A surface point cloud high-density noise processing method, system, terminal and storage medium of a blue-green laser radar

Abstract

This invention relates to the field of point cloud noise processing technology, and discloses a method, system, terminal, and storage medium for high-density noise processing of surface point clouds using blue-green lidar. The method includes: acquiring blue-green laser reflection echo signals from a target scene area and filtering them to obtain near-target surface point cloud data; adaptively calculating the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and obtaining local spatial distribution features based on all spatial structure features and all density distribution features; encoding all point cloud spatial distribution features in the local spatial distribution features to obtain terrain surface point clouds, and classifying them to obtain target surface point clouds. This invention can automatically identify target point clouds and accurately remove non-target noise point clouds when processing sparse point clouds in large-scale integrated land and water scenes, improving the efficiency and accuracy of point cloud denoising processing.

Description

Technical Field

[0001] This invention relates to the field of point cloud noise processing technology, and in particular to a method, system, terminal, and computer-readable storage medium for high-density noise processing of surface point clouds in blue-green lidar. Background Technology

[0002] Accurate water topographic mapping is a crucial prerequisite for water resource management. This mapping technology can be broadly categorized into three types: shipborne single-beam (or multi-beam) sonar bathymetry, airborne laser bathymetry, and satellite remote sensing data bathymetry. Compared to the limitations of traditional shipborne surveying methods and satellite remote sensing in mixed land-water areas, airborne laser bathymetry (ALB) leverages the transmission properties of water to visible light in the 470nm-680nm wavelength range. Using a 532nm blue-green laser as the active detection band, it enables integrated land-water exploration of nearshore areas such as coastal and inland wetlands. This offers significant advantages in describing the topographic features of water areas and constructing 3D models of water bodies. However, due to environmental interference, system errors, and atmospheric conditions, the target point cloud acquired by the ALB system is relatively sparse, with high spatial noise density near the target surface. This noise severely impacts data quality and subsequent analysis and applications. Therefore, effective removal of noise from the point cloud is necessary during the preprocessing stage.

[0003] However, existing denoising methods face problems such as over-reliance on manual labor, poor denoising effect, and difficulty in achieving a balance between noise removal and feature preservation when processing sparse point clouds in large-scale integrated water and land scenes, resulting in low efficiency and low accuracy of point cloud denoising processing.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and storage medium for high-density noise processing of surface point clouds in blue-green lidar. This invention aims to address the problems of low efficiency and low accuracy in point cloud denoising processing when existing technologies process sparse point clouds in large-scale integrated land and water scenarios, which rely excessively on manual labor and struggle to achieve a balance between noise removal and feature preservation.

[0006] To achieve the above objectives, the present invention provides a method for high-density noise processing of surface point clouds in blue-green lidar, the method comprising the following steps:

[0007] The blue-green laser reflection echo signal of the target scene area is acquired, and the blue-green laser reflection echo signal is filtered to obtain near-target surface point cloud data.

[0008] Adaptive calculation is performed on the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and local spatial distribution features are obtained based on all the spatial structure features and all the density distribution features.

[0009] All point cloud spatial distribution features in the local spatial distribution features are encoded to obtain terrain surface point clouds, and the terrain surface point clouds are classified to obtain target surface point clouds with near-surface noise suppression.

[0010] Optionally, the method for processing high-density noise in surface point clouds of the blue-green lidar, wherein acquiring the blue-green laser reflection echo signal of the target scene area and filtering the blue-green laser reflection echo signal to obtain near-target surface point cloud data specifically includes:

[0011] The blue-green laser reflection echo signal of the target scene area is obtained according to the distance gating strategy, and the blue-green laser reflection echo signal is subjected to near-target surface region distance gating filtering in the laser propagation direction to obtain the target blue-green laser reflection echo signal containing the target surface echo.

[0012] Extreme value sliding is performed on the first and last ends of the target blue-green laser reflected echo signal to obtain an extreme value sequence. The extreme value sequence is then updated to obtain a local extreme value sequence.

[0013] The preset time position of the target blue-green laser reflected echo signal and the extreme position of the local extreme value sequence are used to perform sequence calculation to obtain a local maximum value sequence and a local minimum value sequence. The extreme value time position is obtained based on the local maximum value sequence and the local minimum value sequence. The standard deviation of the local maximum value sequence and the local minimum value sequence in the preset direction is calculated to obtain the standard deviation calculation result.

[0014] The signal extraction time range is obtained based on the extreme time location and the standard deviation calculation results. Then, the target blue-green laser reflection echo signal is extracted and the reflection spatial location is calculated based on the signal extraction time range to obtain near-target surface point cloud data.

[0015] Optionally, the method for processing high-density noise in the surface point cloud of the blue-green lidar, wherein the adaptive calculation of the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features specifically includes:

[0016] A voxelized index is established based on the near-target surface point cloud data. A variable search radius is obtained based on the voxelized index. The covariance matrix of the near-target surface point cloud data is searched and calculated based on the variable search radius to obtain the target covariance matrix.

[0017] Based on the target covariance matrix, feature calculation is performed on the local point cloud corresponding to the variable search radius to obtain multiple spatial distribution feature values. Based on all the spatial distribution feature values, spatial features are constructed to obtain multiple spatial structure features.

[0018] Obtain the maximum and minimum eigenvalues ​​of all the spatial distribution eigenvalues, and obtain the plane constructed by the target eigenvector corresponding to the maximum eigenvalue and the target eigenvector corresponding to the minimum eigenvalue. Project the local point cloud onto the plane according to a preset distance to obtain the projection result.

[0019] Based on the projection results, the three-dimensional coordinates of all projection points in the plane are transformed into two-dimensional coordinates to obtain the two-dimensional coordinates of all projection points, and multiple density distribution features are obtained based on all the two-dimensional coordinates.

[0020] Optionally, the method for processing high-density noise in surface point clouds of the blue-green lidar, wherein the step of searching and calculating the covariance matrix of the near-target surface point cloud data based on the variable search radius to obtain the target covariance matrix specifically includes:

[0021] The point cloud data of the near target surface is searched according to the variable search radius to obtain a local point cloud. The average coordinates of all laser points in the local point cloud are calculated to obtain multiple averages. The target average of the local point cloud is obtained based on all the averages.

[0022] The coordinates of all laser points are centered according to the target mean to obtain multiple centered coordinates. A matrix is ​​constructed based on all the centered coordinates to obtain a centered coordinate matrix. The covariance of the centered coordinate matrix is ​​then calculated to obtain the target covariance matrix.

[0023] Optionally, in the method for processing high-density noise in the surface point cloud of the blue-green lidar, the step of performing a two-dimensional projection transformation on the three-dimensional coordinates of all projection points in the plane based on the projection result specifically involves:

[0024] ;

[0025] ;

[0026] in, The second two-dimensional coordinate projected onto the plane Each horizontal axis The second two-dimensional coordinate projected onto the plane One vertical axis, For the first One projection point, The target feature vector corresponding to the largest eigenvalue. This is the target feature vector corresponding to the smallest feature value.

[0027] Optionally, the high-density noise processing method for surface point clouds of the blue-green lidar, wherein encoding all point cloud spatial distribution features in the local spatial distribution features to obtain a terrain surface point cloud, and classifying the terrain surface point cloud to obtain a target surface point cloud with near-surface noise suppression, specifically includes:

[0028] All point cloud spatial distribution features in the local spatial distribution features are classified to obtain feature classification results. The feature classification results are encoded to obtain multiple spatial distribution feature statistical codes. Based on all the spatial distribution feature statistical codes, a terrain surface point cloud with feature codes is obtained.

[0029] A point cloud classification method is constructed based on the statistical encoding of all the spatial distribution features. The point cloud on the terrain surface is classified according to the point cloud classification method to obtain the point cloud classification result. The noisy point cloud in the point cloud classification result is removed to obtain the target surface point cloud with near-surface noise suppression.

[0030] Optionally, the method for processing high-density noise in surface point clouds of the blue-green lidar, wherein the point cloud classification method based on statistical encoding of all the spatial distribution features specifically includes:

[0031] Obtain the classification model and label all the spatial distribution features statistically encoded to obtain the training sample set;

[0032] The classification model is trained using the training sample set to obtain the training results, and a point cloud classification method is constructed based on the training results.

[0033] Optionally, the surface point cloud high-density noise processing method for the blue-green lidar includes a surface point cloud high-density noise processing system comprising:

[0034] The signal filtering module is used to acquire the blue-green laser reflection echo signal of the target scene area, and to filter the blue-green laser reflection echo signal to obtain near-target surface point cloud data.

[0035] The feature calculation module is used to adaptively calculate the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and to obtain local spatial distribution features based on all the spatial structure features and all the density distribution features.

[0036] The point cloud classification module is used to encode all point cloud spatial distribution features in the local spatial distribution features to obtain terrain surface point clouds, and to classify the terrain surface point clouds to obtain target surface point clouds with near-surface noise suppression.

[0037] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a surface point cloud high-density noise processing program for blue-green lidar stored in the memory and executable on the processor, wherein when the surface point cloud high-density noise processing program for blue-green lidar is executed by the processor, the steps of the surface point cloud high-density noise processing method for blue-green lidar as described above are implemented.

[0038] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a surface point cloud high-density noise processing program for a blue-green lidar, and when the surface point cloud high-density noise processing program for the blue-green lidar is executed by a processor, it implements the steps of the surface point cloud high-density noise processing method for the blue-green lidar as described above.

[0039] In this invention, blue-green laser reflection echo signals of the target scene area are acquired, and the blue-green laser reflection echo signals are filtered to obtain near-target surface point cloud data. The spatial distribution features of local point clouds in the near-target surface point cloud data are adaptively calculated to obtain multiple spatial structure features and multiple density distribution features. Based on all the spatial structure features and density distribution features, local spatial distribution features are obtained. All point cloud spatial distribution features in the local spatial distribution features are encoded to obtain terrain surface point clouds. The terrain surface point clouds are then classified to obtain near-surface noise-suppressed target surface point clouds. This invention can automatically identify target point clouds and accurately remove non-target noise point clouds when processing sparse point clouds in large-scale integrated land and water scenes, improving the efficiency and accuracy of point cloud denoising. Attached Figure Description

[0040] Figure 1 This is a flowchart of a preferred embodiment of the surface point cloud high-density noise processing method for blue-green lidar of the present invention;

[0041] Figure 2 This is a schematic diagram of the integrated land and water point cloud corresponding to the target scene area in a preferred embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the direct-pass filtering result of blue-green lidar under strong interference conditions in a preferred embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram illustrating the selection of the effective portion of the echo waveform in a preferred embodiment of the present invention;

[0044] Figure 5 This is a schematic diagram of the covariance matrix of the point cloud within the variable search radius in this invention;

[0045] Figure 6 This is a schematic diagram of the point cloud near the target surface in a preferred embodiment of the present invention;

[0046] Figure 7 This is a schematic diagram of the clustering effect after dimensionality reduction of the near-target surface point cloud in a preferred embodiment of the present invention;

[0047] Figure 8 This is a schematic diagram of point cloud feature statistical encoding for the first statistical count in a preferred embodiment of the present invention;

[0048] Figure 9 This is a schematic diagram of point cloud feature statistical encoding for the second statistical count in a preferred embodiment of the present invention;

[0049] Figure 10 This is a schematic diagram of point cloud feature statistical encoding for the third statistical counting in a preferred embodiment of the present invention;

[0050] Figure 11 This is a schematic diagram of point cloud feature statistical encoding for the fourth statistical counting in a preferred embodiment of the present invention;

[0051] Figure 12 This is a schematic diagram of the labeled training sample set in a preferred embodiment of the present invention;

[0052] Figure 13 This is a schematic diagram comparing the classification confusion matrices of the SVM model and the RF model in a preferred embodiment of the present invention;

[0053] Figure 14 This is a schematic diagram comparing the classification ROC curves of the SVM model and the RF model in a preferred embodiment of the present invention;

[0054] Figure 15 This is a schematic diagram of point cloud classification results in a preferred embodiment of the present invention;

[0055] Figure 16 This is a schematic diagram of the point cloud on the target surface in a preferred embodiment of the present invention;

[0056] Figure 17 This is a structural diagram of a preferred embodiment of the surface point cloud high-density noise processing system for blue-green lidar of the present invention;

[0057] Figure 18 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0059] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0060] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0061] The surface point cloud high-density noise processing method for blue-green lidar according to a preferred embodiment of the present invention, such as... Figure 1 As shown, the method for processing high-density noise in the surface point cloud of the blue-green lidar includes the following steps:

[0062] Step S10: Obtain the blue-green laser reflection echo signal of the target scene area, and filter the blue-green laser reflection echo signal to obtain near-target surface point cloud data.

[0063] Specifically, unlike traditional land-based lidar, the blue-green lasers used in ALB (Airborne Laser Bathymetry) systems operate in the visible light band. Their shorter wavelengths make them susceptible to scattering and absorption interference from suspended matter, bubbles, and interfaces between different media in water. Furthermore, due to variations in the flight platform's attitude and scanning methods, the spatial distribution density of the point cloud is uneven, with target point clouds being sparse and heavily interfered with, while the density of non-target noise points increases significantly. Existing pass-through filtering methods, due to their simple rules, suffer from high false negative rates when processing high-noise areas. Geometric and statistical feature methods are ill-suited to adapting to the inconsistent variations in blue-green laser point clouds across different scales and scenarios. Clustering classification methods based on point cloud density are easily affected by the sparsity and density variations of the point cloud, resulting in low target extraction accuracy. Therefore, to ensure the accuracy of target surface point cloud extraction, the conventional processing of point clouds detected by airborne blue-green lidar still requires a significant amount of manual interaction, resulting in low automation. To address the aforementioned issues, this invention proposes a method for processing high-density noise in surface point clouds of blue-green lidar. It employs a direct-pass filtering method based on signal propagation distance to remove outlier noise points with elevation anomalies, such as those at high altitudes and below the ground surface, thus initially extracting effective point clouds near the surface. A variable search radius structural feature extraction strategy based on dimensional space is introduced, dynamically adjusting the neighborhood search radius according to changes in point cloud density, and identifying planar structural regions using the minimum information entropy criterion. A variable radius principal component analysis method is used to estimate the point cloud normal vector and principal direction information, and clustering analysis is performed on the local projection plane formed by the normal vector and principal direction to enhance the difference between target points and noise points in local spatial structure.

[0064] In this embodiment of the invention, the processing effect of ALB in a mixed land-water area (i.e., the target scene area) is illustrated. The lakebed of the target scene area is mainly composed of silt deposits, which have low reflectivity to blue-green lasers, affecting the intensity of laser echoes. Furthermore, the water is relatively turbid. Based on a comprehensive analysis of the lakebed, water quality, and water depth, using ALB for integrated land-water detection can meet the requirements of land area measurement, ensuring the stability of echo signal identification and classification accuracy in complex water body junction areas. It further improves the distinction between nearshore vegetation and shallow nearshore water areas, fully considering the scene structure characteristics exhibited by the surface spatial positions obtained from blue-green laser radar detection. The corresponding integrated land-water point cloud is as follows: Figure 2 As shown ( Figure 2The coordinates in the image represent elevation values. The ALB receiver continuously records the intensity changes of the blue-green laser reflection echo signal by continuously recording the detection pulse echo signal (i.e., the blue-green laser reflection echo signal). Since the ALB uses the visible light band for detection, it is easily affected by strong scattering from the medium in the transmission optical path. Considering the randomness in the spatial distribution of the point cloud generated by scattering interference in the medium and the point cloud calculated from the target surface echo detection, elevation statistical analysis can be performed based on the elevation distribution of the point cloud. Elevation portions of the point cloud exceeding the statistical confidence interval are removed using a direct-pass filtering method. Figure 3 As shown, the high-density white point cloud is a non-target surface point cloud generated by atmospheric scattering and system noise.

[0065] To address the changes in signal propagation distance and intensity, point cloud pass-through filtering is performed on the obtained blue-green laser reflection echo signals. Specifically, preprocessing is performed using the propagation distance characteristics of the blue-green laser reflection echo signals. For non-surface signals (e.g., high-altitude clutter) and subsurface anomalies (e.g., false echoes, negative elevation points caused by equipment errors) in the point cloud, firstly, pass-through filtering is used to extract the signal portion that may contain target surface echoes based on the intensity variation characteristics of the blue-green laser reflection echo signals. Then, referencing flight scan parameters, a PMT (Photomultiplier Tube) or APD (Avalanche Photomultiplier) is used. Based on distance gating and echo signal filtering using avalanche photodiodes, this method estimates the target surface elevation variation range based on prior information. Point clouds exceeding the target surface elevation variation range in the elevation direction are removed, thereby obtaining near-target surface point clouds located at or near the actual ground surface. This mainly filters isolated points at high altitudes and erroneous points below the ground, reducing the risk of error propagation and improving the overall authenticity and representativeness of the point cloud. It can effectively improve identification accuracy, especially for complex underwater areas. The point cloud direct-pass filtering process includes direct-pass filtering based on the target surface detection distance and direct-pass filtering based on the target surface intensity variation.

[0066] For direct-pass filtering based on target surface detection distance, considering that the backscattered echo of the laser beam returning in the near field region will enter the photosensitive surface of the photomultiplier tube before the blue-green laser reflection echo signal, and because the returned energy is close to the detector position, it is easy to cause signal saturation of the photomultiplier tube and a decrease in photosensitive sensitivity, thus affecting the detection of weak echo signals of far-field targets. Therefore, this invention employs a range-gating strategy in the receiving end, utilizing a high-precision clock in the synchronization control board to combine high-precision time synchronization technology with photomultiplier tube delay technology. This enables range-gating of the laser propagation distance, obtaining the target blue-green laser reflection echo signal containing the target surface echo, thereby maintaining the photomultiplier tube's sensitivity for detecting distant targets. Specifically, based on a high-precision timing circuit of a lidar with a timing accuracy ≤1ns, after sensing the laser pulse emission, a TTL level signal (binary digital level signal) is sent to the photomultiplier tube after a delay. The delay can be flexibly set according to the target distance. During the delay, a reverse voltage of approximately 10V is applied between the photocathode and dynode of the PMT. Electron transfer between the photocathode and dynode allows the PMT to operate in an extremely low gain state, avoiding performance damage from strong echo signals. After the delay ends, a TTL level signal is sent to the PMT, canceling the reverse voltage and allowing the PMT to operate in a normal gain state, thus achieving range-gating-based variable gain detection by the PMT.

[0067] For direct-pass filtering based on target surface intensity changes, in blue-green lidar detection, to ensure that the target's reflected waveform is included in the received blue-green laser reflected echo signal, and to ensure a long effective distance in the lidar's transmission path for Earth observation, the receiver typically uses a relatively long sampling time as the fixed receiving length of the blue-green laser reflected echo signal. The effective portion of the blue-green lidar signal mainly includes echoes from land surfaces and reflected signals returning to the water surface, water body, and bottom during water penetration. The remaining portion of the blue-green lidar signal usually reflects the background noise affecting the signal during propagation. Removing the non-target surface echo signal portion can effectively reduce noise interference in the blue-green laser reflected echo signal. Conventional signal selection usually uses a response threshold method, i.e., using an echo intensity exceeding a certain threshold as the basis for the effective signal range. This selection method has the following problems:

[0068] (1) Signal integrity is difficult to guarantee. Traditional threshold limiting methods rely on the selection of thresholds, which can easily cause interruption of the effective part of the signal in areas with low intensity.

[0069] (2) For individual blue-green laser reflection echo signals with a large distance between two echoes (i.e. poor energy continuity), the uniform and fixed threshold lacks consideration for signal fluctuations caused by noise, which is not conducive to the simulation of the curve.

[0070] Therefore, this invention divides the blue-green laser reflection echo signal obtained by the receiver into three parts according to the main components of the pulse signal, namely the first part of the blue-green laser reflection echo signal. Part Two: Blue-Green Laser Reflection Echo Signal and the third part of the blue-green laser reflection echo signal The corresponding segmentation formula is:

[0071] ;

[0072] in, This represents the time position of each signal intensity sampling point in the blue-green laser reflected echo signal. This represents the starting time and position of the blue-green laser reflected echo signal in the point cloud on the target surface. This represents the termination time position of the blue-green laser reflection echo signal on the target surface point cloud. Typically, signal analysis treats the noise encountered by laser energy propagating in space as Gaussian white noise during filtering. From the blue-green laser reflection echo signal, it can be seen that the influence of the laser pulse energy gradually decreases at the tail end of the signal, while the noise intensity remains similar. Therefore, the beginning and end portions of the blue-green laser reflection echo signal are important references for propagation noise research. The statistical characteristics of the effective echo portion and the noise echo portion are significantly different. In this invention, a threshold of three times the standard deviation of the effective echo portion is used, and sampling points are continuously added from both sides to the effective portion of the waveform to update the data sequence. When the standard deviations of both are greater than their respective thresholds, it is considered that at this moment, not only is there background noise, but it is also affected by the reflected pulse. At this point, the data is truncated by selecting the nearest local minimum position (i.e., the local minimum value) of the sampling point.

[0073] To ensure the effective portion is truncated at local minima, values ​​are slidably taken from the beginning and end of the target blue-green laser reflection echo signal towards the extrema of the effective portion, forming an extremum sequence. During the continuous updating of the extremum sequence, when a new extremum is added, a confidence interval of three times the standard error is used. That is, when the standard deviation of the new sequence exceeds three times the standard error of the previous sequence, the effective portion of the target blue-green laser reflection echo signal is considered reached, and the sequence updating stops, resulting in a local extremum sequence. The signal after the previously added minimum value is considered the effective portion of the target blue-green laser reflection echo signal. By calculating the first derivative, a sequence calculation is performed from the preset time position of the target blue-green laser reflection echo signal to the extremum position of the local extremum sequence, yielding a local maximum sequence and a local minimum sequence. The expression for the local maximum sequence is:

[0074] ;

[0075] The expression for the local minimum sequence is:

[0076] ;

[0077] in, For the first The time corresponding to each maximum value The total number of maxima. For the first The time corresponding to each minimum value The total number of minimum values; and Simplified to and Then the extreme value time interval between the local maximum sequence and the local minimum sequence is:

[0078] ;

[0079] in, The extreme value time interval between the local maximum sequence and the local minimum sequence is... The extreme value time interval between the local minimum sequence and the local maximum sequence is defined; the standard deviation of the local maximum sequence and the local minimum sequence in a preset direction (e.g., from left to right and from right to left) is calculated to obtain the standard deviation calculation result; the standard deviation from left to right is:

[0080] ;

[0081] The standard deviation from right to left is:

[0082] ;

[0083] in, Let be the standard deviation of the sequence of local maxima from left to right. Let be the standard deviation of the local minimum sequence from left to right. Let be the standard deviation of the local maximum sequence from right to top left. Let be the standard deviation of the local minimum sequence from right to top left. This is the blue-green laser reflection echo signal. This represents the average value of the blue-green laser reflection echo signal in the first part. This represents the average value of the blue-green laser reflection echo signal in the second part; subsequently, we can obtain... and The corresponding expression is:

[0084] ;

[0085] ;

[0086] final, The portion corresponding to the time range (i.e., the signal extraction time range) is the signal portion to be extracted. Data extraction is performed on the target's blue-green laser reflection echo signal according to the signal extraction time range to obtain near-target surface point cloud data. Therefore, through statistical analysis and direct-pass filtering of the intensity changes of each echo signal, interference caused by non-target signal components in excessively high or low spatial ranges in the laser signal propagation path can be effectively eliminated, providing basic data for further lidar target point cloud extraction.

[0087] In this embodiment of the invention, the blue-green laser reflection echo signal is used as the processing object. An example is taken with the target surface detection distance range set to a maximum of 10.76m and a minimum of 9.52m. The effective time range is calculated using the following formula:

[0088] ;

[0089] ;

[0090] in, The minimum effective waveform duration. The maximum duration of the effective waveform. The speed at which a laser travels through the air. Let be the refractive index of the laser in air. The refractive index of the laser in seawater, This represents the range of variation of the inclination moment in water. To be the minimum value, The maximum value, The half-width at half maximum (WHM) of the laser emission echo signal. The half-width at half-maximum (WHM) of the laser reflection echo signal, and the relevant parameter values ​​based on the echo signal intensity and signal propagation distance range in this embodiment of the invention, are shown in Table 1.

[0091] Table 1: Relevant parameter values ​​based on echo signal strength and signal propagation distance range

[0092]

[0093] Since the sampling frequency of the original waveform data is 1 ns, the estimated effective waveform duration range is: This reflects the length of the laser response segment in the land surface and water body of the target scene area, which should be within... Within the range. To illustrate the processing effect of this method, firstly, a single echo waveform was randomly selected for an effective portion selection experiment, and the results are as follows. Figure 4 As shown, Figure 4 Part a in the diagram represents the effective portion of the original waveform signal. Figure 4 Part b in the middle represents the effective portion of the echo waveform extracted using the thresholding method. Figure 4 Part c represents the effective portion of the echo waveform obtained using this invention, and it is compared using a conventional thresholding method. The original waveform is processed using an adaptive effective portion selection method based on local features, resulting in the truncation point of the effective portion of the waveform. At this point, the three-fold mean error of the front-end waveform sequence is... The mean square error of the tail sequence is The difference between the two is significant. Using the conventional three times the mean square error at the tail end to represent the overall characteristics of the noise contained in the waveform would be unreasonable. However, due to… Figure 4 As shown in part b, using a single threshold method is more likely to cause misjudgment or truncation of the valid waveform. Figure 4 As can be seen from part c, this invention utilizes the local statistical characteristics of the waveform itself. The effective part of the extracted echo waveform has good continuity, effectively removes the redundant parts in the original waveform, and completely contains the intensity response information of the laser during its propagation on the land surface and in water.

[0094] Step S20: Adaptively calculate the spatial distribution features of the local point cloud in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and obtain the local spatial distribution features based on all the spatial structure features and all the density distribution features.

[0095] Specifically, after obtaining the near-target surface point cloud data, such as Figure 5As shown, a voxelized index (e.g., an octree) is established using the near-target surface point cloud data, and iteratively incremented according to a specific step size. Expand the variable search radius centered on the target point to perform the search (e.g., by one). Expanding the search radius centered on the target point and performing the search yields the following covariance matrix: , according to two Expanding the search radius centered on the target point and performing the search yields the following covariance matrix: and according to indivual Expanding the search radius centered on the target point and performing the search yields the following covariance matrix: ,in, for To prevent the variable search radius from expanding indefinitely as the number of points gradually increases, this invention sets a maximum threshold for the variable search radius and sets the range of neighborhood radius variation with reference to the point cloud density. ,in, The minimum value of the neighborhood radius. This is the maximum value of the neighborhood radius, and the value within the radius is calculated after each expansion of the search radius. The covariance matrix of each point cloud coordinate is calculated, and a feature space containing structural features is constructed based on the calculated eigenvalues ​​and eigenvectors of the matrix. Based on the covariance matrix obtained in each search, the spatial distribution eigenvalues ​​of the three-dimensional covariance matrix of the point cloud spatial reduction table within its corresponding neighborhood can be calculated. and its corresponding eigenvectors (wherein, the normalized target feature vectors are respectively) The point cloud data of the near-target surface is searched according to the variable search radius to obtain a local point cloud. The average coordinates of all laser points in the local point cloud are calculated to obtain multiple averages. The target average of the local point cloud is obtained based on all the averages. The local point cloud contains... There are 1 laser points, and the coordinates of each laser point are: The target mean of the local point cloud for:

[0096] ;

[0097] The coordinates of all laser points are centered based on the target mean, i.e., the mean is subtracted from the coordinates of each laser point to obtain multiple centered coordinates. Let the centered coordinates be... Then the expression corresponding to the centered coordinates is:

[0098] ;

[0099] Then, a matrix is ​​constructed based on all the centralized coordinates to obtain the centralized coordinate matrix. The corresponding expression is:

[0100] ;

[0101] The covariance of the centered coordinate matrix is ​​calculated to obtain the target covariance matrix. In the target covariance matrix, each row represents the centered coordinates of a laser point. The expression is:

[0102] ;

[0103] Then, based on the target covariance matrix, feature calculations are performed on the local point cloud corresponding to the variable search radius to obtain multiple spatial distribution feature values; let the spatial distribution feature values ​​of the local point cloud be... and eigenvectors ,satisfy:

[0104] ;

[0105] The formula for calculating the spatial distribution eigenvalues ​​and eigenvectors of a local point cloud is as follows:

[0106] ;

[0107] in, Using a unit matrix, this invention can more accurately adapt to the problem of uneven point cloud density compared to methods based on fixed radius or k-nearest neighbors, and is especially suitable for situations where the point density of airborne blue-green laser data fluctuates greatly in areas such as near the coast and complex terrain.

[0108] Considering the significant value of the spatial distribution characteristics exhibited by point clouds in target surface point cloud extraction, this section improves the accuracy of point cloud structural feature recognition and enhances the point cloud distribution attributes and discriminative ability of the target's spatial structure by calculating and evaluating local morphological differences on the target surface. During the variable search radius change process, the spatial distribution characteristic values ​​of the three-dimensional covariance matrix of the corresponding neighborhood point cloud spatial reduction table are calculated (respectively...). , and ,in, ) and their corresponding feature vectors (respectively) 、 and ),pass , and The relationships between them construct multi-dimensional spatial features of point clouds ,in, The first point cloud multidimensional spatial features, For the second point cloud multidimensional spatial features, The expression for the third point cloud multidimensional spatial feature is:

[0109] ;

[0110] ;

[0111] ;

[0112] As the variable search radius increases, in order to make the point cloud selected in the local space better reflect the local features of the target surface, this part adopts the method of minimizing information entropy to construct the entropy function. The corresponding expression is:

[0113] ;

[0114] That is, the iteration terminates when the variable search radius is expanded until the entropy function of the point cloud structure feature value in the target point and its radius neighborhood is minimized. If the entropy function value has not converged when the maximum search radius is reached, the spatial distribution feature value corresponding to the maximum search radius is recorded. By selecting the optimal neighborhood radius with the minimum entropy as the target, the foundation for subsequent classification is laid.

[0115] Target surface point clouds exhibit more pronounced spatial regularity compared to interference point clouds caused by medium scattering and system random noise. However, the non-uniform distribution of target surface point clouds, influenced by high-density noise in air and water, results in significant differences, making it difficult for conventional density-based 3D point cloud classification methods to effectively separate target surface point clouds from noise point clouds (such as...). Figure 6 (As shown). Therefore, in order to fully utilize the spatial distribution and local density differences between the point cloud and random noise on the target surface, this invention employs a method combining adaptive perspective point cloud dimensionality reduction and density clustering to estimate the spatial distribution and density characteristics of the point cloud in a local area. The clustering effect after point cloud reduction is as follows: Figure 7 As shown. First, the laser point cloud within the search range is considered when the above iteration terminates after the variable search radius. Projected to the largest eigenvalue (i.e., the spatial distribution eigenvalue) ) and the minimum eigenvalue (i.e., spatial distribution eigenvalue) The target feature vector corresponding to ) and The constructed plane is used to achieve dimensionality reduction calculations that conform to the distribution trend of the point cloud. Let the distance from each point within the range to the projection plane be... for:

[0116] ;

[0117] Point cloud projection onto surface The formula for calculating the three-dimensional coordinates is:

[0118] ;

[0119] The two-dimensional coordinates of the projection point on the plane can then be obtained, and the corresponding calculation formula is:

[0120] ;

[0121] ;

[0122] in, The second two-dimensional coordinate projected onto the plane Each horizontal axis The second two-dimensional coordinate projected onto the plane By mapping the three-dimensional spatial distribution of the point cloud onto a two-dimensional local plane that conforms to the distribution direction of the point cloud, dimensionality reduction of the local point cloud space is achieved. Since the target surface point cloud has a stable distribution around the normal, while the noise point cloud tends to be more random and disordered, the former exhibits a compact and continuous clustering structure on the dimensionality-reduced plane. A density clustering method is used for binary classification estimation of the target surface point cloud and random noise interference, assuming a viewpoint-adaptive point cloud density feature... for:

[0123] ;

[0124] This invention significantly improves the spatial difference between noise points and target points, solving the problem of difficulty in noise identification caused by uneven point cloud density. By leveraging the differences in distribution patterns in the dimensionality-reduced planar space, it enhances the ability to identify randomly distributed points (e.g., water surface reflections, environmental interference echoes, etc.), providing a more discriminative feature foundation for subsequent classification models.

[0125] Step S30: Encode all point cloud spatial distribution features in the local spatial distribution features to obtain terrain surface point clouds, and classify the terrain surface point clouds to obtain target surface point clouds with near-surface noise suppression.

[0126] Specifically, after obtaining the local spatial distribution features, considering that the spatial distribution features of the point cloud on the target surface should have a certain stability and consistency under different radii or different viewpoints, firstly, feature classification is performed on all the point cloud spatial distribution features in the local spatial distribution features to obtain the feature classification results. It is assumed that each point is counted as a laser point within the search range during the traversal process. Next, then the first The spatial statistical feature classification results are as follows:

[0127] ;

[0128] ;

[0129] ;

[0130] ;

[0131] in, For the first statistical count, For the second statistical count, This is the third statistical count. This is the fourth statistical count. ,set up , , , ,in, Statistical encoding of point cloud features for the first statistical count (e.g.) Figure 8 (as shown) Statistical encoding of point cloud features for the second statistical count (e.g.) Figure 9 (as shown) Statistical encoding of point cloud features for third statistical counting (e.g.) Figure 10 (as shown) Statistical encoding of point cloud features for the fourth statistical count (e.g.) Figure 11 As shown), where, Figure 8 , Figure 9 , Figure 10 and Figure 11 If the coordinates are all elevation values, then the spatial distribution characteristics of each laser point are statistically coded. for:

[0132] ;

[0133] Based on the statistical encoding of all the spatial distribution features, a terrain surface point cloud with feature codes is obtained. Then, a point cloud classification method based on multidimensional feature codes is constructed. Specifically, a classification model is obtained, and all the spatial distribution feature statistical codes are labeled to obtain a structured dataset. The structured dataset is then divided into a training sample set and a test set using a standard 7:3 ratio. The corresponding training sample set is shown below. Figure 12 As shown ( Figure 12 Part (a) is the training sample set from the first-person perspective. Figure 12Part (b) is the training sample set from the second perspective. The classification model is a supervised classification machine learning model, specifically an SVM (Support Vector Machine) model and a RF (Random Forest) model. The SVM model learns the discrimination boundary of nonlinear point cloud samples by partitioning the hyperplane in the feature space, while the RF model improves robustness to complex distribution samples such as outliers and sparse points through an ensemble learning strategy. The classification model is trained using the training sample set to obtain training results, and a point cloud classification method is constructed based on these results. In this embodiment, the prediction results output by the SVM and RF models are plotted as a confusion matrix, as shown below. Figure 13 As shown ( Figure 13 Part (a) is the confusion matrix of the SVM model. Figure 13 (b) is the confusion matrix of the RF model. Figure 13 The numbers in the table represent the number of test samples. The predicted probabilities of the SVM model and the RF model are plotted as ROC (receiver operating characteristic) curves for comparison. Figure 14 As shown, after model training, the SVM and RF models were evaluated on the test set. The results show that both models have high classification accuracy. The SVM model achieved an overall classification accuracy of 97.88% on 8811 test samples, with an AUC (Area Under the Curve) of 0.9945 and a target class F1 score as high as 0.9867. The RF model achieved an accuracy of 97.79% on the same dataset, with an AUC of 0.9954 and a target class F1 score of 0.9861, indicating that the model has a strong ability to identify main class target points in point clouds and also has good robustness to minority class samples. In addition, to evaluate the model's generalization ability on different data subsets, a three-fold cross-validation method was used in the experiment for rapid performance verification on the training set. The average accuracy of SVM was 0.9759, and the average accuracy of RF was 0.9767, with both standard deviations below 0.002, further demonstrating the stability of the invention in practical applications.

[0134] The point cloud on the terrain surface is classified according to the aforementioned point cloud classification method to obtain the point cloud classification result, such as... Figure 15 As shown ( Figure 15 (The mid-coordinate is the elevation value), and the near-surface noise point cloud of the point cloud classification result is removed to obtain the noise-suppressed target surface point cloud, such as... Figure 16 As shown ( Figure 16(The mid-coordinate represents the elevation value). The above model experiments demonstrate that users can select the classifier type according to different application scenarios. By introducing this module, the entire blue-green lidar point cloud noise suppression process has transformed from traditional rule-based filtering methods to multi-dimensional spatial feature encoding and machine learning intelligent discrimination. This not only significantly improves the extraction accuracy of near-surface target point clouds but also significantly reduces the cost of manual intervention. It has promising prospects for promotion and application in automated point cloud preprocessing, target extraction, and feature analysis.

[0135] Furthermore, such as Figure 17 As shown, based on the above-mentioned method for high-density noise processing of surface point clouds in blue-green lidar, the present invention also provides a corresponding system for high-density noise processing of surface point clouds in blue-green lidar, wherein the system includes:

[0136] The signal filtering module 51 is used to acquire the blue-green laser reflection echo signal of the target scene area, and to filter the blue-green laser reflection echo signal to obtain near-target surface point cloud data.

[0137] The feature calculation module 52 is used to adaptively calculate the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and to obtain local spatial distribution features based on all the spatial structure features and all the density distribution features.

[0138] The point cloud classification module 53 is used to encode all the point cloud spatial distribution features in the local spatial distribution features to obtain the terrain surface point cloud, and to classify the terrain surface point cloud to obtain the target surface point cloud with near-surface noise suppression.

[0139] Furthermore, such as Figure 18 As shown, based on the above-mentioned high-density noise processing method for surface point clouds of blue-green lidar, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 18 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0140] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a surface point cloud high-density noise processing program 40 for a blue-green lidar, which can be executed by the processor 10 to implement the surface point cloud high-density noise processing method for blue-green lidar in this application.

[0141] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the surface point cloud high-density noise processing method of the blue-green lidar.

[0142] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface.

[0143] In one embodiment, when the processor 10 executes the surface point cloud high-density noise processing program 40 of the blue-green lidar in the memory 20, the following steps are performed:

[0144] The blue-green laser reflection echo signal of the target scene area is acquired, and the blue-green laser reflection echo signal is filtered to obtain near-target surface point cloud data.

[0145] Adaptive calculation is performed on the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and local spatial distribution features are obtained based on all the spatial structure features and all the density distribution features.

[0146] All point cloud spatial distribution features in the local spatial distribution features are encoded to obtain terrain surface point clouds, and the terrain surface point clouds are classified to obtain target surface point clouds with near-surface noise suppression.

[0147] Specifically, acquiring the blue-green laser reflection echo signal of the target scene area and filtering the blue-green laser reflection echo signal to obtain near-target surface point cloud data includes:

[0148] The blue-green laser reflection echo signal of the target scene area is obtained according to the distance gating strategy, and the blue-green laser reflection echo signal is subjected to near-target surface region distance gating filtering in the laser propagation direction to obtain the target blue-green laser reflection echo signal containing the target surface echo.

[0149] Extreme value sliding is performed on the first and last ends of the target blue-green laser reflected echo signal to obtain an extreme value sequence. The extreme value sequence is then updated to obtain a local extreme value sequence.

[0150] The preset time position of the target blue-green laser reflected echo signal and the extreme position of the local extreme value sequence are used to perform sequence calculation to obtain a local maximum value sequence and a local minimum value sequence. The extreme value time position is obtained based on the local maximum value sequence and the local minimum value sequence. The standard deviation of the local maximum value sequence and the local minimum value sequence in the preset direction is calculated to obtain the standard deviation calculation result.

[0151] The signal extraction time range is obtained based on the extreme time location and the standard deviation calculation results. Then, the target blue-green laser reflection echo signal is extracted and the reflection spatial location is calculated based on the signal extraction time range to obtain near-target surface point cloud data.

[0152] Specifically, the adaptive calculation of the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features includes:

[0153] A voxelized index is established based on the near-target surface point cloud data. A variable search radius is obtained based on the voxelized index. The covariance matrix of the near-target surface point cloud data is searched and calculated based on the variable search radius to obtain the target covariance matrix.

[0154] Based on the target covariance matrix, feature calculation is performed on the local point cloud corresponding to the variable search radius to obtain multiple spatial distribution feature values. Based on all the spatial distribution feature values, spatial features are constructed to obtain multiple spatial structure features.

[0155] Obtain the maximum and minimum eigenvalues ​​of all the spatial distribution eigenvalues, and obtain the plane constructed by the target eigenvector corresponding to the maximum eigenvalue and the target eigenvector corresponding to the minimum eigenvalue. Project the local point cloud onto the plane according to a preset distance to obtain the projection result.

[0156] Based on the projection results, the three-dimensional coordinates of all projection points in the plane are transformed into two-dimensional coordinates to obtain the two-dimensional coordinates of all projection points, and multiple density distribution features are obtained based on all the two-dimensional coordinates.

[0157] Specifically, the step of searching and calculating the covariance matrix of the near-target surface point cloud data based on the variable search radius to obtain the target covariance matrix includes:

[0158] The point cloud data of the near target surface is searched according to the variable search radius to obtain a local point cloud. The average coordinates of all laser points in the local point cloud are calculated to obtain multiple averages. The target average of the local point cloud is obtained based on all the averages.

[0159] The coordinates of all laser points are centered according to the target mean to obtain multiple centered coordinates. A matrix is ​​constructed based on all the centered coordinates to obtain a centered coordinate matrix. The covariance of the centered coordinate matrix is ​​then calculated to obtain the target covariance matrix.

[0160] Specifically, the step of performing a two-dimensional projection transformation on the three-dimensional coordinates of all projection points in the plane based on the projection result includes:

[0161] ;

[0162] ;

[0163] in, The second two-dimensional coordinate projected onto the plane Each horizontal axis The second two-dimensional coordinate projected onto the plane One vertical axis, For the first One projection point, The target feature vector corresponding to the largest eigenvalue. This is the target feature vector corresponding to the smallest feature value.

[0164] Specifically, the process of encoding all point cloud spatial distribution features in the local spatial distribution features to obtain a terrain surface point cloud, and classifying the terrain surface point cloud to obtain a target surface point cloud with near-surface noise suppression, includes:

[0165] All point cloud spatial distribution features in the local spatial distribution features are classified to obtain feature classification results. The feature classification results are encoded to obtain multiple spatial distribution feature statistical codes. Based on all the spatial distribution feature statistical codes, a terrain surface point cloud with feature codes is obtained.

[0166] A point cloud classification method is constructed based on the statistical encoding of all the spatial distribution features. The point cloud on the terrain surface is classified according to the point cloud classification method to obtain the point cloud classification result. The noisy point cloud in the point cloud classification result is removed to obtain the target surface point cloud with near-surface noise suppression.

[0167] The point cloud classification method based on statistical encoding of all the spatial distribution features specifically includes:

[0168] Obtain the classification model and label all the spatial distribution features statistically encoded to obtain the training sample set;

[0169] The classification model is trained using the training sample set to obtain the training results, and a point cloud classification method is constructed based on the training results.

[0170] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a surface point cloud high-density noise processing program for a blue-green lidar, and the surface point cloud high-density noise processing program for the blue-green lidar, when executed by a processor, implements the steps of the surface point cloud high-density noise processing method for the blue-green lidar as described above.

[0171] In summary, this invention provides a method, system, terminal, and storage medium for high-density noise processing of surface point clouds in blue-green lidar. The method includes: acquiring blue-green laser reflection echo signals from a target scene area; filtering the blue-green laser reflection echo signals to obtain near-target surface point cloud data; adaptively calculating the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and obtaining local spatial distribution features based on all the spatial structure features and all the density distribution features; encoding all point cloud spatial distribution features in the local spatial distribution features to obtain terrain surface point clouds; and classifying the terrain surface point clouds to obtain target surface point clouds with near-surface noise suppression. This invention can automatically identify target point clouds and accurately remove non-target noise point clouds when processing sparse point clouds in large-scale integrated land and water scenes, improving the efficiency and accuracy of point cloud denoising.

[0172] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0173] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0174] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for processing high-density noise in surface point clouds of a blue-green lidar system, characterized in that, The method for processing high-density noise in the surface point cloud of the blue-green lidar includes: The blue-green laser reflection echo signal of the target scene area is acquired, and the blue-green laser reflection echo signal is filtered to obtain near-target surface point cloud data. Adaptive calculation is performed on the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and local spatial distribution features are obtained based on all the spatial structure features and all the density distribution features. All point cloud spatial distribution features in the local spatial distribution features are encoded to obtain terrain surface point clouds, and the terrain surface point clouds are classified to obtain target surface point clouds with near-surface noise suppression.

2. The method for processing high-density noise in surface point clouds of blue-green lidar according to claim 1, characterized in that, The process of acquiring the blue-green laser reflection echo signal of the target scene area and filtering the blue-green laser reflection echo signal to obtain near-target surface point cloud data specifically includes: The blue-green laser reflection echo signal of the target scene area is obtained according to the distance gating strategy, and the blue-green laser reflection echo signal is subjected to near-target surface region distance gating filtering in the laser propagation direction to obtain the target blue-green laser reflection echo signal containing the target surface echo. Extreme value sliding is performed on the first and last ends of the target blue-green laser reflected echo signal to obtain an extreme value sequence. The extreme value sequence is then updated to obtain a local extreme value sequence. The preset time position of the target blue-green laser reflected echo signal and the extreme position of the local extreme value sequence are used to perform sequence calculation to obtain a local maximum value sequence and a local minimum value sequence. The extreme value time position is obtained based on the local maximum value sequence and the local minimum value sequence. The standard deviation of the local maximum value sequence and the local minimum value sequence in the preset direction is calculated to obtain the standard deviation calculation result. The signal extraction time range is obtained based on the extreme time location and the standard deviation calculation results. Then, the target blue-green laser reflection echo signal is extracted and the reflection spatial location is calculated based on the signal extraction time range to obtain near-target surface point cloud data.

3. The method for processing high-density noise in surface point clouds of blue-green lidar according to claim 1, characterized in that, The adaptive calculation of the spatial distribution features of local point clouds in the near-target surface point cloud data yields multiple spatial structure features and multiple density distribution features, specifically including: A voxelized index is established based on the near-target surface point cloud data. A variable search radius is obtained based on the voxelized index. The covariance matrix of the near-target surface point cloud data is searched and calculated based on the variable search radius to obtain the target covariance matrix. Based on the target covariance matrix, feature calculation is performed on the local point cloud corresponding to the variable search radius to obtain multiple spatial distribution feature values. Based on all the spatial distribution feature values, spatial features are constructed to obtain multiple spatial structure features. Obtain the maximum and minimum eigenvalues ​​of all the spatial distribution eigenvalues, and obtain the plane constructed by the target eigenvector corresponding to the maximum eigenvalue and the target eigenvector corresponding to the minimum eigenvalue. Project the local point cloud onto the plane according to a preset distance to obtain the projection result. Based on the projection results, the three-dimensional coordinates of all projection points in the plane are transformed into two-dimensional coordinates to obtain the two-dimensional coordinates of all projection points, and multiple density distribution features are obtained based on all the two-dimensional coordinates.

4. The method for processing high-density noise in surface point clouds of blue-green lidar according to claim 3, characterized in that, The step of searching and calculating the covariance matrix of the near-target surface point cloud data based on the variable search radius to obtain the target covariance matrix specifically includes: The point cloud data of the near target surface is searched according to the variable search radius to obtain a local point cloud. The average coordinates of all laser points in the local point cloud are calculated to obtain multiple averages. The target average of the local point cloud is obtained based on all the averages. The coordinates of all laser points are centered according to the target mean to obtain multiple centered coordinates. A matrix is ​​constructed based on all the centered coordinates to obtain a centered coordinate matrix. The covariance of the centered coordinate matrix is ​​then calculated to obtain the target covariance matrix.

5. The method for processing high-density noise in surface point clouds of blue-green lidar according to claim 3, characterized in that, The step of performing a two-dimensional projection transformation on the three-dimensional coordinates of all projection points in the plane based on the projection result is specifically as follows: ； ； in, The second two-dimensional coordinate projected onto the plane Each horizontal axis The second two-dimensional coordinate projected onto the plane One vertical axis, For the first One projection point, The target feature vector corresponding to the largest eigenvalue. This is the target feature vector corresponding to the smallest feature value.

6. The method for processing high-density noise in surface point clouds of blue-green lidar according to claim 1, characterized in that, The process of encoding all point cloud spatial distribution features in the local spatial distribution features to obtain a terrain surface point cloud, and classifying the terrain surface point cloud to obtain a target surface point cloud with near-surface noise suppression, specifically includes: All point cloud spatial distribution features in the local spatial distribution features are classified to obtain feature classification results. The feature classification results are encoded to obtain multiple spatial distribution feature statistical codes. Based on all the spatial distribution feature statistical codes, a terrain surface point cloud with feature codes is obtained. A point cloud classification method is constructed based on the statistical encoding of all the spatial distribution features. The point cloud on the terrain surface is classified according to the point cloud classification method to obtain the point cloud classification result. The noisy point cloud in the point cloud classification result is removed to obtain the target surface point cloud with near-surface noise suppression.

7. The method for processing high-density noise in surface point clouds of blue-green lidar according to claim 6, characterized in that, The point cloud classification method based on statistical encoding of all the spatial distribution features specifically includes: Obtain the classification model and label all the spatial distribution features statistically encoded to obtain the training sample set; The classification model is trained using the training sample set to obtain the training results, and a point cloud classification method is constructed based on the training results.

8. A surface point cloud high-density noise processing system for blue-green lidar, characterized in that, The surface point cloud high-density noise processing system of the blue-green lidar includes: The signal filtering module is used to acquire the blue-green laser reflection echo signal of the target scene area, and to filter the blue-green laser reflection echo signal to obtain near-target surface point cloud data. The feature calculation module is used to adaptively calculate the spatial distribution features of local point clouds in the near-target surface point cloud data to obtain multiple spatial structure features and multiple density distribution features, and to obtain local spatial distribution features based on all the spatial structure features and all the density distribution features. The point cloud classification module is used to encode all point cloud spatial distribution features in the local spatial distribution features to obtain terrain surface point clouds, and to classify the terrain surface point clouds to obtain target surface point clouds with near-surface noise suppression.

9. A terminal, characterized in that, The terminal includes a memory, a processor, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the surface point cloud high-density noise processing method for blue-green lidar as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which stores a surface point cloud high-density noise processing program for a blue-green lidar. When the blue-green lidar surface point cloud high-density noise processing program is executed by a processor, it implements the steps of the surface point cloud high-density noise processing method for a blue-green lidar as described in any one of claims 1-7.

Images