A photon data denoising and classification method and system

Abstract

The present invention relates to a photon data denoising and classification method and system, comprising: acquiring target photon data; the target photon data is collected by a photon counting laser radar; photon data within a preset elevation threshold range is extracted multiple times, photon denoising is performed on the extracted photon data using a horizontal ellipse LOF model, and the photon data is further denoised using a rotating ellipse LOF model, thereby achieving accurate extraction and classification of the photon data; terrain slope is used as the rotation angle of the rotating ellipse LOF model, the influence of complex terrain on photon distribution is taken into account, and the accuracy of photon denoising is improved without significantly increasing the amount of calculation.

Description

Technical Field

[0001] The present invention relates to the field of satellite-borne laser radar technology, and in particular to a photon data denoising and classification method and system. Background Art

[0002] NASA launched the ICESat-2 satellite in 2018, one of its main purposes is to measure the height of vegetation canopies as a basis for estimating large-scale biomass and biomass changes. The ATLAS (Advanced Topographic Laser Altimeter System) photon counting lidar system carried by ICESat-2 can detect signals at the single photon level by emitting low-energy, high-frequency laser pulses, obtain three-dimensional spatial data with smaller spots and higher density, and achieve accurate characterization of the observed object. Compared with airborne lidar, satellite-borne lidar is easier to achieve large-scale observations, and its data range covers the entire world, providing important data support for forest remote sensing.

[0003] Since photon counting lidar emits low-energy weak signals, its signal receiver is sensitive to weak signals and is easily affected by noise (e.g. solar background noise, system noise, atmospheric scattering noise) during data collection. Therefore, effectively denoising and classifying photon data is an important prerequisite for subsequent forest parameter inversion. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for photon data denoising and classification.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] The present invention provides a photon data denoising and classification method, comprising:

[0007] Step 1: Acquire target photon data; the target photon data is collected by a photon counting laser radar;

[0008] Step 2: extracting the target photon data whose elevation is within the first preset elevation threshold range to obtain photon data of a rough screening of the elevation histogram;

[0009] Step 3: Divide the photon data of the elevation histogram coarse screening into units in the along-track direction, extract the photon data of the elevation histogram coarse screening whose elevation in each divided unit is within the corresponding second preset elevation threshold range, and obtain the original data of the elevation coarse screening; each divided unit corresponds to a second preset elevation threshold range;

[0010] Step 4: Calculate the LOFE score of the raw data of each elevation rough screening using the horizontal ellipse LOF model;

[0011] Step 5: extracting photon data whose LOFE scores meet a preset LOFE threshold from the raw data of the elevation rough screening to obtain coarsely denoised signal photon data;

[0012] Step 6: Divide the roughly denoised signal photon data into sliding windows in the along-track direction to obtain a plurality of first sliding windows;

[0013] Step 7: extracting the coarsely denoised signal photon data with the smallest elevation in each of the first sliding windows to obtain coarsely classified ground photons;

[0014] Step 8: Divide the raw data of the elevation rough screening into intervals along the track direction to obtain a plurality of first divided intervals;

[0015] Step 9: extracting the raw data of the rough elevation screening whose elevation in each of the first divided intervals is within the range of the third preset elevation threshold, and obtaining the raw data of the fine elevation screening; the raw data of the fine elevation screening includes the roughly classified ground photons;

[0016] Step 10: Divide the original data of the elevation fine screening into intervals along the track direction to obtain a plurality of second divided intervals;

[0017] Step 11: Calculate the terrain slope of each first divided interval according to the along-track distance and elevation of the roughly classified ground photons in the original data of the elevation fine screening in each second divided interval;

[0018] Step 12: For each of the raw data of the elevation fine screening, the terrain slope corresponding to the second divided interval to which the raw data of the elevation fine screening belongs is used as the rotation angle of the rotating ellipse LOF model to calculate the LOFR score of the raw data of the elevation fine screening;

[0019] Step 13: extracting photon data whose LOFR scores meet a preset LOFR threshold from the raw data of the elevation fine screening to obtain finely denoised signal photon data;

[0020] Step 14: Classify the photons in each second sliding window in the along-track direction according to the elevation of each photon data in the finely denoised signal photon data in each second sliding window to obtain finely classified crown photons, ground photons and canopy photons.

[0021] Optionally, the step 9 specifically includes:

[0022] Calculating the average elevation and standard deviation elevation of the roughly classified ground photon data within each of the first divided intervals;

[0023] The third preset elevation threshold range of each first divided interval is calculated according to the average elevation and the standard deviation elevation, and the original data of the elevation coarse screening whose elevations in each first divided interval are within the third preset elevation threshold range are extracted to obtain the original data of the elevation fine screening.

[0024] Optionally, the step 2 specifically includes:

[0025] Dividing the target photon data from top to bottom in the elevation direction to obtain a plurality of first division units;

[0026] Calculate the number of photons in a front-end preset number of first division units, and calculate a first average value and a first standard deviation of the number of photons in the front-end preset number of first division units;

[0027] Calculating the number of photons in a preset number of first division units at the end, and calculating a second average value and a second standard deviation of the number of photons in the preset number of first division units at the end;

[0028] Calculating a unit noise level based on the first mean value, the first standard deviation, the second mean value, and the second standard deviation;

[0029] The elevation of the first division unit where the first noise level exceeds the unit noise level is used as an upper bound; the elevation of the first division unit where the last noise level exceeds the unit noise level is used as a lower bound; the upper bound and the lower bound are used to determine the first preset elevation threshold range;

[0030] The target photon data whose elevation is within the first preset elevation threshold range is extracted to obtain photon data of a coarse screening of the elevation histogram.

[0031] Optionally, the unit noise level calculation formula is:

[0032]

[0033] Among them, Noise represents the unit noise level, Noise High Indicates the average noise at the beginning of the signal, Noise Low represents the signal end average noise, N1 represents the first average value, σ1 represents the first standard deviation, N2 represents the second average value, σ2 represents the second standard deviation, and α is the noise parameter.

[0034] Optionally, the step 3 specifically includes:

[0035] Dividing the photon data of the coarse screening of the elevation histogram into a plurality of second division units in the along-track direction;

[0036] Calculate the average elevation, 25% quantile elevation and 75% quantile elevation of the photon elevation in each of the second divided units;

[0037] The second preset elevation threshold range corresponding to each second division unit is set according to the average elevation, 25% quantile elevation and 75% quantile elevation of each second division unit, and the coarse-screened photon data of the elevation histogram in each second division unit whose elevation is within the corresponding second preset elevation threshold range is extracted to obtain the coarse-screened original data of the elevation.

[0038] Optionally, step 4 specifically includes:

[0039] For any photon P, the LOFE score of the photon P is calculated by the following steps:

[0040] Calculate the reachable distance between any two photons O and P, where photon O is a photon in the horizontal ellipse K neighborhood of photon P:

[0041] Reach_distances K (P, O) = max{K-distances K (O), d(P, O)}

[0042] Among them, Reach_distances K (P, O) represents the reachable distance between O and P, K-distances K (O) represents the KNN distance of the horizontal elliptical K neighborhood of photon O, and d(P, O) represents the distance between photons O and P;

[0043] According to the reachable distance, obtaining the local reachable density of the photon P;

[0044] The LOFE score of the photon P is obtained according to the local reachable density and the average local reachable density of photons in the horizontal ellipse K neighborhood of the photon P.

[0045] Optionally, the KNN distance calculation formula of the horizontal ellipse K neighborhood is:

[0046]

[0047] Among them, P and O represent two different photons, Δx represents the distance between P and O photons in the along-track direction, Δy represents the difference between P and O photons in the elevation direction, and a and b represent the major and minor axis parameters of the horizontal elliptical search area, respectively.

[0048] Optionally, the terrain slope calculation formula in step 11 is:

[0049]

[0050] Among them, α i represents the terrain slope of the ith interval, h p represents the elevation of the coarsely classified ground photon p, h q represents the elevation of the coarsely classified ground photon q, x p represents the along-track distance of the coarsely classified ground photon p, x q Represents the along-track distance of the coarsely classified ground photon q.

[0051] Optionally, the KNN distance calculation formula of the rotating ellipse LOF model search area is:

[0052]

[0053] Δx=cosα i ×(x p -x q )+sinα i ×(h p -h q )

[0054] Δy=sinα i ×(x p -x q )-cosα i ×(h p -h q )

[0055] in, represents the KNN distance of the search area of ​​the rotating ellipse LOF model, α i represents the terrain slope of the i-th interval, Δx represents the distance between p and q photons along the track, Δy represents the difference in elevation between p and q photons, and h p represents the elevation of photon p, h q represents the elevation of photon q, x p Represents photon p The distance along the track, x q represents the distance along the track of photon q.

[0056] The present invention provides a photon data denoising and classification system, comprising:

[0057] A target photon data acquisition module is used to acquire target photon data; the target photon data is collected by a photon counting laser radar;

[0058] The photon data acquisition module of the coarse screening of the elevation histogram is used to extract the target photon data whose elevation is within the first preset elevation threshold range to obtain the photon data of the coarse screening of the elevation histogram;

[0059] The raw data acquisition module of the elevation rough screening is used to divide the photon data of the elevation histogram rough screening into units in the along-track direction, extract the photon data of the elevation histogram rough screening whose elevation in each divided unit is within the corresponding second preset elevation threshold range, and obtain the raw data of the elevation rough screening; each of the divided units corresponds to a second preset elevation threshold range;

[0060] A LOFE score calculation module, for calculating the LOFE score of each raw data of the elevation rough screening using a horizontal ellipse LOF model;

[0061] A coarse denoising signal photon data acquisition module is used to extract the photon data whose LOFE score meets the preset LOFE threshold in the raw data of the elevation coarse screening, and obtain the coarse denoising signal photon data;

[0062] A first division module, used for performing sliding window division on the coarsely denoised signal photon data in the along-track direction to obtain a plurality of first sliding windows;

[0063] A coarsely classified ground photon acquisition module is used to extract the coarsely denoised signal photon data with the smallest elevation in each of the first sliding windows to obtain coarsely classified ground photons;

[0064] A second division module is used to divide the raw data of the elevation rough screening into intervals in the along-track direction to obtain a plurality of first division intervals;

[0065] The raw data acquisition module of the elevation fine screening is used to extract the raw data of the elevation coarse screening whose elevation in each of the first divided intervals is within the range of the third preset elevation threshold value to obtain the raw data of the elevation fine screening; the raw data of the elevation fine screening includes the coarsely classified ground photons;

[0066] A third division module is used to divide the original data of the elevation fine screening into intervals in the along-track direction to obtain a plurality of second divided intervals;

[0067] A terrain slope calculation module, used for calculating the terrain slope of each first divided interval according to the along-track distance and elevation of the coarsely classified ground photons in the original data of the elevation fine screening in each second divided interval;

[0068] A LOFR score calculation module, for each of the raw data of the elevation fine screening, using the terrain slope corresponding to the second divided interval to which the raw data of the elevation fine screening belongs as the rotation angle of the rotating ellipse LOF model to calculate the LOFR score of the raw data of the elevation fine screening;

[0069] A signal photon data acquisition module for fine denoising is used to extract photon data whose LOFR score meets a preset LOFR threshold in the original data of the elevation fine screening to obtain signal photon data for fine denoising;

[0070] The classification module is used to classify the photons in each second sliding window in the along-track direction according to the elevation size of each photon data in the finely denoised signal photon data in each second sliding window, so as to obtain finely classified crown photons, ground photons and canopy photons.

[0071] According to the specific embodiment provided by the present invention, the present invention discloses the following technical effects: the present invention provides a photon data denoising and classification method and system, including: acquiring target photon data; the target photon data is collected by a photon counting laser radar; extracting the target photon data whose elevation is within a first preset elevation threshold range to obtain photon data of a rough screen of an elevation histogram; dividing the photon data of the rough screen of the elevation histogram into units in the along-track direction, extracting the photon data of the rough screen of the elevation histogram whose elevation in each divided unit is within a corresponding second preset elevation threshold range to obtain the original rough screen of the elevation. data; each division unit corresponds to a second preset elevation threshold range; the LOFE score of each raw data of the elevation rough screening is calculated using the horizontal ellipse LOF model; the photon data whose LOFE score meets the preset LOFE threshold in the raw data of the elevation rough screening are extracted to obtain the coarse denoised signal photon data; the coarse denoised signal photon data is divided into sliding windows in the along-track direction to obtain a plurality of first sliding windows; the coarse denoised signal photon data with the smallest elevation in each of the first sliding windows is extracted to obtain the coarsely classified ground photons; the raw data of the elevation rough screening are divided into sliding windows in the along-track direction to obtain a plurality of first sliding windows; the coarse denoised signal photon data with the smallest elevation in each of the first sliding windows is extracted to obtain the coarsely classified ground photons; Divide the interval upward to obtain a plurality of first divided intervals; extract the raw data of the rough elevation screening in each of the first divided intervals whose elevation is within the range of the third preset elevation threshold to obtain the raw data of the fine elevation screening; the raw data of the fine elevation screening includes the coarsely classified ground photons; divide the raw data of the fine elevation screening in the along-track direction to obtain a plurality of second divided intervals; calculate the terrain slope of each of the first divided intervals according to the along-track distance and elevation of the coarsely classified ground photons in the raw data of the fine elevation screening in each of the second divided intervals; for each of the elevation The original data of the fine screening is used, and the terrain slope corresponding to the second divided interval to which the original data of the elevation fine screening belongs is used as the rotation angle of the rotating ellipse LOF model to calculate the LOFR score of the original data of the elevation fine screening; the photon data whose LOFR score meets the preset LOFR threshold in the original data of the elevation fine screening is extracted to obtain the finely denoised signal photon data; the photons in each second sliding window are classified according to the elevation size of each photon data in the finely denoised signal photon data in each second sliding window in the along-track direction to obtain finely classified crown photons, ground photons and canopy photons. The present invention extracts photon data within the preset elevation threshold range for multiple times, performs photon denoising on the extracted photon data using the horizontal ellipse LOF model, and further denoises the photon data using the rotating ellipse LOF model, thereby realizing accurate extraction and classification of photon data; the terrain slope is used as the rotation angle of the rotating ellipse LOF model, the influence of complex terrain on photon distribution is taken into account, and the accuracy of photon denoising is improved without significantly increasing the amount of calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0073] Figure 1 A flow chart of the photon data denoising and classification method provided in Example 1 of the present invention;

[0074] Figure 2 The photon data classification effect diagram of the Baotianman test area provided in Example 1 of the present invention;

[0075] Figure 3 A partial enlarged view of the photon data classification effect of the Baotianman test area provided in Example 1 of the present invention;

[0076] Figure 4 The photon data classification effect diagram of the Saihanba test area provided in Example 1 of the present invention;

[0077] Figure 5 A local enlarged view of the photon data classification effect of the Saihanba test area provided in Example 1 of the present invention;

[0078] Figure 6 This is a block diagram of the photon data denoising and classification system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0079] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0080] The purpose of the present invention is to provide a method and system for photon data denoising and classification.

[0081] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0082] Example 1

[0083] This embodiment provides a photon data denoising and classification method, see Figure 1 , the method comprising:

[0084] Step 1: Obtain target photon data; the target photon data is collected by a photon counting laser radar.

[0085] This embodiment uses ATL03 photon data in the ATLAS data as target photon data.

[0086] Step 2: extracting the target photon data whose elevation is within the first preset elevation threshold range to obtain photon data of a rough screening of the elevation histogram;

[0087] Step 3: Divide the photon data of the elevation histogram coarse screening into units in the along-track direction, extract the photon data of the elevation histogram coarse screening whose elevation in each divided unit is within the corresponding second preset elevation threshold range, and obtain the original data of the elevation coarse screening; each divided unit corresponds to a second preset elevation threshold range;

[0088] Step 4: Calculate the LOFE score of the raw data of each elevation rough screening using the horizontal ellipse LOF model;

[0089] Step 5: extracting photon data whose LOFE scores meet a preset LOFE threshold from the raw data of the elevation rough screening to obtain coarsely denoised signal photon data;

[0090] Step 6: Divide the roughly denoised signal photon data into sliding windows in the along-track direction to obtain a plurality of first sliding windows;

[0091] Step 7: extracting the coarsely denoised signal photon data with the smallest elevation in each of the first sliding windows to obtain coarsely classified ground photons;

[0092] Step 8: Divide the raw data of the elevation rough screening into intervals along the track direction to obtain a plurality of first divided intervals;

[0093] Step 9: extracting the raw data of the rough elevation screening whose elevation in each of the first divided intervals is within the range of the third preset elevation threshold, and obtaining the raw data of the fine elevation screening; the raw data of the fine elevation screening includes the roughly classified ground photons;

[0094] Step 10: Divide the original data of the elevation fine screening into intervals along the track direction to obtain a plurality of second divided intervals;

[0095] Step 11: Calculate the terrain slope of each first divided interval according to the along-track distance and elevation of the roughly classified ground photons in the original data of the elevation fine screening in each second divided interval;

[0096] Step 12: For each of the raw data of the elevation fine screening, the terrain slope corresponding to the second divided interval to which the raw data of the elevation fine screening belongs is used as the rotation angle of the rotating ellipse LOF model to calculate the LOFR score of the raw data of the elevation fine screening;

[0097] Step 13: extracting photon data whose LOFR scores meet a preset LOFR threshold from the raw data of the elevation fine screening to obtain finely denoised signal photon data;

[0098] Step 14: Classify the photons in each second sliding window in the along-track direction according to the elevation of each photon data in the finely denoised signal photon data in each second sliding window to obtain finely classified crown photons, ground photons and canopy photons.

[0099] In this embodiment, step 2 may include the following steps:

[0100] Dividing the target photon data from top to bottom in the elevation direction to obtain a plurality of first division units;

[0101] Calculate the number of photons in a front-end preset number of first division units, and calculate a first average value and a first standard deviation of the number of photons in the front-end preset number of first division units;

[0102] Calculating the number of photons in a preset number of first division units at the end, and calculating a second average value and a second standard deviation of the number of photons in the preset number of first division units at the end;

[0103] Calculating a unit noise level based on the first mean value, the first standard deviation, the second mean value, and the second standard deviation;

[0104] The elevation of the first division unit where the first noise level exceeds the unit noise level is used as an upper bound; the elevation of the first division unit where the last noise level exceeds the unit noise level is used as a lower bound; the upper bound and the lower bound are used to determine the first preset elevation threshold range;

[0105] The target photon data whose elevation is within the first preset elevation threshold range is extracted to obtain photon data of a coarse screening of the elevation histogram.

[0106] The following is a specific introduction to step 2 by way of example:

[0107] Project the ATL03 photon data onto the along-track profile and preliminarily screen the photon data. First, convert the ICESat-2 photon data into along-track distance-elevation coordinates based on the time information and longitude and latitude information, and create a histogram in the elevation direction to count the number of photons in each unit. Each unit has the same length, generally 1 meter. Count the number of photons in the front 50 first division units and the last 50 first division units from high to low along the elevation direction, and calculate the first average value N1 and the first standard deviation σ1 of the number of photons, as well as the second average value N2 and the second standard deviation and standard deviation σ2 of the number of photons in the last 50 first division units. The average value and standard deviation calculation formula are:

[0108]

[0109]

[0110]

[0111]

[0112] Where i represents the i-th first division unit, P i Represents the number of photons in the i-th first division unit.

[0113] Based on the first mean value, the first standard deviation, the second mean value and the second standard deviation obtained above, the unit noise level is calculated, and the calculation formula is:

[0114]

[0115] Among them, Noise represents the unit noise level, Noise High Indicates the average noise at the beginning of the signal, Noise Low represents the signal end average noise, N1 represents the first average value, σ1 represents the first standard deviation, N2 represents the second average value, σ2 represents the second standard deviation, α is the noise parameter, 1≤α≤3.

[0116] According to the unit noise level obtained above, the number of photons in each unit is compared with the noise level Noise from top to bottom along the elevation direction to determine the starting first division unit and the ending first division unit, thereby determining the first preset elevation threshold range, and removing the target photon data whose elevation exceeds the upper limit of the first preset elevation threshold range or is less than the lower limit of the first preset elevation threshold range to obtain the photon data of the coarse screening of the elevation histogram.

[0117] In this embodiment, step 3 may include the following steps:

[0118] The photon data obtained by coarse screening of the elevation histogram is divided into a plurality of second division units in the along-track direction.

[0119] The average elevation, the 25% quantile elevation and the 75% quantile elevation of the photon elevation in each of the second divided units are calculated.

[0120] According to the average height Height of each of the second division units mean 、25% quantile elevationHeight 25 and the 75% quantile elevation Height 75 A second preset elevation threshold range is set corresponding to each of the second division units, and the photon data of the elevation histogram coarse screening in each second division unit whose elevation is within the corresponding second preset elevation threshold range is extracted to obtain the original data of the elevation coarse screening.

[0121] The calculation formula for the second preset elevation threshold range is:

[0122]

[0123] Among them, Hei g ht top is the upper limit of the second preset elevation threshold range, Hei g ht bottom The lower limit of the second preset elevation threshold range.

[0124] In this embodiment, steps 4 and 5 are the stages of coarse denoising of the raw data obtained from the rough elevation screening in step 3, and the coarse denoised signal photon data and the coarsely classified surface photon data are extracted using the horizontal ellipse LOF model (Local Outlier Factor with Ellipse search area, LOFE).

[0125] Taking photon P as an example, first calculate the LOFE score of photon P. For any photon P, the LOFE score of the photon P can be calculated by the following steps:

[0126] Calculate the reachable distance between any two photons O and P, where photon O is a photon in the horizontal ellipse K neighborhood of photon P:

[0127] Reach_distances K (P, O) = max{K-distances K (O), d(P, O)}

[0128] Among them, Reach_distances K (P, O) represents the reachable distance between O and P, K-distances K(O) represents the KNN distance of the horizontal ellipse K neighborhood of photon O, and d(P, O) represents the distance between two photons O and P. The KNN distance calculation formula of the horizontal ellipse K neighborhood is:

[0129]

[0130] Among them, P and O represent two different photons, Δx represents the distance between P and O photons in the along-track direction, Δy represents the difference between P and O photons in the elevation direction, and a and b represent the major and minor axis parameters of the horizontal elliptical search area, respectively.

[0131] According to the reachable distance, the local reachable density of the photon P is obtained, and the local reachable density of the photon P is expressed as:

[0132]

[0133] Among them, Lrd k (P) represents the local reachable density of photon P, N k (P) represents the photons in the K neighborhood of p, |N k (P)| is the number of photons in the K-neighborhood of p.

[0134] The LOFE score of the photon P is obtained based on the local reachable density obtained above and the average local reachable density of photons in the horizontal ellipse K neighborhood of the photon P.

[0135]

[0136] Where LOFE(P) is the LOFE score of photon P, represents the average local reachable density of photons in the horizontal ellipse K neighborhood of photon P, N k (P) represents the photons in the K neighborhood of P.

[0137] After calculating the LOFE score of each raw data of the elevation rough screening, the raw data of the elevation rough screening is screened based on its LOFE score. For example, the photon data whose LOFE score meets the preset LOFE threshold value in the raw data of the elevation rough screening is extracted, and the photons less than the preset LOFE threshold value T are classified as signal photons to obtain the roughly denoised signal photon data, and the photons whose LOFE score is greater than the preset LOFE threshold value T are classified as noise.

[0138] In this embodiment, after obtaining the coarse denoised signal photon data, equal-length, partially overlapping first sliding windows are established for the coarse denoised signal photon data in the along-track direction, and the signal photons with the smallest elevation in each first sliding window are classified into coarsely classified ground photons.

[0139] In this embodiment, step 9 specifically includes:

[0140] Calculating the average elevation and standard deviation elevation of the roughly classified ground photons in each of the first divided intervals;

[0141] The third preset elevation threshold range of each first divided interval is calculated according to the average elevation and the standard deviation elevation, and the coarsely denoised signal photon data with elevation within the third preset elevation threshold range in each first divided interval is extracted to obtain the original data of elevation fine screening, specifically:

[0142] In some cases, the local density of noise may be close to that of signal photons. Therefore, such photons are removed by limiting the range in elevation. The signal photons are divided into a number of first division intervals along the track, the average elevation and standard deviation elevation of the ground photons in each first division interval are calculated, and a third preset elevation threshold range of the signal photons is set. The coarsely denoised signal photon data far from the ground is removed to obtain the original data of the elevation fine screening.

[0143] In this embodiment, steps 10-14 may be specifically as follows:

[0144] In the density-based photon denoising algorithm, the search area close to the real distribution of the signal has obvious advantages. Compared with the traditional circular domain search, the horizontal elliptical search area emphasizes the distribution characteristics of photons in the horizontal direction, and has achieved better performance, which is also the advantage of the LOFE algorithm. However, in complex terrain, the distribution of signal photons is affected by the terrain, the spatial distribution characteristics change, and the performance of the horizontal elliptical search area is unstable in different terrains. Therefore, how to adaptively improve the elliptical search area is an important research content of this type of method. The existing algorithms usually use the method of traversing angles to select the optimal rotation angle and then rotate the elliptical search area, that is, traverse all angles with equal lengths (3° or 5°), calculate the density features under the angle respectively, and finally select the angle with the maximum density feature as the final rotation angle. However, during the traversal process, the density calculation involved in each angle will bring a huge amount of calculation, reducing the efficiency of the algorithm. Therefore, this embodiment selects the slope calculated based on the ground photon in the coarse denoising as the rotation angle of the search area, thereby improving the accuracy of photon denoising without significantly increasing the amount of calculation.

[0145] Based on the original data of elevation fine screening obtained above, a rotating ellipse LOF model (LOFR) is established according to the slope to improve the adaptability of the LOFE algorithm to terrain.

[0146] The original data of the elevation fine screening is divided into intervals along the track direction to obtain a number of second divided intervals, and the terrain slope of each second divided interval is calculated according to the along-track distance and elevation of the original data of the elevation fine screening in each second divided interval. The terrain slope calculation formula is:

[0147]

[0148] Among them, α i represents the terrain slope of the ith interval, h p represents the elevation of the coarsely classified ground photon p, h q represents the elevation of the coarsely classified ground photon q, x p represents the along-track distance of the coarsely classified ground photon p, x q represents the along-track distance of the coarsely classified ground photon q. p and q are the leftmost and rightmost coarsely classified ground photons in the i-th interval, respectively.

[0149] The terrain slope of each second divided interval obtained above is used as the rotation angle of the rotating ellipse LOF model to calculate the LOFR score of the original data of the elevation fine screening, wherein the KNN distance calculation formula of the rotating ellipse LOF model search area is:

[0150]

[0151] Δx=cosα i ×(x p -x q )+sinα i ×(h p -h q )

[0152] Δy=sinα i ×(x p -x q )-cosα i ×(h p -h q )

[0153] in, represents the KNN distance of the search area of ​​the rotating ellipse LOF model, α i represents the terrain slope of the i-th interval, Δx represents the distance between p and q photons along the track, Δy represents the difference in elevation between p and q photons, and h p represents the elevation of photon p, h q represents the elevation of photon q, x p represents the distance along the track of photon p, x q represents the distance along the track of photon q.

[0154] The LOFR score of the original data of each elevation fine-screening is calculated by using the KNN distance based on the rotating ellipse search area, and the photon data whose LOFR score meets the preset LOFR threshold in the original data of the elevation fine-screening is extracted. In this embodiment, a preset LOFR threshold T′ is set, and photons smaller than the preset LOFR threshold T′ are classified as signal photons to obtain finely denoised signal photon data, and photons with LOFR scores greater than the preset LOFR threshold T′ are classified as noise.

[0155] After the above-mentioned photon denoising process, a second sliding window of equal length and partial overlap is established in the along-track direction, and the photons with the maximum elevation and the photons with the minimum elevation in each second sliding window are classified into crown photons and ground photons, respectively, and the signal photons with elevations between the maximum elevation and the minimum elevation are classified as canopy photons.

[0156] Finally, all ICESat-2 photons are classified into ground photons, canopy photons, canopy top photons, and noise. This example uses Baotianman and Saihanba areas as test areas to demonstrate the actual denoising effect. The actual denoising effect is as follows: Figure 2 , 3 , 4, and 5 (ALS_Surface and ALS_Terrain are comparisons from airborne lidar data). Figure 2 This is the photon data classification effect diagram of the Baotianman test area; Figure 3 This is a partial enlarged picture of the photon data classification effect of the Baotianman test area; Figure 4 This is the photon data classification effect diagram of the Saihanba test area; Figure 5 This is a partial enlarged picture of the photon data classification effect in the Saihanba test area. Noise represents noise, Ground Photons represents ground photons, Canopy Photons represents canopy photons, and TOCPhotons represents canopy photons. Elevation represents the elevation of the photon, and Along-track Distance represents the distance along the track of the photon.

[0157] The present invention extracts photon data within a preset elevation threshold range multiple times, performs photon denoising on the extracted photon data using a horizontal ellipse LOF model, and further denoises the photon data using a rotating ellipse LOF model, thereby achieving accurate extraction and classification of photon data. By using the terrain slope as the rotation angle of the rotating ellipse LOF model, the rotation angle of the ellipse search area can be adaptively determined without introducing other terrain prior knowledge, thereby achieving accurate classification of ICESat-2 photon data. Compared with other current denoising algorithms, the rotating search area can improve the adaptability of the algorithm to complex terrain, and the rotation angle based on the terrain slope can determine the rotation angle without significantly increasing the computational complexity, which is suitable for large-scale ICESat-2 / ATLAS data processing.

[0158] Example 2

[0159] This embodiment provides a photon data denoising and classification system, see Figure 6 , the photon data denoising and classification system comprises:

[0160] The target photon data acquisition module T1 is used to acquire the target photon data; the target photon data is collected by a photon counting laser radar;

[0161] The photon data acquisition module T2 for rough screening of the elevation histogram is used to extract the target photon data whose elevation is within the first preset elevation threshold range to obtain the photon data for rough screening of the elevation histogram;

[0162] The raw data acquisition module T3 for rough elevation screening is used to divide the photon data of the rough elevation histogram into units in the along-track direction, extract the photon data of the rough elevation histogram in each divided unit whose elevation is within the corresponding second preset elevation threshold range, and obtain the raw data of the rough elevation screening; each of the divided units corresponds to a second preset elevation threshold range;

[0163] LOFE score calculation module T4, used for calculating the LOFE score of each raw data of the elevation rough screening by using the horizontal ellipse LOF model;

[0164] A coarse denoising signal photon data acquisition module T5 is used to extract the photon data whose LOFE score meets the preset LOFE threshold in the raw data of the elevation coarse screening, and obtain the coarse denoising signal photon data;

[0165] A first division module T6, used for performing sliding window division on the coarsely denoised signal photon data in the along-track direction to obtain a plurality of first sliding windows;

[0166] A coarsely classified ground photon acquisition module T7, used to extract the coarsely denoised signal photon data with the smallest elevation in each of the first sliding windows to obtain coarsely classified ground photons;

[0167] The second division module T8 is used to divide the raw data of the elevation rough screening into intervals along the track direction to obtain a plurality of first division intervals;

[0168] The raw data acquisition module T9 for elevation fine screening is used to extract the raw data of the elevation coarse screening in each of the first divided intervals whose elevation is within the range of the third preset elevation threshold, and obtain the raw data of the elevation fine screening; the raw data of the elevation fine screening includes the coarsely classified ground photons;

[0169] The third division module T10 is used to divide the original data of the elevation fine screening into intervals in the along-track direction to obtain a plurality of second divided intervals;

[0170] A terrain slope calculation module T11, used to calculate the terrain slope of each first divided interval according to the along-track distance and elevation of the coarsely classified ground photons in the original data of the elevation fine screening in each second divided interval;

[0171] The LOFR score calculation module T12 is used to calculate the LOFR score of each of the raw data of the elevation fine screening by using the terrain slope corresponding to the second divided interval to which the raw data of the elevation fine screening belongs as the rotation angle of the rotating ellipse LOF model;

[0172] A fine denoising signal photon data acquisition module T13 is used to extract the photon data whose LOFR score meets the preset LOFR threshold in the original data of the elevation fine screening, and obtain the fine denoising signal photon data;

[0173] The classification module T14 is used to classify the photons in each second sliding window in the along-track direction according to the elevation size of each photon data in the finely denoised signal photon data in each second sliding window, and obtain finely classified crown photons, ground photons and canopy photons.

[0174] As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0175] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A photon data denoising and classification method, characterized in that: include: Step 1: Acquire target photon data; the target photon data is collected by a photon counting laser radar; Step 2: extracting the target photon data whose elevation is within the first preset elevation threshold range to obtain photon data of a rough screening of the elevation histogram; Step 3: Divide the photon data of the elevation histogram coarse screening into units in the along-track direction, extract the photon data of the elevation histogram coarse screening whose elevation in each divided unit is within the corresponding second preset elevation threshold range, and obtain the original data of the elevation coarse screening; each divided unit corresponds to a second preset elevation threshold range; Step 4: Calculate the LOFE score of the raw data of each elevation rough screening using the horizontal ellipse LOF model; Step 5: extracting photon data whose LOFE scores meet a preset LOFE threshold from the raw data of the elevation rough screening to obtain coarsely denoised signal photon data; Step 6: Divide the roughly denoised signal photon data into sliding windows in the along-track direction to obtain a plurality of first sliding windows; Step 7: extracting the coarsely denoised signal photon data with the smallest elevation in each of the first sliding windows to obtain coarsely classified ground photons; Step 8: Divide the raw data of the elevation rough screening into intervals along the track direction to obtain a plurality of first divided intervals; Step 9: extracting the raw data of the rough elevation screening whose elevation in each of the first divided intervals is within the range of the third preset elevation threshold, and obtaining the raw data of the fine elevation screening; the raw data of the fine elevation screening includes the roughly classified ground photons; Step 10: Divide the original data of the elevation fine screening into intervals along the track direction to obtain a plurality of second divided intervals; Step 11: Calculate the terrain slope of each first divided interval according to the along-track distance and elevation of the roughly classified ground photons in the original data of the elevation fine screening in each second divided interval; Step 12: For each of the raw data of the elevation fine screening, the terrain slope corresponding to the second divided interval to which the raw data of the elevation fine screening belongs is used as the rotation angle of the rotating ellipse LOF model to calculate the LOFR score of the raw data of the elevation fine screening; Step 13: extracting photon data whose LOFR scores meet a preset LOFR threshold from the raw data of the elevation fine screening to obtain finely denoised signal photon data; Step 14: Classify the photons in each second sliding window in the along-track direction according to the elevation of each photon data in the finely denoised signal photon data in each second sliding window to obtain finely classified crown photons, ground photons and canopy photons.

2. The photon data denoising and classification method according to claim 1, characterized in that: The step 9 specifically includes: Calculating the average elevation and standard deviation elevation of the coarsely classified ground photon data within each of the first divided intervals; The third preset elevation threshold range of each first divided interval is calculated according to the average elevation and the standard deviation elevation, and the coarsely denoised signal photon data with elevations within the third preset elevation threshold range in each first divided interval is extracted to obtain the original data of elevation fine screening.

3. The photon data denoising and classification method according to claim 1, characterized in that: The step 2 specifically includes: Dividing the target photon data from top to bottom in the elevation direction to obtain a plurality of first division units; Calculate the number of photons in a front-end preset number of first division units, and calculate a first average value and a first standard deviation of the number of photons in the front-end preset number of first division units; Calculating the number of photons in a terminal preset number of first division units, and calculating a second average value and a second standard deviation of the number of photons in the terminal preset number of first division units; Calculating a unit noise level based on the first mean value, the first standard deviation, the second mean value, and the second standard deviation; The elevation of the first division unit where the first noise level exceeds the unit noise level is used as an upper bound; the elevation of the first division unit where the last noise level exceeds the unit noise level is used as a lower bound; the upper bound and the lower bound are used to determine the first preset elevation threshold range; The target photon data whose elevation is within the first preset elevation threshold range is extracted to obtain photon data of a coarse screening of the elevation histogram.

4. The photon data denoising and classification method according to claim 3, characterized in that: The unit noise level calculation formula is: Among them, Noise represents the unit noise level, Noise High Indicates the average noise at the beginning of the signal, Noise Low represents the signal end average noise, N1 represents the first average value, σ1 represents the first standard deviation, N2 represents the second average value, σ2 represents the second standard deviation, and α is the noise parameter.

5. The photon data denoising and classification method according to claim 1, characterized in that: The step 3 specifically includes: Dividing the photon data of the coarse screening of the elevation histogram into a plurality of second division units in the along-track direction; Calculate the average elevation, 25% quantile elevation and 75% quantile elevation of the photon elevation in each of the second divided units; The second preset elevation threshold range corresponding to each second division unit is set according to the average elevation, 25% quantile elevation and 75% quantile elevation of each second division unit, and the coarse-screened photon data of the elevation histogram in each second division unit whose elevation is within the corresponding second preset elevation threshold range is extracted to obtain the coarse-screened original data of the elevation.

6. The photon data denoising and classification method according to claim 1, characterized in that: The step 4 specifically includes: For any photon P, the LOFE score of the photon P is calculated by the following steps: Calculate the reachable distance between any two photons O and P, where photon O is a photon in the horizontal ellipse K neighborhood of photon P: Reach_distances K (P,O)=max{K-distances K (O),d(P,O)} Among them, Reach_distances K (P, O) represents the reachable distance between O and P, K-distances K (O) represents the KNN distance of the horizontal elliptical K neighborhood of photon O, and d(P, O) represents the distance between photons O and P; According to the reachable distance, obtaining the local reachable density of the photon P; The LOFE score of the photon P is obtained according to the local reachable density and the average local reachable density of photons in the horizontal ellipse K neighborhood of the photon P.

7. The photon data denoising and classification method according to claim 6, characterized in that: The KNN distance calculation formula of the horizontal ellipse K neighborhood is: Among them, P and O represent two different photons, Δx represents the distance between P and O photons in the along-track direction, Δy represents the difference between P and O photons in the elevation direction, and a and b represent the major and minor axis parameters of the horizontal elliptical search area, respectively.

8. The photon data denoising and classification method according to claim 1, characterized in that: The calculation formula of the terrain slope in step 11 is: Among them, α i represents the terrain slope of the ith interval, h p represents the elevation of the coarsely classified ground photon p, h q represents the elevation of the coarsely classified ground photon q, x p represents the along-track distance of the coarsely classified ground photon p, x q Represents the along-track distance of the coarsely classified ground photon q.

9. The photon data denoising and classification method according to claim 1, characterized in that: The KNN distance calculation formula of the rotating ellipse LOF model search area is: Δx=cosα i ×(x p -x q )+sinα i ×(h p -h q ) Δy=sinα i ×(x p -x q )-cosα i ×(h p -h q ) in, represents the KNN distance of the search area of ​​the rotating ellipse LOF model, α i represents the terrain slope of the i-th interval, Δx represents the distance between p and q photons along the track, Δy represents the difference in elevation between p and q photons, and h p represents the elevation of photon p, h q represents the elevation of photon q, x p represents the distance along the track of photon p, x q represents the distance along the track of photon q.

10. A photon data denoising and classification system, characterized in that: include: A target photon data acquisition module is used to acquire target photon data; the target photon data is collected by a photon counting laser radar; The photon data acquisition module of the coarse screening of the elevation histogram is used to extract the target photon data whose elevation is within the first preset elevation threshold range to obtain the photon data of the coarse screening of the elevation histogram; The raw data acquisition module of the elevation rough screening is used to divide the photon data of the elevation histogram rough screening into units in the along-track direction, extract the photon data of the elevation histogram rough screening whose elevation in each divided unit is within the corresponding second preset elevation threshold range, and obtain the raw data of the elevation rough screening; each of the divided units corresponds to a second preset elevation threshold range; A LOFE score calculation module, for calculating the LOFE score of each raw data of the elevation rough screening using a horizontal ellipse LOF model; A coarse denoising signal photon data acquisition module is used to extract the photon data whose LOFE score meets the preset LOFE threshold in the raw data of the elevation coarse screening, and obtain the coarse denoising signal photon data; A first division module, used for performing sliding window division on the coarsely denoised signal photon data in the along-track direction to obtain a plurality of first sliding windows; A coarsely classified ground photon acquisition module is used to extract the coarsely denoised signal photon data with the smallest elevation in each of the first sliding windows to obtain coarsely classified ground photons; A second division module is used to divide the raw data of the elevation rough screening into intervals in the along-track direction to obtain a plurality of first division intervals; The raw data acquisition module of the elevation fine screening is used to extract the raw data of the elevation coarse screening whose elevation in each of the first divided intervals is within the range of the third preset elevation threshold value to obtain the raw data of the elevation fine screening; the raw data of the elevation fine screening includes the coarsely classified ground photons; A third division module is used to divide the original data of the elevation fine screening into intervals in the along-track direction to obtain a plurality of second divided intervals; A terrain slope calculation module, used for calculating the terrain slope of each first divided interval according to the along-track distance and elevation of the coarsely classified ground photons in the original data of the elevation fine screening in each second divided interval; A LOFR score calculation module, for each of the raw data of the elevation fine screening, using the terrain slope corresponding to the second divided interval to which the raw data of the elevation fine screening belongs as the rotation angle of the rotating ellipse LOF model to calculate the LOFR score of the raw data of the elevation fine screening; A signal photon data acquisition module for fine denoising is used to extract photon data whose LOFR score meets a preset LOFR threshold in the original data of the elevation fine screening to obtain signal photon data for fine denoising; The classification module is used to classify the photons in each second sliding window in the along-track direction according to the elevation size of each photon data in the finely denoised signal photon data in each second sliding window, so as to obtain finely classified crown photons, ground photons and canopy photons.

Images