A noise identification and elimination method and system

Abstract

The application provides a noise identification and elimination method and system, and relates to the technical field of point cloud data processing, and comprises the following steps: obtaining point cloud data of a target area and performing standardization processing on the point cloud data to obtain standardized point cloud data, and determining layering parameters according to the vertical direction distribution characteristics of the standardized point cloud data, and then dividing the standardized point cloud data into multiple vertical layers; determining noise intensity according to the number of photons of each vertical layer; performing significance test on each vertical layer to obtain noise photons in each vertical layer, and performing an elimination operation on the noise photons to obtain preliminary screening point cloud data; performing cluster noise identification according to the area of the vertical layer and the number of photons of the preliminary screening point cloud data in each vertical layer, obtaining the remaining noise photons of the preliminary screening point cloud data in each vertical layer, and performing an elimination operation on the remaining noise photons to obtain the denoised point cloud data of the target area. The application effectively solves the precision problem of noise identification and elimination of laser radar point cloud data in complex environments such as shallow sea.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of point cloud data processing, in particular to a noise identification and elimination method and system. BACKGROUND

[0002] Photon counting laser radar has become a key tool for high-precision terrain measurement and environmental monitoring by accurately recording the reflection time information of laser photons. In actual application, the data collected by the laser radar is often disturbed by environmental factors and usually contains a large amount of noise. Especially in complex environments such as shallow sea, laser signals need to go through reflection and refraction of various media such as water surface and bottom, making the data noise more complex.

[0003] In related technologies, the existing noise identification and elimination methods mostly use simple filtering or statistical analysis methods based on threshold values. Since the density difference between signal photons and noise photons is small in shallow sea environment, the traditional threshold method is prone to misjudgment, resulting in loss of effective signal photons and residual noise photons. SUMMARY

[0004] The problem solved by the present application is how to improve the accuracy of noise identification and elimination.

[0005] To solve the above problems, the present application provides a noise identification and elimination method and system.

[0006] In a first aspect, the present application provides a noise identification and elimination method, comprising:

[0007] Obtaining point cloud data of a target area, and performing standardization processing on the point cloud data to obtain standardized point cloud data of the target area;

[0008] Determining layering parameters according to the vertical direction distribution characteristics of the standardized point cloud data, and dividing the standardized point cloud data into multiple vertical layers according to the layering parameters;

[0009] Determining noise intensity of each vertical layer according to the number of photons of the standardized point cloud data in each vertical layer;

[0010] Performing saliency test on each vertical layer according to the noise intensity to obtain noise photons of the standardized point cloud data in each vertical layer, and performing elimination operation on the noise photons to obtain preliminary screening point cloud data;

[0011] Performing cluster noise identification according to the area of each vertical layer and the number of photons of the preliminary screening point cloud data in each vertical layer to obtain residual noise photons of the preliminary screening point cloud data in each vertical layer, and performing elimination operation on the residual noise photons to obtain denoised point cloud data of the target area.

[0012] Optionally, the standardization processing of the point cloud data comprises:

[0013] acquiring three-dimensional coordinates of each photon in the point cloud data;

[0014] determining a median of X coordinates, a median of Y coordinates and a median of Z coordinates of the point cloud data according to the three-dimensional coordinates of the photon;

[0015] determining an absolute deviation of the median of X coordinates, an absolute deviation of the median of Y coordinates and an absolute deviation of the median of Z coordinates of the point cloud data according to the median of X coordinates, the median of Y coordinates and the median of Z coordinates;

[0016] standardizing the three-dimensional coordinates of each photon according to the absolute deviation of the median of X coordinates, the absolute deviation of the median of Y coordinates and the absolute deviation of the median of Z coordinates, to obtain the standardized point cloud data.

[0017] Optionally, the determining of the layering parameters according to the vertical direction distribution characteristics of the standardized point cloud data comprises:

[0018] determining a quartile range of the standardized point cloud data according to the coordinate values of all photons in the standardized point cloud data in the vertical direction;

[0019] obtaining a slice thickness of the standardized point cloud data according to the total number of photons in the standardized point cloud data and the quartile range;

[0020] obtaining a total number of slices according to the maximum and minimum coordinate values of the photons in the standardized point cloud data in the vertical direction and the slice thickness;

[0021] taking the total number of slices and the slice thickness as the layering parameters.

[0022] Optionally, the dividing of the standardized point cloud data into a plurality of vertical layers according to the layering parameters comprises:

[0023] determining a coordinate value range of the standardized point cloud data in the vertical direction according to the maximum and minimum coordinate values of the photons in the standardized point cloud data in the vertical direction;

[0024] dividing a plurality of continuous vertical intervals from the minimum coordinate value of the coordinate value range in the vertical direction to the maximum coordinate value in the vertical direction at intervals of the slice thickness, wherein the number of the vertical intervals is consistent with the total number of slices;

[0025] Traverse each of the photons in the standardized point cloud data, and according to the coordinate value of each of the photons in the vertical direction, classify it into the corresponding vertical interval, and take the set of all photon points contained in each of the vertical intervals as a vertical layer.

[0026] Optionally, the determining of the noise intensity of each of the vertical layers according to the number of photons in each of the vertical layers in the standardized point cloud data comprises:

[0027] Obtaining the number of photons of the standardized point cloud data in each of the vertical layers, and determining the projection area of each of the vertical layers in the horizontal plane;

[0028] According to the projection area, determining the projection area of the projection area, and taking the projection area as the area of the vertical layer;

[0029] The number of photons of each of the vertical layers is operated by ratio operation with the area of the vertical layer, and the noise intensity of the vertical layer is obtained.

[0030] Optionally, the significance test of each of the vertical layers according to the noise intensity, the noise photons in each of the vertical layers in the standardized point cloud data, and the noise photons are removed to obtain the preliminary screening point cloud data, comprising:

[0031] According to the noise intensity of each of the vertical layers, determining the expected number of neighbors within a preset neighborhood radius in the vertical layer;

[0032] According to the expected number of neighbors, determining the neighborhood radius of the vertical layer;

[0033] Traverse each of the photons in the vertical layer to obtain the number of observed photons within the neighborhood radius of each of the photons;

[0034] According to the number of observed photons, determining the noise probability of each of the photons;

[0035] According to the noise probability, determining the noise photons in the vertical layer;

[0036] The noise photons in each of the vertical layers are removed to obtain the preliminary screening point cloud data.

[0037] Optionally, the determining of the noise photons in the vertical layer according to the noise probability comprises:

[0038] According to the noise probability and the preset significance threshold, it is judged whether the photon in the vertical layer is the noise photon;

[0039] When the noise probability of the photon is greater than or equal to the preset significance threshold, it is determined that the photon belongs to the noise photon.

[0040] When the noise probability of the photon is less than the preset significance threshold, it is determined that the photon does not belong to the noise photon.

[0041] Optionally, the clustering noise recognition according to the area of the vertical layer and the number of photons of the preliminary screening point cloud data in each vertical layer comprises:

[0042] According to the area of each vertical layer and the number of photons of the preliminary screening point cloud data in the vertical layer, the connected radius and the minimum neighborhood point number of the vertical layer are determined.

[0043] The connected radius is taken as a neighborhood radius, and the photons of the vertical layer are processed by clustering in combination with the minimum neighborhood point number, to obtain isolated points in the vertical layer.

[0044] The photons corresponding to the isolated points are taken as the remaining noise photons.

[0045] Optionally, the rejection operation on the remaining noise photons to obtain the denoised point cloud data of the target region comprises:

[0046] The rejection operation is performed on the remaining noise photons in each vertical layer to obtain a denoised vertical layer.

[0047] According to the photons in all the denoised vertical layers, the denoised point cloud data is generated.

[0048] In a second aspect, a noise recognition and rejection system is provided, comprising:

[0049] An acquisition unit is configured to acquire point cloud data of a target region, and perform standardization processing on the point cloud data to obtain standardized point cloud data of the target region.

[0050] A division unit is configured to determine layering parameters according to the vertical direction distribution characteristics of the standardized point cloud data, and divide the standardized point cloud data into a plurality of vertical layers according to the layering parameters.

[0051] A noise intensity analysis unit is configured to determine the noise intensity of each vertical layer according to the number of photons of the standardized point cloud data in the vertical layer.

[0052] A noise elimination unit is configured to perform a saliency test on each of the vertical layers according to the noise intensity, to obtain noise photons of the standardized point cloud data in each of the vertical layers, and to perform an elimination operation on the noise photons to obtain preliminary screening point cloud data;

[0053] A de-noising processing unit is configured to perform a cluster noise identification according to the area of the vertical layer and the number of photons of the preliminary screening point cloud data in each of the vertical layers, to obtain residual noise photons of the preliminary screening point cloud data in each of the vertical layers, and to perform an elimination operation on the residual noise photons to obtain de-noised point cloud data of the target region.

[0054] The noise identification and elimination method and system can obtain point cloud data of a target region and perform standardization processing to obtain standardized point cloud data, which can eliminate dimensional differences and abnormal values of the data, making subsequent processing more stable and accurate, and providing a unified data basis for subsequent layering and noise identification, and avoiding misjudgment caused by dimensional differences of the data. According to vertical distribution characteristics of the standardized point cloud data, layering parameters are determined, and the point cloud data is divided into multiple vertical layers. In a shallow sea environment, laser signals will experience reflection and refraction of multiple media such as water surface and water bottom, and noise distribution has obvious vertical direction differences. Through vertical layering, noise of different media can be processed separately, avoiding misjudgment in global processing. At the same time, noise characteristics of different vertical layers are processed specifically, improving the accuracy of noise identification. The number of photons is used to evaluate the noise intensity, and combined with saliency test, noise photons can be more accurately identified, effectively reducing misjudgment, avoiding loss of effective signal photons, reducing noise photon residues, and improving the accuracy of noise elimination. At the same time, cluster noise identification can further refine noise identification, accurately eliminate noise in local areas, make up for noise that may be missed by saliency test, further improve the accuracy of noise identification, and ensure that noise photons are completely eliminated and effective signal photons are retained in the final obtained point cloud data.

[0055] The present application can accurately identify complex noise distribution in a shallow sea environment through vertical layering and cluster noise identification, avoiding misjudgment of traditional methods. Through the combination of saliency test and cluster noise identification, effective signal photons can be maximally retained while noise photons are eliminated. The technical scheme of the present application is particularly suitable for complex environments such as shallow seas, and can effectively deal with noise problems caused by reflection and refraction of laser signals in multiple media. At the same time, the de-noised point cloud data obtained finally has higher quality, which can provide more accurate data support for subsequent terrain measurement and environmental monitoring.

[0056] In summary, the present application effectively solves the precision problem of noise recognition and elimination of lidar point cloud data in complex environments such as shallow sea by standardization processing, vertical layering, noise intensity evaluation, significance test and cluster noise identification operations. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 A flow chart of the noise recognition and elimination method of the embodiment of the present application is shown in the figure.

[0058] Figure 2 A schematic diagram of the collected original three-dimensional photon data of the embodiment of the present application is shown in the figure.

[0059] Figure 3 A schematic diagram of the three-dimensional photon data layering of the embodiment of the present application is shown in the figure.

[0060] Figure 4 A schematic diagram of the photon probability density distribution and water surface photon demarcation line determination of the embodiment of the present application is shown in the figure.

[0061] Figure 5 A schematic diagram of the water surface and water bottom signal photon extraction result of the embodiment of the present application is shown in the figure.

[0062] Figure 6 A structural block diagram of the noise recognition and elimination system of the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0063] In order to make the above objectives, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms, and should not be interpreted as being limited to the embodiments described herein, on the contrary, these embodiments are provided to make the present application more thorough and complete. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes, and are not intended to limit the scope of protection of the present application.

[0064] It should be understood that each step described in the method embodiment of the present application can be executed in different order and / or in parallel. In addition, the method embodiment can include additional steps and / or omit the execution of the steps shown. The scope of the present application is not limited in this respect.

[0065] The term "include," and derivations thereof, is an open term that means "including, but not limited to"; the term "based on" means "based, at least in part, on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optional" means "optional in at least some embodiments". Related definitions are given throughout the description. It is to be noted that the concepts mentioned in the present application are illustrative and not restrictive, and those skilled in the art should understand that "one", "a plurality of" or "a plurality of" mentioned in the present application are illustrative and not restrictive, and those skilled in the art should understand that "one" or "a plurality of" should be understood as "one or more" unless otherwise explicitly indicated in the context.

[0066] It should be noted that the modification of "one" or "a plurality" mentioned in the present application is illustrative and not restrictive, and those skilled in the art should understand that "one" or "a plurality" should be understood as "one or more" unless otherwise explicitly indicated in the context.

[0067] In combination Figure 1 As shown, the noise recognition and elimination method provided by the embodiments of the present application comprises:

[0068] Obtaining point cloud data of a target area, and performing standardization processing on the point cloud data to obtain standardized point cloud data of the target area.

[0069] Specifically, the target area is scanned by a photon counting laser radar system to collect photon point cloud data in the area, which contains the position information of photons in three-dimensional space. The obtained point cloud data is processed to eliminate the dimensional difference of different coordinate axes and the influence of outliers in the data. Specifically, the median and median absolute deviation of each coordinate axis (X, Y, Z) are calculated, and then the coordinates of each photon are converted according to these statistics, so that the data is converted into a form of zero mean and unit standard deviation, thereby obtaining the standardized point cloud data.

[0070] According to the vertical direction distribution characteristics of the standardized point cloud data, the layering parameters are determined, and the standardized point cloud data is divided into a plurality of vertical layers according to the layering parameters.

[0071] Specifically, firstly, the vertical distribution of the standardized point cloud data is analyzed, and parameters such as interquartile range (IVR) are used to determine the layering parameters to help determine the appropriate slice thickness. Simultaneously, the total number of photons and the vertical coordinate range (maximum and minimum values) are combined to determine the total number of slices. Slice thickness and the total number of slices together constitute the layering parameters, used to determine how to divide the entire data range into multiple layers vertically. After determining the layering parameters, the standardized point cloud data is divided into multiple vertical layers according to these parameters. Each vertical layer contains photon points within a certain range, allowing data at different depth ranges to be processed separately. Through layered processing, more refined analysis and processing can be performed on the characteristics of each vertical layer, such as subsequent noise identification and removal operations, thereby improving the accuracy and efficiency of the processing.

[0072] The noise intensity of the vertical layer is determined based on the number of photons in each vertical layer using the standardized point cloud data.

[0073] Specifically, the noise intensity of each layer is calculated, for each layer... Calculate its noise intensity.

[0074] Noise intensity This reflects the density of noise photons in this layer and can be calculated using the following formula:

[0075] ;

[0076] in: Let represent the noise intensity of the j-th layer, and let represent the number of noise photons per unit area of ​​that layer. Indicates the first j The number of photon points in the layer. Indicates the first j The area of ​​a layer is usually the projected area of ​​that layer on the xy plane.

[0077] For each vertical layer, the number of photons within that layer is counted to provide basic data for subsequent noise intensity calculations. By analyzing the x and y coordinate ranges of all photons within the layer, the coverage area of ​​that layer in the xy-plane is calculated, thus obtaining its projected area. Using the formula described above, the noise intensity of each vertical layer is calculated based on the number of photons and the area of ​​that region. Noise intensity reflects the density of noise photons in that layer, i.e., the number of noise photons per unit area. In this way, the distribution of noise photons in each vertical layer can be quantified, providing important data for further noise identification and removal.

[0078] According to the noise intensity, a significance test is performed on each of the vertical layers to obtain noise photons of the standardized point cloud data in each of the vertical layers, and the noise photons are removed to obtain preliminary screening point cloud data.

[0079] Specifically, in each vertical layer, the neighborhood radius is adaptively adjusted according to the noise intensity of the layer. First, according to the formula:

[0080] ;

[0081] wherein: denotes the noise intensity of the i-th layer. j denotes the neighborhood radius to be measured. denotes the area of the neighborhood. denotes the number of neighbors that can theoretically be seen within this radius under the pure noise condition (expected value). The expected number of neighbors within a certain radius under the pure noise condition is calculated. Then, to control the false positive risk, the neighborhood radius

[0082] , is selected to satisfy the probability of at least one neighboring point appearing within the radius neighborhood under the pure noise assumption not more than :

[0083] ;

[0084] wherein: denotes the number of points actually counted in the neighborhood with a radius of ; is the upper limit probability of at least one neighboring point appearing within the radius neighborhood; denotes the probability of at least one neighboring point appearing within the neighborhood with a radius of , denotes the expected probability of at least one point appearing within the neighborhood with a radius of under the pure noise condition, denotes the expected number of points within the neighborhood with a radius of under the pure noise condition.

[0085] Therefore, the target expected number of neighbors corresponding to this upper limit probability is:

[0086] ;

[0087] Thus, the uniform neighborhood radius within the layer is obtained as:

[0088] ;

[0089] wherein: ​This refers to the uniform radius used in this layer; the larger the radius, the denser the noise. It will automatically shrink, and vice versa.

[0090] Based on the above formula, the target expected number of neighbors is determined, satisfying the pure noise assumption, where the probability of having at least one neighboring point within the radius neighborhood does not exceed a certain threshold. Finally, a uniform neighborhood radius is obtained within the layer. The neighborhood radius is adjusted based on noise intensity; the denser the noise, the larger the neighborhood radius. It will automatically shrink, and vice versa.

[0091] Each layer of photon points may be affected by different noise levels. This embodiment uses a significance test and the Benjamini-Hochberg procedure to control the false positive rate, thereby more accurately eliminating noisy photons. The significance test (…) p- (Value calculation) is used to measure whether a photon point is a noise photon. p- The value is calculated based on the Poisson distribution, representing the probability that such an extreme number of photons would be observed when the photon point is assumed to be noise. p - Value calculation formula:

[0092] ;

[0093] in: k This indicates the number of observed photons at that point. Indicates the first j The noise intensity of the layer, Indicates the first i A photon point p The - value represents the probability that the photon is a noise photon. When processing multiple... p When the value is -, the Benjamini-Hochberg procedure is used to control the false positive rate. This is achieved by setting a significance threshold. According to p- The significance threshold is used to eliminate noise photons. The formula for the significance threshold is:

[0094] ;

[0095] in: No. j The significance threshold of the layer No. j The number of photon points in the layer, if p - Value greater than the significance threshold If the photon is not identified as a noise photon, it will be removed from the point cloud data.

[0096] According to the area of the vertical layer and the number of photons of the preliminary screening point cloud data in each vertical layer, cluster noise recognition is performed to obtain the remaining noise photons of the preliminary screening point cloud data in each vertical layer, and the remaining noise photons are removed to obtain the denoised point cloud data of the target region.

[0097] Specifically, in each vertical layer, connectivity analysis is performed. First, to ensure connectivity between photon points, a connectivity radius is defined, which is the maximum allowed distance between two points. If the distance between two points is less than the connectivity radius, they are considered to be connected and belong to the same signal region. The formula for calculating the connectivity radius is as follows:

[0098] ;

[0099] wherein: represents the connectivity radius of the jth layer. represents the area of the jth layer, usually the projection area of the layer in the x-y plane, represents the number of photon points in the jth layer.

[0100] Then, by defining the connectivity radius, it is ensured that the photon points in the same signal region are spatially connected. If the distance between two points is less than the connectivity radius, they are considered to be connected and belong to the same signal region. The goal of cluster analysis is to cluster photon points through the DBSCAN clustering algorithm and distinguish signal photons and noise photons based on spatial density. In particular, the connectivity radius is taken as the neighborhood radius Eps of DBSCAN, and the minimum number of neighborhood points MinPts is adaptively set according to the number of photon points in each layer, thereby improving the removal accuracy of noise photons and ensuring the accurate extraction of signal photons. DBSCAN is a density-based clustering algorithm that defines core points through neighborhood radius Eps and minimum neighborhood point number MinPts , and identifies signal photons through density expansion. Its basic principle is as follows:

[0101] Definition of core point:

[0102] ;

[0103] wherein: represents the p neighborhood of point Eps , i.e. all points with a distance less than or equal to p from point Eps . : The Euclidean distance between point p and point q .

[0104] If the number of points in MinPts is greater than or equal to p , the point is considered a core point and is directly density reachable, if the point satisfies the following conditions: ; and is a core point, then and are considered to be directly density reachable.

[0105] In order to adapt to the change of data density of different layers, the minimum neighborhood point number MinPts is adaptively set based on the number of photon points of each layer. Generally, the setting of MinPts depends on the number of photon points of the layer and the density of the layer. The following adaptive setting method is adopted:

[0106] ;

[0107] wherein: represents the number of photon points of the jth layer. represents a constant for adjusting the photon point density of the layer, usually taking 3 or 4. Starting from each core point, expand the cluster and add all directly density reachable points to the cluster. Remove noise points from the data. Noise points are usually points with low density and cannot form clusters. After the noise identification and removal operation, the remaining noise photons in each vertical layer are removed, and the denoised point cloud data of the target area is finally obtained. These data retain the signal photons and remove the noise photons, thereby improving the processing accuracy of the photon counting lidar data and providing more accurate data support for subsequent applications such as depth measurement and ocean environment monitoring.

[0108] The noise recognition and elimination method of the embodiment, by acquiring the point cloud data of the target area and performing standardization processing thereon, obtains standardized point cloud data, the standardization processing can eliminate the dimensional difference and abnormal value of the data, so that the subsequent processing is more stable and accurate, and at the same time provides a unified data basis for subsequent layering and noise recognition, avoiding misjudgment caused by dimensional difference of data. Then, according to the vertical direction distribution characteristics of the standardized point cloud data, the layering parameters are determined, and the point cloud data is divided into multiple vertical layers. In the shallow sea environment, the laser signal will experience reflection and refraction of various media such as water surface and water bottom, and the noise distribution has obvious vertical direction difference. Through vertical layering, the noise of different media can be processed separately, avoiding misjudgment in global processing. At the same time, according to the noise characteristics of different vertical layers, targeted processing is carried out to improve the accuracy of noise recognition. By evaluating the noise intensity through the number of photons and combining the significance test, the noise photons can be more accurately identified, effectively reducing the misjudgment, avoiding the loss of effective signal photons, and at the same time reducing the residual noise photons, improving the precision of noise elimination. At the same time, the cluster noise recognition can further refine the noise recognition, accurately eliminate the noise in the local area, make up for the noise that may be missed by the significance test, further improve the accuracy of noise recognition, and ensure that the noise photons in the finally obtained point cloud data are completely eliminated and the effective signal photons are retained.

[0109] The embodiment can accurately identify the complex noise distribution in the shallow sea environment through vertical layering and cluster noise recognition, avoiding the misjudgment of traditional methods. And through the combination of significance test and cluster noise recognition, the effective signal photons are maximized while the noise photons are eliminated. The technical scheme of the present application is particularly suitable for complex environments such as shallow sea, and can effectively deal with the noise problem caused by reflection and refraction of laser signal in multiple media. At the same time, the quality of the finally obtained denoised point cloud data is higher, which can provide more accurate data support for subsequent terrain measurement and environment monitoring.

[0110] In summary, the embodiment effectively solves the precision problem of laser radar point cloud data noise recognition and elimination in complex environments such as shallow sea through standardization processing, vertical layering, noise intensity evaluation, significance test and cluster noise recognition.

[0111] Optionally, the standardization processing of the point cloud data obtains the standardized point cloud data of the target area, comprising:

[0112] acquiring the three-dimensional coordinates of each photon in the point cloud data;

[0113] determining the median of X coordinate, the median of Y coordinate and the median of Z coordinate of the point cloud data according to the three-dimensional coordinates of the photon;

[0114] According to the X coordinate median, the Y coordinate median and the Z coordinate median, determine the X coordinate median absolute deviation, the Y coordinate median absolute deviation and the Z coordinate median absolute deviation of the point cloud data;

[0115] According to the X coordinate median absolute deviation, the Y coordinate median absolute deviation and the Z coordinate median absolute deviation, standardize the three-dimensional coordinates of each photon to obtain the standardized point cloud data.

[0116] Specifically, the original point cloud data of a target region is obtained from a photon counting laser radar system. These data are presented in the form of a three-dimensional point cloud, and the position of each photon is represented by its three-dimensional coordinates (x, y, z). The x coordinate values of all photons are sorted from small to large, and the value located in the middle position is found, which is the X coordinate median. The y coordinate values of all photons are sorted from small to large, and the value located in the middle position is found, which is the Y coordinate median. The z coordinate values of all photons are sorted from small to large, and the value located in the middle position is found, which is the Z coordinate median. If the number of photons is even, take the average of the two middle values as the median. By calculating the median of each coordinate axis, a central reference point can be provided for subsequent standardization processing.

[0117] For the x coordinate of each photon, calculate the absolute difference between it and the X coordinate median, and then sort these absolute differences from small to large, and find the value located in the middle position, which is the X coordinate median absolute deviation. For the y coordinate of each photon, calculate the absolute difference between it and the Y coordinate median, and then sort these absolute differences from small to large, and find the value located in the middle position, which is the Y coordinate median absolute deviation. For the z coordinate of each photon, calculate the absolute difference between it and the Z coordinate median, and then sort these absolute differences from small to large, and find the value located in the middle position, which is the Z coordinate median absolute deviation. If the number of photons is even, take the average of the two middle values as the median absolute deviation. By calculating the median absolute deviation (MAD) of each coordinate axis, a scale reference can be provided for subsequent standardization processing.

[0118] For the x coordinate of each photon, subtract the X coordinate median from the x coordinate value of the photon, and then divide by the X coordinate median absolute deviation to obtain the standardized x coordinate value. For the y coordinate of each photon, subtract the Y coordinate median from the y coordinate value of the photon, and then divide by the Y coordinate median absolute deviation to obtain the standardized y coordinate value. For the z coordinate of each photon, subtract the Z coordinate median from the z coordinate value of the photon, and then divide by the Z coordinate median absolute deviation to obtain the standardized z coordinate value.

[0119] In a preferred embodiment of the present application, as shown inFigure 2 The raw three-dimensional photon point cloud data obtained by the airborne photon counting lidar (PCL) system is shown. The initial data form without noise processing is intuitively displayed. A large number of photons are presented in the form of discrete points in three-dimensional space (x, y, z axes correspond to plane position and depth, respectively, x axis points to the east, and the value increases to represent eastward movement; y axis points to the north, and the value increases to represent northward movement; z axis is the elevation, upward is positive, and the value increases to represent height increase). It contains effective signal photons (such as water surface and bottom reflected photons) and noise photons (such as sun background and bubble interference photons), and the overall point distribution has no obvious hierarchical or clustering rule.

[0120] The raw data collected by the photon counting lidar can be represented as a three-dimensional point cloud:

[0121] ;

[0122] Among them: is the coordinate of the i-th photon in three-dimensional space, is the total number of photon points collected.

[0123] At the same time, in order to eliminate the dimensional differences of different coordinate axes in the data and the influence of outliers, the obtained photon point cloud data is standardized. Specifically, the median (med(x), med(y), med(z)) and the median absolute deviation (MAD(x), MAD(y), MAD(z)) of each coordinate axis (x, y, z) are calculated, and then the coordinates of each photon point are converted according to the standardization formula to obtain the standardized point cloud data.

[0124] The standardization formula is:

[0125] ;

[0126] Among them: med(x), med(y), med(z) are the medians of the x, y, and z coordinates, respectively. x , y , z are the median absolute deviations of the x, y, and z coordinates, respectively. The deviation of the data points from the median is used to measure the dispersion of the data. MAD(x), MAD(y), MAD(z) x y z The standardized data form is:

[0127]

[0128] ;

[0129] ​​​​​At this time, all the photon point cloud data is converted into a form of zero mean and unit standard deviation, and after standardization, all the photon point cloud data is converted into a form of zero mean and unit standard deviation, which provides a unified coordinate system for subsequent data analysis and processing, and facilitates more accurate calculation and analysis.

[0130] In this optional embodiment, the coordinates of each photon are converted into a form of zero mean and unit standard deviation through standardization. Thus, the dimensional differences on different coordinate axes are eliminated, and the data has comparability in each dimension, which provides a unified coordinate system for subsequent data analysis and processing; the standardized point cloud data can be better used for subsequent processing steps such as noise identification and elimination.

[0131] Optionally, the determining of the layering parameter according to the vertical direction distribution feature of the standardized point cloud data comprises:

[0132] determining a quartile range of the standardized point cloud data according to the coordinate values of all photons in the standardized point cloud data in the vertical direction;

[0133] obtaining a slice thickness of the standardized point cloud data according to the total number of photons in the standardized point cloud data in combination with the quartile range;

[0134] obtaining a total number of slices according to the maximum and minimum coordinate values of the photons in the standardized point cloud data in the vertical direction in combination with the slice thickness;

[0135] taking the total number of slices and the slice thickness as the layering parameter.

[0136] Specifically, first, the coordinate values of all photons in the vertical direction (usually the z-axis) are found, and these coordinate values are sorted from small to large, and then three points that divide the data into four equal parts are found, i.e. the first quartile (Q1), the second quartile (median, Q2) and the third quartile (Q3). The interquartile range (IQR) is the difference between the third quartile (Q3) and the first quartile (Q1), which is used to reflect the dispersion of the data in the vertical direction. The slice thickness is determined using the interquartile range (IQR) and the total number of photons in the point cloud data, wherein the slice thickness determines how thick the data is divided into layers in the vertical direction.

[0137] In a preferred embodiment of the present invention, the standardized point cloud data is analyzed to obtain its distribution in the vertical direction (z-axis direction). Through statistical analysis and other methods, the density variations and distribution patterns of the data within different depth ranges are understood, thus providing a basis for determining the stratification parameters. Based on the vertical distribution characteristics, an appropriate stratification method and parameters are selected. For example, the slice thickness can be automatically determined using the Freedman–Diaconis rule. ).

[0138] ;

[0139] in: IQR(z) :for z The interquartile range of dimensional data, which is the difference between the upper quartile (Q3) and the lower quartile (Q1), is used to measure the middle range of the data. N It refers to the number of points in the data. K It represents the total number of slices, calculated using the ratio of the maximum and minimum water depths to the slice thickness.

[0140] The Freedman-Diaconis rule is used to determine the data segmentation width, dynamically adjusting the slice width by calculating the interquartile range (IQR) of the data. This method adapts to the distribution characteristics of the data. Compared to simple fixed-width slicing, this method is more adaptive. Specifically, it calculates the interquartile range of the data along the z-axis (…). IQR(z) The value is the difference between the upper quartile (Q3) and the lower quartile (Q1), and the slice thickness is then calculated using the formula described above. After determining the slice thickness, the total number of slices K is calculated based on the maximum and minimum water depths.

[0141] like Figure 3 As shown, the result of vertically slicing (layering) the original 3D photon point cloud according to the Freedman–Diaconis rule separates photon data at different depths. The slice thickness Δz is automatically determined by calculating the interquartile range (IQR) of the z-axis (depth dimension) data, dividing the original point cloud into multiple independent depth layers. Figure 3 (Different regions or colors are used to distinguish each layer), so that the shallow layer with strong background photons, the deep layer with sparse signal photons, and the near-bottom layer with strong reflection photons are in different layers, thus avoiding interference between data from different depths.

[0142] After slicing, the point cloud data is divided into multiple layers:

[0143] ;

[0144] in: It is the set of photons in the j-th layer. represents the i-th photon point in the point cloud data, i represents the set of all photon points in the entire point cloud data set, P represents the i-th photon point in the point cloud data, represents the start depth value of the i-th layer, j represents the depth value of the i-th photon point, represents the depth value of the i-th photon point, i represents the thickness of the slice, i.e., the depth interval of each layer, represents the index of the layer, which takes a value ranging from 1 to j , K is the total number of layers, K represents the total number of layers after slicing. K According to the determined layering parameters, the standardized point cloud data is divided into multiple vertical layers. Each vertical layer contains a range of photon points, and the photon points belonging to the same depth range are classified into a layer, forming multiple layered point cloud data, so as to subsequently perform separate processing on different depth layers and better cope with the differences between signal photons and noise photons at different depths.

[0145] In this embodiment, the Freedman-Diaconis rule is combined with IQR and the total number of photons to dynamically calculate the slice thickness, so that the slice thickness can adapt to the distribution characteristics of the data, rather than simply using a fixed value. The maximum and minimum values of the coordinates of the photons in the vertical direction in the standardized point cloud data are obtained, and these two values define the range of the data in the vertical direction. Then, the range is divided by the previously calculated slice thickness to obtain the total number of slices, which represents how many layers the data is divided into in the vertical direction for subsequent separate processing of each layer. The calculated total number of slices and slice thickness are used as layering parameters. These two parameters together determine how to divide the standardized point cloud data into multiple layers in the vertical direction for subsequent separate processing of each layer, such as noise identification and removal. The layering parameters are determined based on the distribution characteristics of the data itself, making the processing process more adaptive and targeted.

[0146]

[0147] ​In this optional embodiment, the noise photons can be more accurately identified and removed through a hierarchical clustering strategy. In shallow water environments, due to complex environmental factors such as strong solar background, sea surface reflection, water attenuation, etc., traditional noise removal methods often cannot effectively distinguish between signal photons and noise photons. The present application processes different depth photon points separately, and then combines connectivity analysis and adaptive DBSCAN clustering algorithm to adjust the noise removal standard according to the density characteristics of different depth layers, thereby significantly improving the accuracy of data processing. At the same time, by using hierarchical processing, the complex three-dimensional point cloud data is decomposed into multiple depth layers for separate processing. This hierarchical strategy not only simplifies the complexity of data processing, but also enables more efficient allocation of computing resources to each depth layer. By performing connectivity analysis and clustering analysis independently in each layer, noise photons can be more quickly identified and removed, thereby improving the overall efficiency of data processing. This embodiment is suitable for a variety of application scenarios, including ocean remote sensing, environmental monitoring, underwater terrain investigation, etc. Through adaptive parameter setting, the processing strategy can be automatically adjusted according to the data characteristics in different environments, thereby effectively identifying and removing noise photons in various complex environments. This strong adaptability makes the present application widely applicable in practical applications.

[0148] Optionally, the dividing the standardized point cloud data into a plurality of vertical layers according to the hierarchical parameters comprises:

[0149] Determining a coordinate value range of the standardized point cloud data in the vertical direction according to a maximum value and a minimum value of the coordinates of the photons in the vertical direction in the standardized point cloud data;

[0150] Dividing a plurality of continuous vertical intervals from the minimum value of the coordinate value range of the vertical direction to the maximum value of the coordinate value range of the vertical direction at intervals of the slice thickness, wherein the number of the vertical intervals is consistent with the total number of the slices;

[0151] Traversing each of the photons in the standardized point cloud data, and according to the coordinate value of each of the photons in the vertical direction, the photons are classified into the corresponding vertical interval, and all the photon points contained in each of the vertical intervals are taken as a vertical layer.

[0152] Specifically, according to the coordinate values of all the photons in the vertical direction (usually the z-axis) in the standardized point cloud data, the maximum value and the minimum value in these coordinate values are found, and these two values define the coordinate value range of the entire data in the vertical direction, which is the basis for subsequent division of vertical layers.

[0153] According to the determined slice thickness, a plurality of continuous vertical intervals are divided in sequence from the minimum value of the vertical direction coordinate value range, with the slice thickness as the interval. These vertical intervals are arranged in sequence from the minimum value to the maximum value along the vertical direction, and each interval represents a vertical layer. At the same time, the number of the divided vertical intervals is consistent with the total number of slices obtained, so as to ensure that the entire vertical direction range is covered. Each of the photons in the standardized point cloud data is traversed to view the coordinate value of each photon in the vertical direction. According to the coordinate value, it is determined which vertical interval the photon belongs to, and then it is classified into the corresponding vertical interval. Finally, all the photon point sets contained in each vertical interval constitute a vertical layer, and the entire standardized point cloud data is divided into a plurality of vertical layers. The photon points in each layer are adjacent and have similar depth characteristics in the vertical direction. Through the above steps, the standardized point cloud data is divided into a plurality of vertical layers according to the depth characteristics in the vertical direction, providing more refined processing units for subsequent noise identification and elimination operations.

[0154] In this optional embodiment, by dividing the standardized point cloud data into a plurality of vertical layers, fine processing can be performed on the data in different depth ranges. In a shallow water environment, photon data at different depths has different noise characteristics and signal distribution rules. Layered processing makes the data in each vertical layer have relatively consistent characteristics, thereby providing a more accurate processing basis for subsequent noise identification and elimination.

[0155] Moreover, after the layered processing, the number of photon points in each vertical layer is relatively small, and has similar depth characteristics, so that when noise identification is performed in each layer, signal photons and noise photons can be more accurately distinguished. By independently performing noise identification and elimination in each vertical layer, the misjudgment problem caused by the difference in noise characteristics at different depths during global processing can be avoided, thereby improving the accuracy of noise identification.

[0156] In practical applications, different vertical layers can have different noise intensities and signal distribution patterns. Through the hierarchical processing of the embodiment, the noise identification and rejection strategy can be adaptively adjusted according to the characteristics of each layer. For example, in the shallow layer with strong noise, a stricter noise rejection standard can be adopted; while in the deep layer with weak signal, a more moderate strategy can be adopted to avoid false deletion of signal photons. This adaptability enables the algorithm to better cope with data processing needs in complex environments. Hierarchical processing decomposes complex data processing tasks into multiple relatively simple subtasks, each of which only involves data within a vertical layer. This decomposition not only simplifies the complexity of data processing, but also enables more efficient allocation of computing resources to each vertical layer. By processing multiple vertical layers in parallel, the overall data processing efficiency can be significantly improved. When identifying and rejecting noise within each vertical layer, high-density signal clusters and isolated noise points can be more accurately identified. In this way, false deletion of signal photons can be effectively avoided, thereby preserving the integrity of signal photons. This is crucial for subsequent applications such as marine remote sensing, environmental monitoring, and underwater terrain investigation, as the integrity of signal photons directly affects the accuracy and reliability of these applications.

[0157] Through hierarchical processing, the data processing process within each vertical layer is more consistent and repeatable. This consistency enables data collected at different times or different locations to be processed in the same way, thereby improving the reproducibility of data processing results.

[0158] Optionally, the method further comprises:

[0159] acquiring the number of photons of the standardized point cloud data within each vertical layer, and determining the projection area of each vertical layer on the horizontal plane;

[0160] determining the projection area of the projection area as the area of the vertical layer;

[0161] performing a ratio operation on the number of photons of each vertical layer and the area of the vertical layer to obtain the noise intensity of the vertical layer.

[0162] Specifically, by traversing the standardized point cloud data, each photon is assigned to a corresponding vertical layer according to its vertical direction coordinate value, and then the total number of photons in each vertical layer is counted. At the same time, the projection area of each vertical layer on the horizontal plane is determined, which refers to the projection range of all photons in the vertical layer on the horizontal plane (usually the x-y plane), which is determined by analyzing the x and y coordinates of the photons. The projection area is calculated by determining the boundary of the projection area, for example, by finding the maximum and minimum values of the x and y coordinates in the projection area to determine a rectangular area, and calculating the area of the rectangle. This projection area will be used as the area of the vertical layer for subsequent noise intensity calculation. The noise intensity is obtained by dividing the number of photons in the vertical layer by its corresponding area. This ratio reflects the photon density per unit area in the vertical layer and is used as a quantitative indicator of noise intensity. The higher the noise intensity, the greater the noise photon density in the vertical layer, and vice versa.

[0163] In this optional embodiment, by calculating the noise intensity of each vertical layer, the noise level in each depth range can be accurately quantified, which provides an important reference for subsequent noise identification and removal. Different vertical layers at different depths may have different noise characteristics. By calculating the noise intensity of each vertical layer separately, the noise changes in different depth ranges can be better adapted to, thereby improving the relevance and effectiveness of noise processing. The quantification results of noise intensity can be used to optimize the subsequent noise processing strategy. For example, stricter noise removal standards can be used in layers with high noise intensity, while more moderate strategies can be used in layers with low noise intensity, thereby effectively removing noise photons while ensuring the integrity of signal photons. By accurately quantifying the noise intensity, noise photons can be more accurately identified and removed, thereby improving the overall reliability of data processing.

[0164] Optionally, the noise photons in each vertical layer are obtained by performing a significance test on each vertical layer according to the noise intensity, and the noise photons are removed to obtain preliminary screening point cloud data, including:

[0165] According to the noise intensity of each vertical layer, the expected number of neighbors within a preset neighborhood radius in the vertical layer is determined;

[0166] According to the expected number of neighbors, the neighborhood radius of the vertical layer is determined;

[0167] Each photon in the vertical layer is traversed to obtain the number of observed photons within the neighborhood radius of each photon;

[0168] According to the number of observed photons, the noise probability of each photon is determined;

[0169] determining the noise photons in the vertical layer according to the noise probability;

[0170] performing a rejection operation on the noise photons in each of the vertical layers to obtain the preliminary screened point cloud data.

[0171] Specifically, in each vertical layer, the expected number of neighbors within a preset neighborhood radius is determined according to the noise intensity of the layer, and the noise intensity is used to reflect the density of noise photons in the layer. In the embodiment, the expected number of neighbors can be estimated according to the noise intensity, which is the number of neighbor photons that should theoretically exist within a given neighborhood radius. The expected number of neighbors is an important basis for subsequent judgment of whether a photon is a noise photon. Based on the expected number of neighbors, the neighborhood radius of each vertical layer is determined. At the same time, the selection of the neighborhood radius needs to ensure that the distribution of photons within the radius can reflect the characteristics of noise. If the expected number of neighbors is high, it means that the noise photons are relatively dense, and the neighborhood radius can be appropriately reduced. Conversely, if the expected number of neighbors is low, the neighborhood radius can be appropriately increased. In this way, the neighborhood radius can adapt to the vertical layers with different noise intensities.

[0172] By calculating the distance between each photon and its surrounding photons, all photons in each vertical layer are traversed. For each photon, the actual observed number of photons within a preset neighborhood radius is found. The photons with a distance less than or equal to the neighborhood radius are counted as neighbor photons, and the number of these neighbor photons is counted. The observed number of photons reflects the actual distribution of adjacent photons in the local area of each photon. According to the observed number of photons within the neighborhood radius of each photon, combined with the noise intensity of the vertical layer, the noise probability of each photon is calculated. The noise probability is used to reflect the possibility that the photon is a noise photon. Specifically, if the observed number of photons is much lower than the expected number of neighbors, it means that the photon is more likely to be a noise photon, and its noise probability is higher. Conversely, if the observed number of photons is close to or higher than the expected number of neighbors, it means that the photon is more likely to be a signal photon, and its noise probability is lower. According to the noise probability of each photon, it is determined whether the photon is a noise photon. In an optional embodiment, a threshold can be set, and when the noise probability of a photon is higher than the threshold, it is determined to be a noise photon; otherwise, it is determined to be a signal photon. In this way, noise photons can be distinguished from signal photons. In each vertical layer, the points determined to be noise photons are removed from the point cloud data, and the data in each vertical layer is preliminarily screened. At the same time, the points that are more likely to be signal photons are retained. The screening and purification results of all vertical layers are combined to obtain the preliminary screened point cloud data, which is the preliminary result after noise rejection and provides a more accurate basis for further data processing and analysis.

[0173] In the optional embodiment, the noise photons in each vertical layer can be accurately identified through the significance test. The neighborhood radius and the expected number of neighbors are adaptively adjusted according to the noise intensity of each vertical layer, so that the algorithm can better adapt to data with different noise levels, improving the flexibility and adaptability of the processing. By calculating the noise probability and setting a reasonable threshold, the situation of misjudging signal photons as noise photons can be effectively reduced, while avoiding missing the real noise photons, thereby improving the reliability of data processing. By removing the noise photons, the point cloud data after preliminary screening is obtained, which significantly improves the quality of the data. The data after preliminary screening is closer to the real signal distribution, providing more accurate data support for subsequent applications such as marine remote sensing, environmental monitoring and underwater terrain investigation.

[0174] Optionally, the determining the noise photons in the vertical layer according to the noise probability comprises:

[0175] determining whether the photon in the vertical layer is the noise photon according to the noise probability and a preset significance threshold;

[0176] wherein, when the noise probability of the photon is greater than or equal to the preset significance threshold, it is determined that the photon belongs to the noise photon;

[0177] when the noise probability of the photon is less than the preset significance threshold, it is determined that the photon does not belong to the noise photon.

[0178] Specifically, in each vertical layer, whether a photon is a noise photon is determined by comparing the noise probability of the photon with a preset significance threshold. In the embodiment, the significance threshold is a pre-set reference value for distinguishing noise photons and signal photons. If the noise probability of a photon is greater than or equal to the preset significance threshold, it means that the photon is more likely to be a noise photon. In this case, the photon is determined to be a noise photon. The determination method of the embodiment is based on the level of noise probability, ensuring that only those photons with higher noise probability are identified as noise photons. On the contrary, if the noise probability of a photon is less than the preset significance threshold, it means that the photon is more likely to be a signal photon. In this case, the photon is determined not to be a noise photon. This determination method helps to retain those photons with lower noise probability, thereby reducing the possibility of misjudgment.

[0179] In the optional embodiment, by setting the significance threshold, the noise photons can be accurately identified, and based on the level of noise probability, it is ensured that only those photons with higher noise probability are identified as noise photons, thereby improving the accuracy of noise identification. By comparing the noise probability with the significance threshold, the situation of misjudging signal photons as noise photons can be effectively reduced, while avoiding missing the real noise photons.

[0180] Optionally, the clustering noise recognition is performed according to the area of the vertical layer and the number of photons of the preliminary screening point cloud data in each vertical layer, to obtain the remaining noise photons of the preliminary screening point cloud data in each vertical layer, including:

[0181] According to the area of each vertical layer and the number of photons of the preliminary screening point cloud data in the vertical layer, the connected radius and the minimum neighborhood point number of the vertical layer are determined;

[0182] The connected radius is taken as the neighborhood radius, and the photons of the vertical layer are processed by clustering in combination with the minimum neighborhood point number, to obtain the isolated points in the vertical layer;

[0183] The photons corresponding to the isolated points are taken as the remaining noise photons.

[0184] Specifically, in each vertical layer, in combination with the area of the layer and the number of photons after preliminary screening, a suitable connected radius and minimum neighborhood point number are determined, the connected radius is used to define the spatial neighborhood relationship between photons, and the minimum neighborhood point number is used to judge whether a photon belongs to a core point of a cluster. In this embodiment, the selection of these two parameters can be adjusted according to the specific characteristics of the vertical layer to adapt to photons data of different densities and distributions. By using the determined connected radius as the neighborhood radius and in combination with the minimum neighborhood point number, the photons in each vertical layer are processed by clustering. The purpose of clustering is to divide the photons into different clusters, in which high-density photons are aggregated together to form signal clusters, and low-density photons can be regarded as noise. In this process, those photons that do not meet the minimum neighborhood point number requirement are identified as isolated points, which are considered as noise photons. After clustering, the photons corresponding to the identified isolated points are determined as the remaining noise photons. These photons are sparse in spatial distribution and do not meet the aggregation characteristics of signal photons, so they are removed. In this way, the noise photons in the preliminary screening point cloud data can be further removed, thereby improving the purity and quality of the data.

[0185] In this optional embodiment, through clustering processing, noise photons that may still exist after preliminary screening can be identified and removed, further improving the purity and quality of the data. The connected radius and the minimum neighborhood point number are dynamically adjusted according to the area of each vertical layer and the number of photons, so that the clustering processing can adapt to data of different densities and distributions, improving the adaptability and robustness of the algorithm. Through clustering processing, signal photons and noise photons can be effectively distinguished, and the integrity of signal photons is preserved, providing more accurate data support for subsequent analysis and application. Clustering processing can more accurately identify noise photons, especially in complex environments, this method can effectively remove noise interference, improving the accuracy and reliability of data processing.

[0186] Optionally, the rejecting operation on the residual noise photons obtains the denoised point cloud data of the target region, and the rejecting operation on the residual noise photons comprises:

[0187] rejecting the residual noise photons in each of the vertical layers to obtain a denoised vertical layer;

[0188] generating the denoised point cloud data according to the photons in all the denoised vertical layers.

[0189] Specifically, based on the clustering processing result, the points determined as isolated points (i.e. noise photons) are rejected from the data. After the rejecting operation, the data in each vertical layer is more pure and mainly contains signal photons, thereby obtaining a denoised vertical layer. The photon points in all the denoised vertical layers are combined to form the denoised point cloud data of the entire target region. By recombining the data of the vertical layers after noise rejection processing, a complete point cloud data set without noise photons is obtained, which more accurately reflects the real structure and characteristics of the target region and provides high-quality data support for subsequent applications (such as marine remote sensing).

[0190] Finally, based on the denoised point cloud data, layering is performed, wherein the demarcation between the water surface and the water bottom specifically comprises: estimating the probability density distribution of the elevation values in the denoised laser radar point cloud data by kernel density estimation:

[0191] ;

[0192] wherein, represents the probability density estimation value at the elevation . represents the total number of photon points, represents a bandwidth parameter controlling the smoothness, represents a kernel function, is the name of the kernel function, and a Gaussian kernel is usually selected, is the elevation value of the current estimation point, represents the elevation value of the i th photon point, represents the sum of all photon points, i.e. from the first photon point to the th photon point.

[0193] By kernel density estimation, the probability density curve of the elevation distribution is obtained, and the maximum value of the probability density curve is found, i.e.

[0194] ;

[0195] wherein, This indicates the search for a function that... Reaching the maximum value z Value operations, It is elevation z The probability density estimate at point , argmax is used to find the value corresponding to this maximum value. z Value, i.e., boundary depth It can determine the boundary depth between the water surface and underwater. 。 like Figure 4 As shown, the first curve includes the probability density distribution curve of the denoised photon elevation (z-axis), and the second is the water surface and underwater boundary depth (z_split) determined based on this curve. This embodiment uses the kernel density estimation (KDE) method, with a Gaussian kernel as the kernel function, to calculate the probability density of photons at different elevations, generating a smooth density distribution curve (the horizontal axis in the figure represents photon elevation, and the vertical axis represents probability density). The maximum value point of the probability density curve is found, and the elevation corresponding to this point is the boundary depth z_split. The two sides of this point correspond to surface signal photons (high-density concentration area) and underwater photons (relatively low density area), respectively, providing a quantitative basis for subsequent extraction of surface and underwater signals.

[0196] The results of photon extraction of signals from the water surface and bottom are as follows: Figure 5 As shown, the distinction between surface and bottom water signal photons is based on the dividing depth. The photon data is divided into water surface signal photons. underwater signal photons ， The specific formula is as follows:

[0197] .

[0198] In this optional embodiment, by removing residual noise photons from each vertical layer, a denoised vertical layer is obtained, resulting in cleaner point cloud data and preservation of signal photon integrity. This significantly improves data quality and provides a more accurate foundation for subsequent analysis and applications. The denoised point cloud data reduces interference from noise photons, making the data more reliable. The noise-removed point cloud data more clearly reflects the true situation of the target area, making the data more usable in various application scenarios. For example, in marine environmental monitoring, the denoised data can more accurately reflect underwater topography and ecological characteristics.

[0199] This invention also provides a noise identification and rejection system, comprising:

[0200] An acquisition unit is used to acquire point cloud data of a target area and perform standardization processing on the point cloud data to obtain standardized point cloud data of the target area.

[0201] The dividing unit is configured to determine a layering parameter according to a vertical direction distribution feature of the standardized point cloud data, and divide the standardized point cloud data into a plurality of vertical layers according to the layering parameter.

[0202] The noise intensity analysis unit is configured to determine a noise intensity of each of the vertical layers according to a photon number of the standardized point cloud data in each of the vertical layers.

[0203] The noise removal unit is configured to perform a significance test on each of the vertical layers according to the noise intensity, obtain noise photons of the standardized point cloud data in each of the vertical layers, and perform a removal operation on the noise photons to obtain preliminary screening point cloud data.

[0204] The noise removal unit is configured to perform a significance test on each of the vertical layers according to the noise intensity, obtain noise photons of the standardized point cloud data in each of the vertical layers, and perform a removal operation on the noise photons to obtain preliminary screening point cloud data.

[0205] The noise removal unit is configured to perform a significance test on each of the vertical layers according to the noise intensity, obtain noise photons of the standardized point cloud data in each of the vertical layers, and perform a removal operation on the noise photons to obtain preliminary screening point cloud data.

[0206] Although the present application is disclosed as above, the protection scope of the present application is not limited to this. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application, and these changes and modifications shall fall within the protection scope of the present application.

Claims

1. A method for noise identification and removal, characterized in that, include: Obtain point cloud data of the target area and perform standardization processing on the point cloud data to obtain standardized point cloud data of the target area; Based on the vertical distribution characteristics of the standardized point cloud data, the layering parameters are determined, and the standardized point cloud data is divided into multiple vertical layers according to the layering parameters. The noise intensity of the vertical layer is determined based on the number of photons in each vertical layer using the standardized point cloud data. Based on the noise intensity, a saliency test is performed on each vertical layer to obtain the noise photons in each vertical layer of the standardized point cloud data, and the noise photons are removed to obtain the initial screened point cloud data; specifically, this includes: determining the expected number of neighbors within a preset neighborhood radius in each vertical layer based on the noise intensity of each vertical layer; The neighborhood radius of the vertical layer is determined based on the expected number of neighbors; For each photon in the vertical layer, the number of observed photons within the neighborhood radius of each photon is obtained; Based on the observed photon count, determine the noise probability of each photon; The noise photons in the vertical layer are determined based on the noise probability. The noise photons in each of the vertical layers are removed to obtain the initial screening point cloud data; Based on the area of ​​the vertical layer and the number of photons in each vertical layer of the initial screening point cloud data, cluster noise identification is performed to obtain the remaining noise photons in each vertical layer of the initial screening point cloud data. The remaining noise photons are then removed to obtain the denoised point cloud data of the target region.

2. The noise identification and rejection method according to claim 1, characterized in that, The standardization process of the point cloud data to obtain standardized point cloud data of the target region includes: Obtain the three-dimensional coordinates of each photon in the point cloud data; Based on the three-dimensional coordinates of the photon, determine the median of the X-coordinate, median of the Y-coordinate, and median of the Z-coordinate of the point cloud data; Based on the median of the X-coordinate, the median of the Y-coordinate, and the median of the Z-coordinate, determine the absolute deviation of the median of the X-coordinate, the absolute deviation of the median of the Y-coordinate, and the absolute deviation of the median of the Z-coordinate of the point cloud data; The three-dimensional coordinates of each photon are standardized based on the absolute deviation of the median of the X-coordinate, the absolute deviation of the median of the Y-coordinate, and the absolute deviation of the median of the Z-coordinate to obtain the standardized point cloud data.

3. The noise identification and rejection method according to claim 1, characterized in that, The step of determining the layering parameters based on the vertical distribution characteristics of the standardized point cloud data includes: Based on the vertical coordinates of all photons in the standardized point cloud data, the interquartile range of the standardized point cloud data is determined. Based on the total number of photons in the standardized point cloud data and the interquartile range, the slice thickness of the standardized point cloud data is obtained. Based on the maximum and minimum coordinates of the photons in the vertical direction in the standardized point cloud data, and combined with the slice thickness, the total number of slices is obtained; The total number of slices and the slice thickness are used as the layering parameters.

4. The noise identification and rejection method according to claim 3, characterized in that, The step of dividing the standardized point cloud data into multiple vertical layers according to the layering parameters includes: The range of coordinate values ​​in the vertical direction of the standardized point cloud data is determined based on the maximum and minimum coordinate values ​​of the photons in the vertical direction. Using the slice thickness as an interval, starting from the minimum value of the coordinate range in the vertical direction, several consecutive vertical intervals are sequentially divided towards the maximum value of the coordinates, wherein the number of vertical intervals is the same as the total number of slices. Each photon in the standardized point cloud data is traversed, and based on the coordinate value of each photon in the vertical direction, it is assigned to the corresponding vertical interval, and the set of all photons contained in each vertical interval is taken as a vertical layer.

5. The noise identification and rejection method according to claim 1, characterized in that, The step of determining the noise intensity of a vertical layer based on the number of photons in each vertical layer using the standardized point cloud data includes: Obtain the number of photons in the standardized point cloud data within each vertical layer, and determine the projection area of ​​each vertical layer on the horizontal plane; Based on the projection area, determine the projection area of ​​the projection area, and use the projection area as the area of ​​the vertical layer; The noise intensity of the vertical layer is obtained by calculating the ratio of the number of photons in each vertical layer to the area of ​​the vertical layer.

6. The noise identification and rejection method according to claim 1, characterized in that, Determining the noise photon in the vertical layer based on the noise probability includes: Based on the noise probability and the preset saliency threshold, it is determined whether the photon in the vertical layer is the noise photon; Wherein, when the noise probability of the photon is greater than or equal to the preset salience threshold, the photon is determined to belong to the noise photon; When the noise probability of the photon is less than the preset saliency threshold, the photon is determined not to be a noise photon.

7. The noise identification and rejection method according to claim 1, characterized in that, The step of performing cluster noise identification based on the area of ​​the vertical layer and the number of photons in each vertical layer of the initial screening point cloud data to obtain the remaining noise photons in each vertical layer of the initial screening point cloud data includes: Based on the area of ​​the region in each vertical layer and the number of photons in the vertical layer from the initial screening point cloud data, the connectivity radius and minimum number of neighboring points of the vertical layer are determined. Using the connectivity radius as the neighborhood radius and combining it with the minimum number of neighborhood points, the photons in the vertical layer are clustered to obtain isolated points in the vertical layer. The photon corresponding to the isolated point is taken as the remaining noise photon.

8. The noise identification and rejection method according to claim 1, characterized in that, The step of removing the remaining noise photons to obtain the denoised point cloud data of the target region includes: The remaining noise photons in each of the vertical layers are removed to obtain the denoised vertical layers; The denoised point cloud data is generated based on the photons in all the denoised vertical layers.

9. A noise identification and rejection system, characterized in that, include: An acquisition unit is used to acquire point cloud data of a target area and perform standardization processing on the point cloud data to obtain standardized point cloud data of the target area. A partitioning unit is used to determine the layering parameters based on the vertical distribution characteristics of the standardized point cloud data, and to partition the standardized point cloud data into multiple vertical layers based on the layering parameters. The noise intensity analysis unit is used to determine the noise intensity of the vertical layer based on the number of photons in each vertical layer of the standardized point cloud data. A noise removal unit is used to perform a saliency test on each vertical layer based on the noise intensity to obtain noisy photons in each vertical layer of the standardized point cloud data, and to remove the noisy photons to obtain preliminary screened point cloud data. Specifically, this includes: determining the expected number of neighbors within a preset neighborhood radius in each vertical layer based on the noise intensity of each vertical layer; determining the neighborhood radius of the vertical layer based on the expected number of neighbors; traversing each photon in the vertical layer to obtain the number of observed photons for each photon within the neighborhood radius; determining the noise probability of each photon based on the number of observed photons; determining the noisy photons in the vertical layer based on the noise probability; and removing the noisy photons in each vertical layer to obtain the preliminary screened point cloud data. The denoising processing unit is used to perform cluster noise identification based on the area of ​​the vertical layer and the number of photons in each vertical layer of the initial screening point cloud data, to obtain the remaining noise photons in each vertical layer of the initial screening point cloud data, and to remove the remaining noise photons to obtain the denoised point cloud data of the target area.

Images