points with an average distance >0.09+2×0.04=0.17m are removed, resulting in approximately 1200 outliers (0.94%), leaving 126600 points.

[0091] Furthermore, radial basis function interpolation is used for smooth fusion: at the boundary of the missing region in the original point cloud, the width of the fusion region is defined as 0.5m. For points within the fusion region, their final elevation z_final is obtained by a weighted average of the acoustic point cloud elevation z_acoustic and the original point cloud elevation z_lidar, as shown in the following formula:

[0092] The relationship between the weight w and the boundary distance d of the fusion region is w=exp(-d² / (2σ²)), z_final is the final elevation, z_acoustic is the acoustic point cloud elevation, and z_lidar is the original point cloud elevation.

[0093] For example, radial basis function interpolation fusion is performed at the boundary of the hole: the fusion buffer width is defined as 0.5m, containing approximately 3,200 original point cloud boundary points and approximately 2,800 acoustic point cloud boundary points. For any point within the buffer, its final elevation is calculated using a weighted average formula, with σ taken as 0.2m. After fusion, the boundary transition is natural, without obvious steps or wrinkles.

[0094] Furthermore, the key parameters involved can be found in the summary table shown in Table 1 below.

[0095] Table 1:

[0096] The final output complete ground point cloud was imported into ArcGIS to generate a digital elevation model (DEM) with a resolution of 0.2m. Compared with manual RTK field measurements (32 checkpoints evenly distributed across the supplementary measurement area), the RMSE for horizontal position was 6.3cm, and the RMSE for elevation was 4.7cm. The elevation interpolation (using Kriging) of the hole area in the original LiDAR point cloud was compared with the RTK measurement, showing an RMSE of 0.31m. This invention's acoustic supplementation method improves elevation accuracy from decimeters to centimeters, an improvement of over 85%. The hole filling rate (acoustic point cloud coverage area / total hole area) reached 87%, with a small number of extremely narrow gaps (width <5cm) on the order of acoustic wavelength, which are indistinguishable but have no significant impact on DEM generation.

[0097] This implementation identifies missing areas in the ground point cloud caused by tree canopy occlusion based on the original laser point cloud, thus determining the range of holes to be measured, and planning the flight path of the UAV based on the hole range. Furthermore, by using an acoustic sensor array mounted on the UAV to emit coded acoustic pulses and simultaneously collect multi-channel echo signals, the ground under the tree canopy is directly detected using the penetrating power of centimeter-wavelength sound waves through leaf gaps, fundamentally overcoming the physical limitation that light waves cannot penetrate dense tree canopies. The original echo sequence is pulse-compressed and peak-detected to extract the features of the reflection interface, and the ground echo is identified based on these features. The system outputs time delay information, and the distance resolution reaches the centimeter level after pulse compression using a broadband linear frequency modulated signal, meeting the requirements of high-precision terrain mapping. Based on the UAV's pose information and ground echo time delay information, a bistatic synthetic aperture back projection algorithm is used to generate a 3D ground point cloud. The multi-channel sensor array is combined with the UAV's motion trajectory to construct an equivalent large-aperture sonar system, and energy coherent accumulation effectively suppresses random scattering noise from the canopy. Finally, the 3D ground point cloud is registered with the original laser point cloud, discrete point removal is performed, and fusion operations are performed to fill in the missing areas of the ground point cloud, outputting a complete ground point cloud of the canopy-covered area. This successfully solves the technical problems of large-area missing ground point clouds caused by canopy occlusion, limited penetration capabilities of optical and laser methods, and insufficient accuracy and efficiency of existing supplementary measurement methods. It achieves high-precision and high-efficiency supplementary measurement of ground point clouds in canopy-covered areas, significantly improving the integrity and reliability of digital terrain modeling.

[0098] Based on the same idea as the canopy-shaded area ground point cloud supplementation method based on acoustic sensors in the above embodiments, this application also provides a canopy-shaded area ground point cloud supplementation device 500 based on acoustic sensors. This canopy-shaded area ground point cloud supplementation device 500 can be used to execute the above-described canopy-shaded area ground point cloud supplementation method based on acoustic sensors. Figure 5 As shown, the module comprises an occlusion recognition and flight path planning module 501, an encoded transmission and echo acquisition module 502, an echo preprocessing and ground recognition module 503, a synthetic aperture point cloud reconstruction module 504, and a point cloud registration and fusion module 505. In some embodiments, the above modules can be programmable software instructions stored in memory and executable by a processor. It is understood that in other embodiments, the above modules can also be program instructions or firmware embedded in a processor.

[0099] The occlusion recognition and flight path planning module 501 is used to identify the missing area of ​​the ground point cloud caused by tree canopy occlusion based on the original laser point cloud, so as to determine the range of holes to be measured, and to plan the flight path of the UAV based on the range of holes to be measured, wherein the flight path covers the range of holes to be measured. The encoding transmission and echo acquisition module 502 is used to control the UAV to fly along the flight path, transmit encoded acoustic pulses through the acoustic sensor array mounted on the UAV, and simultaneously acquire multi-channel echo signals to record the original echo sequence. The echo preprocessing and ground identification module 503 is used to perform pulse compression and peak detection on the original echo sequence to extract the features of the reflection interface, identify the ground echo based on the features, and output the time delay information of the ground echo. The synthetic aperture point cloud reconstruction module 504 is used to generate a ground three-dimensional point cloud based on the pose information of the UAV and the ground echo delay information. The point cloud registration and fusion module 505 is used to register the ground 3D point cloud with the original laser point cloud, remove discrete points, and fuse them to fill in the missing areas of the ground point cloud and output a complete ground point cloud of the canopy occlusion area.

[0100] Please refer to Figure 6 , Figure 6 This is a schematic diagram of an embodiment of the electronic device of this application. In this embodiment of the invention, the electronic device 600 includes a processor 601, a memory 602, a display 603, and a signal recognizer 604. Figure 6 Only some components of the electronic device 600 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0101] In some embodiments, processor 601 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 602 or process data, such as the ground point cloud supplementation method for tree canopy shading area based on acoustic wave sensor in this invention.

[0102] In some embodiments, display 603 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 603 is used to display information from electronic device 600 and to display visual user applications. Components 601-603 of electronic device 600 communicate with each other via a system bus.

[0103] In one embodiment, when the processor 601 executes the ground point cloud supplementation program for the tree canopy occlusion area based on the acoustic sensor stored in the memory 602, the following steps can be implemented: Based on the original laser point cloud, the missing area of ​​the ground point cloud caused by tree canopy occlusion is identified to determine the range of holes to be measured, and the flight path of the UAV is planned based on the range of holes to be measured, the flight path covering the range of holes to be measured. The drone is controlled to fly along the flight path, and coded acoustic pulses are emitted through the acoustic sensor array mounted on the drone, while multi-channel echo signals are collected simultaneously to record the original echo sequence. The original echo sequence is subjected to pulse compression and peak detection to extract the features of the reflection interface. Based on the features, the ground echo is identified and the time delay information of the ground echo is output. A ground 3D point cloud is generated based on the pose information of the UAV and the ground echo delay information. The ground 3D point cloud is registered with the original laser point cloud, discrete points are removed, and the points are fused to fill in the missing areas of the ground point cloud and output a complete ground point cloud of the canopy occlusion area.

[0104] It should be understood that when the processor 601 executes the aero-optical effect sequence image restoration program in the memory 602, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0105] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 600 mentioned. Electronic device 600 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 600 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0106] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0107] The above provides a detailed description of the ground point cloud supplementation method and device based on acoustic sensors for tree canopy shading areas provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for supplementing ground point cloud measurements in tree canopy-shaded areas based on acoustic sensors, characterized in that, include: Based on the original laser point cloud, the missing area of ​​the ground point cloud caused by tree canopy occlusion is identified to determine the range of holes to be measured, and the flight path of the UAV is planned based on the range of holes to be measured, the flight path covering the range of holes to be measured. The drone is controlled to fly along the flight path, and coded acoustic pulses are emitted through the acoustic sensor array mounted on the drone, while multi-channel echo signals are collected simultaneously to record the original echo sequence. The original echo sequence is subjected to pulse compression and peak detection to extract the features of the reflection interface. Based on the features, the ground echo is identified and the time delay information of the ground echo is output. A ground 3D point cloud is generated based on the pose information of the UAV and the ground echo delay information. The ground 3D point cloud is registered with the original laser point cloud, discrete points are removed, and the points are fused to fill in the missing areas of the ground point cloud and output a complete ground point cloud of the canopy occlusion area.

2. The method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors according to claim 1, characterized in that, The method of identifying missing ground point cloud areas caused by canopy occlusion based on original laser point cloud includes: The original laser point cloud is projected onto a horizontal plane and divided into grids. The number of point clouds in each grid and the elevation difference between the highest and lowest points are counted. If the number of point clouds in any grid is less than the preset number and the elevation difference is greater than the preset difference threshold, it is determined to be an occluded or missing area. Morphological dilation is performed on the grid cells identified as occluded or missing regions to obtain continuous polygons of holes to be measured, which are used as missing regions of the ground point cloud.

3. The method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors according to claim 1, characterized in that, The spacing between the flight paths satisfies the azimuth resolution requirement of the synthetic aperture of the acoustic sensor array, as shown in the following formula: Δd≤λ·R / D Where λ is the wavelength of the sound wave, R is the flight height of the UAV above the top of the tree canopy, and D is the effective aperture of the array; The flight speed v of the UAV satisfies the pulse repetition period constraint, as shown in the following formula: v≤Δd / Tprf Where Tprf is the time interval between adjacent pulse transmissions.

4. The method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors according to claim 1, characterized in that, The encoded acoustic pulse uses a linear frequency modulated signal, and its time-domain expression is as follows: Where A is the transmitted amplitude and T is the pulse width. fc is the center frequency, B is the bandwidth, and the acoustic sensor array uses time division multiple access to transmit sequentially. After each sensor transmits, all sensors receive synchronously to form echo data from multiple transmit and receive channels.

5. The method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors according to claim 1, characterized in that, The pulse compression is achieved through matched filtering, and each channel performs matched filtering independently to obtain the corresponding compressed range profile. The compressed range profile is obtained by calculating the received echo and the copy of the transmitted signal. The peak detection employs a constant false alarm rate (CFAR) detector, and uses a unit average CFAR algorithm on the range image amplitude sequence to detect all peak positions that exceed the adaptive threshold. Each peak corresponds to a two-way delay of a reflective interface.

6. The method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors according to claim 1, characterized in that, The generation of a ground 3D point cloud based on the pose information of the UAV and the ground echo delay information includes: Based on the range of the holes to be measured, a ground candidate grid is constructed to form a three-dimensional candidate space; For each transmitting sensor and each receiving sensor at each UAV trackpoint, calculate the theoretical two-way propagation delay from each transmitting sensor and each receiving sensor to each candidate point in the three-dimensional candidate space. Extract the energy value corresponding to the theoretical two-way propagation delay from the compressed range image corresponding to each transmit and receive channel at each track point, and coherently accumulate the energy values ​​extracted from all valid track points and all transmit and receive channels to obtain the accumulated energy of each candidate point. The accumulated energy is compared with a detection threshold. When the accumulated energy is greater than or equal to the detection threshold, the corresponding candidate point is determined to be a ground point, and the three-dimensional coordinates of the candidate point are output to generate a ground three-dimensional point cloud.

7. The method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors according to claim 1, characterized in that, The process of registering, removing discrete points, and fusing the ground 3D point cloud with the original laser point cloud includes: The ground 3D point cloud and the original laser point cloud are aligned using coordinate system one, and fine registration is performed using the iterative nearest point algorithm. The coordinate system one uses global coordinate transformation parameters provided by the differential global positioning system and the inertial measurement unit. Each point in the ground 3D point cloud has global coordinates. Calculate the average distance from each point in the ground 3D point cloud to a preset number of nearest neighbor points. If the average distance is greater than a preset multiple of the standard deviation of the global average distance, it is determined to be a discrete point and removed. A fusion region is defined at the boundary of the missing region in the original laser point cloud. For points within the fusion region, the final elevation of the points within the fusion region is obtained by a weighted average of the elevation of the acoustic point cloud and the elevation of the original point cloud, as shown in the following formula: The relationship between the weight w and the boundary distance d of the fusion region is w=exp(-d² / (2σ²)), z_final is the final elevation, z_acoustic is the acoustic point cloud elevation, and z_lidar is the original point cloud elevation.

8. A ground point cloud supplementation device for tree canopy shading areas based on an acoustic sensor, characterized in that, include: The occlusion recognition and flight path planning module is used to identify the missing area of ​​the ground point cloud caused by tree canopy occlusion based on the original laser point cloud, so as to determine the range of holes to be measured, and to plan the flight path of the UAV based on the range of holes to be measured, wherein the flight path covers the range of holes to be measured. The encoding transmission and echo acquisition module is used to control the UAV to fly along the flight path, transmit encoded acoustic pulses through an acoustic sensor array mounted on the UAV, and simultaneously acquire multi-channel echo signals to record the original echo sequence. The echo preprocessing and ground identification module is used to perform pulse compression and peak detection on the original echo sequence to extract the features of the reflection interface, identify the ground echo based on the features, and output the time delay information of the ground echo. A synthetic aperture point cloud reconstruction module is used to generate a ground three-dimensional point cloud based on the pose information of the UAV and the ground echo delay information. The point cloud registration and fusion module is used to register the ground 3D point cloud with the original laser point cloud, remove discrete points, and fuse them to fill in the missing areas of the ground point cloud and output a complete ground point cloud of the canopy occlusion area.

9. An electronic device, characterized in that, It includes a signal identifier, a memory, and a processor, wherein the signal identifier is used to identify areas of missing ground point cloud caused by canopy occlusion based on the original laser point cloud; The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the method for supplementing ground point cloud measurements in tree canopy occlusion areas based on acoustic sensors as described in any one of claims 1 to 7.

10. A ground point cloud supplementation system for tree canopy shading areas based on acoustic sensors, characterized in that, The ground point cloud supplementation system for tree canopy occlusion areas based on acoustic sensors includes the electronic device, signal recognition device, and result output device as described in claim 9. The signal recognition device and the result output device are respectively connected to the electronic device. The data acquisition device is used to identify the missing area of ​​the ground point cloud caused by tree canopy occlusion based on the original laser point cloud. The result output device is used to output and display the expanded and complete ground point cloud of the tree canopy occlusion area.

Images