CN111077532A - Surface feature space information acquisition method based on deconvolution and Gaussian decomposition - Google Patents
Surface feature space information acquisition method based on deconvolution and Gaussian decomposition Download PDFInfo
- Publication number
- CN111077532A CN111077532A CN201911156330.9A CN201911156330A CN111077532A CN 111077532 A CN111077532 A CN 111077532A CN 201911156330 A CN201911156330 A CN 201911156330A CN 111077532 A CN111077532 A CN 111077532A
- Authority
- CN
- China
- Prior art keywords
- waveform
- decomposition
- gaussian
- peak
- deconvolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/02—Systems using the reflection of electromagnetic waves other than radio waves
- G01S17/06—Systems determining position data of a target
- G01S17/46—Indirect determination of position data
- G01S17/48—Active triangulation systems, i.e. using the transmission and reflection of electromagnetic waves other than radio waves
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/483—Details of pulse systems
- G01S7/486—Receivers
- G01S7/487—Extracting wanted echo signals, e.g. pulse detection
- G01S7/4876—Extracting wanted echo signals, e.g. pulse detection by removing unwanted signals
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Optical Radar Systems And Details Thereof (AREA)
Abstract
The invention relates to a surface feature space information acquisition method based on deconvolution and Gaussian decomposition, which comprises the following steps: s1: acquiring original waveform data of a satellite laser height measurement system; s2: preprocessing original waveform data, wherein the preprocessing comprises estimation of background noise and removal of random noise; s3: deconvoluting the preprocessed waveform data, and extracting a first waveform parameter which accords with a preset first constraint condition from the deconvoluted waveform data; s4: according to the first waveform parameter, carrying out Gaussian decomposition on the waveform data after deconvolution, and extracting a second waveform parameter which accords with a preset second constraint condition; s5: and acquiring specific ground object space information according to the second waveform parameters. Compared with the prior art, the method is suitable for a large-light-spot system, and has the advantages of high echo decomposition precision, high ranging resolution and the like.
Description
Technical Field
The invention relates to the field of satellite laser height measurement, in particular to a ground object space information acquisition method based on deconvolution and Gaussian decomposition.
Background
Compared with a discrete laser radar system, the full-waveform satellite-borne laser height measurement system not only records the distance information of the earth surface target, but also records the interaction process of the transmitted pulse with the earth surface terrain and the target. Thus, the full waveform parameters have better potential and capability in characterizing terrain features and target vertical distributions. The extraction precision of waveform parameters (wave crest number, peak amplitude, peak position, peak width, skewness and the like) directly influences the accurate expression of terrain and the application of vegetation height extraction, biomass estimation, land surface classification, city modeling and the like.
The complexity of the waveform not only depends on the structure and physical properties of the reflecting ground object, but also relates to the size of the spot of the laser foot. Typically, large spot systems, such as Ice, Cloud, and Land Elevation satellites (Ice, Cloud, and Land Elevation Satellite, ICESat) looking at the ground form a spot of about 70m in diameter, as opposed to a full waveform for a small spot (0.15-3m), which may cover more types of terrain. As shown in figure 1, dense trees, buildings with inclined roofs, asphalt roads and the like exist in the light spots, and complex superposed waves and ground weak echoes exist in the corresponding waveforms of figure 2. In a similar scene, a dense forest area has a challenge of accurately decomposing a waveform due to the fact that a superposed echo and a weak echo are formed by the attenuation action of a canopy structure and the echo in a penetration canopy structure, or a superposed waveform caused by buildings with similar heights and the like.
The full waveform decomposition method comprises Gaussian decomposition, wavelet decomposition, deconvolution, waveform simulation and the like, and the extracted waveform parameters are different according to different principles. Among them, the gaussian decomposition is the most commonly used and successful decomposition method, regardless of the large and small spot system. The gaussian decomposition is based on the assumption that the transmitted pulse is gaussian-like distribution, the scattering cross section of the surface feature can also be expressed by a gaussian model, and the echo waveform is the convolution result of the two, so that the echo waveform can be decomposed by the gaussian model. The gaussian decomposition is generally divided into two steps: estimation of initial waveform parameters and optimal fitting of the waveform. Commonly used waveform fitting methods include Levenberg-Marquardt (LM), Expectation-Maximization (EM), or a modification of both methods, which generally require better initial estimation parameters (peak amplitude, peak position, peak width). The peak amplitude is generally dependent on local maxima, which easily results in missed detection of the superimposed wave or in false detection due to excessive noise.
The deconvolution method considers that the echo waveform is not only reflected from the ground surface feature but also contributes to the system such that the transmission pulse lasts for several nanoseconds, and research by Hancock et al indicates that the longer the transmission pulse (the width corresponding to half the maximum peak amplitude), the higher the degree of blurring of the echo waveform, and the system noise, etc. are. The echo waveform does not represent the vertical distribution of the real surface features. Accurately inverting the structure and physical properties of the terrain necessitates deconvolution of the echo waveform to remove systematic effects. Currently, the deconvolution methods include WF filter, B-spline deconvolution, regularization filter, RL deconvolution, Gold deconvolution, blind deconvolution (blid deconvolution), and the like. Each method has advantages and disadvantages, such as fast calculation of WF filter in frequency domain, but negative amplitude without physical meaning in deconvolution result, and similar situation in B-spline deconvolution result. Regularization filters alleviate the difficulty of numerical computation and effectively suppress the propagation of noise, but rely on an estimate of regularization parameters, and the method is no longer well suited when the noise level cannot be estimated. RL and Gold deconvolution are two iterative calculation methods, and have better performance compared with a WF filter, but the iterative process consumes excessive time and has low calculation efficiency.
The literature research and comparison of the deconvolution method are mostly based on the waveform ranging information extraction of the small-spot system, and the correlation research of the large-spot system is less and the quantitative comparison is lacked.
Disclosure of Invention
The invention aims to provide a ground object space information acquisition method based on deconvolution and Gaussian decomposition, which can accurately decompose a waveform and is suitable for a large-light-spot system, in order to overcome the defects that superimposed waveforms or weak waves exist in satellite laser height measurement waveform data in the prior art and are difficult to distinguish and detect, thereby affecting the distance measurement resolution and the distance measurement precision.
The purpose of the invention can be realized by the following technical scheme:
a surface feature space information acquisition method based on deconvolution and Gaussian decomposition comprises the following steps:
s1: acquiring original waveform data of a satellite laser height measurement system;
s2: preprocessing original waveform data, wherein the preprocessing comprises estimation of background noise and removal of random noise;
s3: deconvoluting the preprocessed waveform data, and extracting a first waveform parameter which accords with a preset first constraint condition from the deconvoluted waveform data;
s4: according to the first waveform parameter, carrying out Gaussian decomposition on the waveform data after deconvolution, and extracting a second waveform parameter which accords with a preset second constraint condition;
s5: and acquiring specific ground object space information according to the second waveform parameters.
Further, in step S3, the preprocessed waveform data is deconvoluted by using a wiener filter, a regularization filter, a richardson method, or a blind deconvolution method.
Further, the blind deconvolution method uses a blind deconvolution term based on a maximum expectation algorithm.
Furthermore, the iteration times of the two iterative solution methods of the Lucy-Richardson method and the blind deconvolution method are 10 times.
Further, the first waveform parameter and the second waveform parameter both include a number of peaks, a peak amplitude, a peak position, and a peak width, and the peak position is a time corresponding to the peak.
Further, in step S4, a gaussian decomposition is performed by a gaussian model, and a computational expression of the gaussian model is:
wherein w (t) is an echo waveform, NlevelIs the noise level, AmIs the amplitude position of the mth Gaussian component, tmIs the peak position of the mth gaussian component,
is the half width of the mth Gaussian component, and N is the number of Gaussian components.
Further, the gaussian decomposition further comprises fitting optimization of the waveform after the gaussian decomposition by using a levenberg-marquardt method.
Further, the first constraint condition includes that the peak amplitude is greater than a noise threshold and the interval between adjacent peak positions is greater than 1ns, the estimation of the background noise specifically includes calculating a mean value and a variance value of the background noise according to multiple sections of original waveform data, and the noise threshold is obtained according to the mean value and the variance value of the background noise.
Further, the first constraint condition further comprises sorting peak values according to peak areas, taking the first six peak values, and obtaining the peak areas by multiplying peak amplitudes by peak half-widths.
In fitting optimization by adopting the Levenberg-Marquardt method, the situation that the peak amplitude is negative may occur, and the situation is not in accordance with the actual situation, so that constraint conditions are added to the amplitude level to detect reasonable waveform components. In addition, the corresponding positions of the peak values should be distributed in the effective signal interval, the positions of the adjacent peak values are larger than the system resolution, and the peak value width should also be larger than the transmission pulse width and have reasonable upper limit constraint.
Further, the second constraint condition includes that the peak amplitude is greater than a noise threshold, the peak position is greater than the start time of the waveform data and less than the end time of the waveform data, the peak half width is (2.5ns,300ns) and the interval between adjacent peak positions is greater than 1ns, the estimation of the background noise specifically includes calculating the mean value and the variance value of the background noise according to the multiple segments of original waveform data, and the noise threshold is obtained according to the mean value and the variance value of the background noise.
Compared with the prior art, the invention has the following advantages:
(1) the invention relates to a ground object space information acquisition method based on deconvolution and Gaussian decomposition, which combines a deconvolution method with a Gaussian decomposition method, firstly preprocesses original waveform data to remove system noise, and then removes the system effect on echo waveforms by using the deconvolution method, thereby solving the defects that the Gaussian decomposition is easy to miss the detection of superposed waves and false detection is caused by overlarge noise, and improving the echo decomposition precision and the distance measurement resolution of the star laser height measurement waveform decomposition method.
(2) The surface feature space information acquisition method based on deconvolution and Gaussian decomposition is not only suitable for a small light spot system, but also suitable for a large light spot system.
(3) In the embodiment of the invention, a conclusion is obtained through a large number of simulation waveform experiments: 1. compared with a direct Gaussian decomposition method, the deconvolution and Gaussian decomposition combination method can obviously improve the detection rate of the Gaussian peak value, and particularly, the wiener filter, the blind deconvolution method and the Gaussian combination method have better effect compared with two deconvolution methods. 2. In true waveform decomposition, the average value of the blind deconvolution and Gaussian combined decomposition Gaussian components is the largest, and the visual fitting effect is better; in addition, compared with the relative ground object height information extracted from the reference scattering cross section obtained by the GLAH05 product parameter comparison analysis, the Google earth image and the airborne point cloud simulation, the wave form fitting precision is higher based on the Blind-Gauss combined decomposition, and the peak value decomposition number is more in line with the actual vertical distribution of the ground object.
(4) The embodiment of the invention shows that the method for acquiring the ground object space information can be applied to the satellite-borne laser height measurement waveform decomposition of complex waveforms in cities and towns or forest areas and the like, and provides a new processing method for the accurate decomposition processing of the satellite-borne laser height measurement waveform.
(5) In the method for acquiring the ground object space information, the iteration times of the Lucy-Richardson method and the blind deconvolution method are 10 times, so that a good recovery effect can be obtained, and the method can be compared with other methods to achieve a rapid and fair waveform decomposition result.
Drawings
FIG. 1 is a distribution diagram of an object in a satellite-borne large light spot foot point;
FIG. 2 is a schematic diagram of waveforms corresponding to distribution of ground objects in a satellite-borne large light spot foot point;
FIG. 3 is a schematic flow chart of a surface feature space information acquisition method based on deconvolution and Gaussian decomposition according to the present invention;
FIG. 4 is an exemplary diagram of a simulated waveform according to an embodiment of the present invention, (4a) is a schematic diagram of a transmitted pulse, (4b) is a schematic diagram of a scattering cross section in reality, and (4c) is a schematic diagram of an echo waveform after convolution;
FIG. 5 is a schematic diagram of a scattering cross-sectional waveform of a real feature in accordance with an embodiment of the present invention;
FIG. 6 is a schematic diagram of the deconvolution of the echo waveform after convolution using a different deconvolution method and comparing the results with the true solid scatter cross-section of FIG. 5;
FIG. 7 is a diagram illustrating the variation of the waveform decomposition consistency rate with increasing SNR ("+" is an anomaly) according to an embodiment of the present invention;
FIG. 8 is a diagram illustrating a comparison between a true value and a distribution diagram of the number of waveform components decomposed by different decomposition methods according to an embodiment of the present invention;
fig. 9 is a schematic diagram illustrating a comprehensive comparison of different combination and decomposition methods according to an embodiment of the present invention, in which a box corresponding to each method in the diagram sequentially represents a coincidence rate, a low detection rate, a false alarm rate, and a multi-detection rate from left to right;
FIG. 10 is a schematic diagram illustrating the variation trend of the false alarm rate with increasing SNR for different combinatorial decomposition methods according to an embodiment of the present invention;
fig. 11 is a schematic diagram of a test scenario selected in the embodiment of the present invention, (11a) is a distribution diagram of town field range in the test range, (11b) is a type diagram of a foot point covered ground object, and (11c) is a schematic diagram of waveform data corresponding to fig. 11 b;
FIG. 12 is a comparison of real reference scattering cross-section results obtained using different deconvolution methods according to an embodiment of the present invention;
FIG. 13 is a comparison graph of the number of peak values of the real waveform combinatorial decomposition method and the direct Gaussian decomposition according to the embodiment of the present invention;
fig. 14 is a comparison graph of a fitting result of different waveform decomposition methods and a simulated scattering cross section, (14a) is a schematic diagram of a decomposition result of a Blind-Gauss combined decomposition method, (14b) is a distribution graph of an actual ground object in a laser foot point (a nail is the center position of a foot mark, an ellipse is the foot mark range, and the background is a Google earth image), (14c) is a schematic diagram of a decomposition result of a GLAH05(Gauss) decomposition method, (14d) is a schematic diagram of a decomposition result of a Lucy-Gauss combined decomposition method, (14e) is a schematic diagram of a decomposition result of a Wnr-Gauss combined decomposition method, (14f) is a schematic diagram of a decomposition result of a Reg-Gauss combined decomposition method, (14g) is a scattering cross section diagram obtained based on high-precision airborne point cloud data simulation, and (14h) is a TIN diagram constructed by airborne point cloud.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
Example 1
As shown in fig. 3, the present embodiment is a surface feature space information obtaining method based on deconvolution and gaussian decomposition, the method includes the following steps:
s1: acquiring original waveform data of a satellite laser height measurement system;
s2: the raw waveform data is pre-processed, including estimation of background noise and removal of random noise. Generally, calculating the mean value and variance of 20 samples of the front part and the rear part of an original waveform as background noise and variance values, and denoising random noise through a Gaussian filter;
s3: deconvoluting the preprocessed waveform data: taking the denoised smooth waveform as the input of a deconvolution method, and respectively processing by a WF filter (wiener filter), a Tikhonov regularization filter (regularization filter), a RL (Lucy-Richardson method) and a Blind deconvolution method (Blind deconvolution method);
extracting initial waveform parameters which accord with a preset first constraint condition from waveform data after deconvolution: and (3) initially estimating the number and the width of Gaussian peaks of the deconvolved waveform, namely the information of the real scattering cross section of the representative ground object by using a knee point method, wherein the maximum value between two adjacent knee points and the corresponding position of the maximum value are respectively used as peak amplitude and peak position initial estimation values.
The first constraint is: the effective peak value needs to be larger than a noise threshold value, and the noise threshold value is usually set to be the noise plus three times of standard deviation; adjacent distinguishable peaks need to be larger than the system ranging resolution by 1 ns; the effective peaks are sorted according to area, and the first 6 important peaks are subjected to Gaussian model fitting and optimization.
The deconvolution method is described in detail below:
in the time domain, the echo waveform can be viewed as being formed by the convolution of the system effect and the surface response. The convolution process blurs the characteristics of the scattering cross section and reduces the resolvability of the ground objects. Deconvolution is used as the inverse process of convolution, and theoretically, the fuzzy effect brought by system contribution can be removed, and separability of ground object targets is improved, so that the ranging resolution is improved. Deconvolution methods can be largely classified into direct methods and iterative methods. In the embodiment, two direct calculation methods of a WF filter and a Tikhonov regularization filter and two iteration methods of RL and Blind Blind deconvolution can be selected.
The following describes the methods in detail:
1) WF filter
Jutzi et al propose a WF filter (formula (1)) to suppress noise by first transforming an echo signal and a transmit pulse into the frequency domain by Fourier transform, dividing the echo waveform by the transmit pulse in the frequency domain to obtain the power spectrum of the scattering cross section of different targets (formula (2)), and finally obtaining the scattering cross section of the targets in the time domain by inverse Fourier transform.
In the formula (I), the compound is shown in the specification,R(f) an echo waveform in the frequency domain;N(f) background noise in the frequency domain; w (f) is a real function, i.e., a wiener filter; rT(f) a transmit pulse of the system;
are the scattering cross-sections of different targets in the frequency domain.
2) Tikhonov regularization filter
Compared with a WF filter, the Tikhonov regularization filter does not need to have power spectrum information prior information of pure signals and noise, and can be directly calculated from background noise only by noise variance and mean value information. The calculation formula for the scattering cross section of the target is as follows:
wherein gamma is a Tikhonov matrix; t isrTransmitting a pulse for the system; r, denoising the echo waveform;
is an estimate of the scattering cross-section of the object.
3) RL method
The RL method is an iterative algorithm in the time domain and is initially used for image restoration. Based on Bayes theory, the iterative process and the remainder calculation formula are as follows:
in the formula (I), the compound is shown in the specification,
is an estimate of the scattering cross-section of the object; r (t) is an echo waveform after denoising; r isiThe remainder of the ith iteration.
4) Blind Blind deconvolution
The Blind deconvolution of Blind is a general term for a method for restoring a case where both a point spread function and a target are uncertain in image restoration. The algorithm can simultaneously restore a point spread function and a target, correspondingly, in the satellite laser height measurement signal processing, the system point spread function and the target response are not exact, so that the denoised waveform and the emission pulse can be respectively used as initial input values of blind deconvolution, the blind deconvolution term based on the maximum expectation algorithm proposed by Fish and the like is adopted in the embodiment, and RL deconvolution is carried out twice each time to respectively restore the target and the point spread function.
For two iterative deconvolution methods of RL and Blind, the number of iterations can affect the effect after deconvolution. Because the noise type of the waveform signal and the image noise distribution generally conform to poisson distribution or gaussian distribution, the embodiment can obtain a better restoration effect by adopting 10 iterations in the process of restoring a general image, and the iteration times of the two iteration solving methods are also set to be 10 times so as to achieve rapid and fair waveform decomposition result comparison.
S4: according to the initial waveform parameters, carrying out Gaussian decomposition on the waveform data after deconvolution, and extracting waveform parameters which accord with a pre-established second constraint condition;
the gaussian decomposition is described in detail below:
gaussian decomposition, i.e. decomposing the echo waveform into a plurality of gaussian components, the model expression of which is as follows:
wherein w (t) is an echo waveform, NlevelIs the noise level, Am、tm、
The amplitude peak position and half width of the mth Gaussian component are respectively, and N is the number of Gaussian components.
And determining the number and width of initial peak values of the waveform after deconvolution according to an inflection point method, wherein the local maximum amplitude and the corresponding position between inflection points are respectively used as the initial values of the peak values and the peak value positions. In order to ensure the reasonability of the initial values, constraint conditions are added, each initial waveform component is sequenced according to the area importance, and the initial components sequenced in the first 6 are subjected to fitting optimization by adopting a Levenberg-Marquardt method. The optimized waveform component may have a negative amplitude condition, which is not in line with the actual condition, so that a constraint condition is added to the amplitude level to detect reasonable waveform components, in addition, the corresponding positions of the peak values should be distributed in an effective signal interval, the positions of the adjacent peak values are greater than the system resolution, and the peak value width should also be greater than the emission pulse width and have reasonable upper limit constraint.
Therefore, the second constraint in this embodiment includes (1) the peak minimum is greater than the noise mean plus 3 times the noise standard deviation; (2) the peak positions are distributed in the effective signal interval; (3) amplitude half-width range (2.5ns < sigma <300 ns); (4) adjacent peak times differ by more than 1 ns.
S5: and extracting the number and the position information of the waveform components according to the waveform parameters meeting the pre-established second constraint condition, acquiring the space information of the specific ground object, and verifying and evaluating the distance measurement precision based on the simulation and the real waveform data.
Test protocol:
to verify the effectiveness of the proposed method, the present embodiment uses simulation and real waveform data for experiments. In the simulation waveform, the robustness of the combined decomposition method under different signal-to-noise ratio conditions is verified by setting floating waveform parameters and signal-to-noise ratios, and then the advantages and disadvantages of different combined decomposition methods are quantitatively evaluated on the evaluation index statistics of batch waveforms after decomposition. Similarly, in the real waveform acquired by ICESat/GLAS, firstly, different combination methods are subjected to deconvolution, then, restored scattering cross sections are compared, then, the decomposition conditions of the different methods are compared in batches aiming at the waveform sample selected from a complex scene, finally, the results of the different combination decomposition methods and the parameters of the GLAH05 product are subjected to comparative analysis, and high-precision airborne point cloud data are adopted to simulate the scattering cross sections in the range of the laser footprint and construct TIN for verification.
1. Index of decomposition evaluation
In order to quantitatively evaluate different combination decomposition methods and direct Gaussian decomposition effects, objective quantitative evaluation indexes are provided:
(1) for simulated waveforms
Defining the consistent waveform components needs to meet the following two conditions, the number of decomposed waveform components is consistent, the mean value error of the peak position of each component is not more than 1ns, and the consistent rate of the defined waveform when the waveform components are consistent is as follows:
1) peak coincidence rate Consisratio: decomposing the number of waveforms with the consistent number of waveform components/the total number of test waveform samples;
in complex scenes such as urban areas or forest areas, superposed waveforms or weak echoes generally occur, and if the number of detected waveform components is less than the number of actual ground objects in different decomposition methods, the less detection rate is defined as follows:
2) low detection rate Lessratio: decomposing the sum of the number of waveform components/the total number of test waveform samples, wherein the number of the waveform components is less than the preset number of the waveform components;
due to the presence of noise, the deconvolution process may amplify the noise, causing the detected waveform components to be spurious, and the false alarm rate is defined as follows:
3) false alarm rate Falseratio: decomposing the total number of waveform components and/or the total number of test waveform samples, wherein the number of the waveform components is larger than the preset number of the waveform components;
compared with direct Gaussian decomposition, different combined decomposition methods have more detected waveform components and the number of the waveform components is less than or equal to the preset number of the waveform components so as to eliminate the influence of false alarm rate, and the multiple detection rate is defined as follows:
4) multi-detection rate Moreratio: the total number of the waveform components is greater than the number of the direct Gaussian components but less than or equal to the preset number of the waveform components/the total number of the test waveform samples;
(2) for ICESat/GLAS waveforms
1) Comparing the number and the mean value of the Gaussian components of different combined decomposition methods with the number and the mean value of the peak value of the GLAH 14;
2) each gaussian component parameter (peak, peak position, peak width) was compared to the GLAH05 product parameters (obtained by direct gaussian decomposition least squares fit);
3) the method includes the steps of firstly comparing and analyzing the height of a ground object obtained by inverting a randomly extracted waveform decomposition result and a scattering cross section extraction result in a laser foot point simulated by high-resolution airborne point cloud data, wherein the principle of waveform simulation is based on a rapid time domain simulation algorithm, namely, a topographic contour line in the laser foot point is converted into an isochrone, and the number of returned photons in a certain time interval is superposed to form a waveform histogram. Sampling is carried out according to the 1GHz/ns sampling rate of an ICESat/GLAS system, and then an in-foot-point object backscattering section of the laser is formed, because the ICESat/GLAS foot point has about 5m plane positioning error on a flat ground surface, the embodiment moves the foot point position within about 5m range during simulation so as to obtain the maximization of a correlation coefficient between a simulation waveform and an actual ICESat/GLAS waveform, and the simulation backscattering section under the maximization of the correlation coefficient is the finally determined backscattering section. In addition, the foot points are projected on the Google earth image according to the positions of the ICESat/GLAS foot points to visually observe the actual ground feature distribution in the foot points.
2. Simulated waveform
2.1 simulation experiment design
In the simulation data of the embodiment, on the matlab platform, it is assumed that both the transmitted pulse and the ground surface true scattering cross section conform to a gaussian model, and the echo waveform is random white gaussian noise added after the convolution of the transmitted pulse and the ground surface true scattering cross section. The signal to noise ratio is incremented from 10db to 40db at 5 steps. The transmission pulse length is 45, the signal-to-noise ratio is set to be 30dB, the energy is 0.75J, and the center position is 25 ns; the sampling number and the sampling frequency of the echo waveform are set according to the parameters of an ICESat/GLAS system: the number of echo samples is 544, and the sample sampling frequency is 1GHz, i.e. 1ns time interval. The maximum number of peaks is 2-6, the amplitude intensity (Amp) is 1-10, the position (Loc) is distributed in the range of 140-380ns, and the minimum value of the width (Sigma) is 3-12 ns. Specific signal-to-noise ratio, amplitude level, peak position, and peak width setting parameters are shown in table 1 below. In order to avoid the contingency of a small amount of echo results, 1000 simulation waveforms are generated by transforming peak parameters and signal-to-noise ratios, and experimental results are statistically analyzed, and fig. 4 is a schematic diagram of a typical echo waveform after a system emission pulse is convolved with a surface scattering cross section in a forming process and a deconvolution effect to restore the scattering cross section:
table 1 simulation waveform parameter set
2.2 simulation waveform deconvolution comparison
As shown in fig. 5 and 6, for example, the effect of deconvolution by the various deconvolution methods is shown in fig. 6, in which one waveform is randomly generated and deconvolution is performed by the above four deconvolution methods: (1) compared with the amplitude of the original waveform, an RL filter (Deconv-Lucy), a Tikhonov regularization filter (Deconv-Reg) and a WF filter (Deconv-Wnr) have enhancement functions; (2) the Blind deconvolution (Deconv-Blind) maintains the same amplitude level as the original waveform; (3) comparing fig. 5 for true scatter cross-sections with 5 components, the various deconvolution methods clearly separated the last two (300- > 320ns) superimposed components.
2.3 simulation waveform decomposition results and discussion
FIG. 7 shows the comparison of the decomposition results between different combined decomposition methods and direct Gaussian decomposition, in which the SNR increases from 10dB to 5 steps to 40dB each time. As can be seen from the figure, (1) compared with the direct Gaussian decomposition method (Gauss), the different combined decomposition methods all achieve higher decomposition consistency rate under different signal-to-noise ratios. (2) The Lucy-Gauss (RL Filter-Gaussian decomposition) mean is relatively lower than the other three combinatorial decomposition methods. (3) Wnr-Gauss (WF filter-Gaussian decomposition method) and Reg-Gauss (Tikhonov regularization filter-Gaussian decomposition method) are slightly higher than the combination decomposition method of Blind-Gauss (Blind deconvolution-Gaussian decomposition method).
As shown in fig. 8, the True values (True) of the gaussian component numbers of 1000 simulated waveforms and the distribution of the gaussian peak numbers after gaussian decomposition (Gauss), deconvolution and gaussian combination decomposition respectively are shown, and as can be seen from fig. 8, compared with the distribution of True peak numbers, different decomposition methods have fewer detection phenomena, but the combined decomposition method for 5-6 relatively complex waveform number distributions is obviously higher than the direct gaussian decomposition method.
Table 2 below lists the mean values of the specific gaussian peak numbers, with the true values in the simulation being 3.675. The direct gaussian decomposition result is 2.988, compared with the direct gaussian decomposition, the different combined decomposition methods are improved, and compared with Lucy-Gauss, the other three combined decompositions are relatively higher, which also proves the statistical result of the histogram.
TABLE 2 simulation waveform different combination decomposition method peak value number average value
As shown in fig. 9, three additional quantitative indicators for different decomposition methods are presented, in addition to the peak coincidence rate (red box): and comparing the results of few detection rates (green frames), false alarm rates (cyan frames) and many detection rates (purple frames). It can be seen that the Lucy-Gauss low detection rate is higher, and the peak value consistency rate is indirectly stated to be lower. The small detection rate of Reg-Gauss changes little, and Wnr-Gauss and Blind-Gauss float much. The median false alarm rate of the Reg-Gauss and Blind-Gauss combined methods is low, which indicates that the false alarm rates of the two methods are low as a whole. Fig. 10 shows in further detail the trend of the false alarm rate as the signal-to-noise ratio increases. Compared with direct Gaussian decomposition, the median levels of the multi-detection rates of different combined decomposition methods are equivalent, and the Reg-Gauss is increased obviously.
As shown in fig. 10, the variation trend of the false alarm rate with the signal-to-noise ratio of different combinatorial decomposition methods is shown, and it can be seen that: (1) the false alarm rates of the Lucy-Gauss, Reg-Gauss and Blind-Gauss three combined decomposition methods are in a descending trend along with the increase of the signal to noise ratio; (2) the Lucy-Gauss has large overall fluctuation, and the false alarm rate can reach 8.7 percent at most under the condition of low signal to noise ratio. The rapid descending trend of the two methods of Reg-Gauss and Blind-Gauss is shown, and the Reg-Gauss approaches to 0 when the signal-to-noise ratio is high, so that the Reg-Gauss has advantages when the signal-to-noise ratio is high compared with other three combined decomposition methods; (3) Wnr-Gauss has a minimum false alarm rate of 1.4% when the signal-to-noise ratio is low, and the increase may be caused by amplification or ringing of noise, and the probability of decrease after the increase occurs as the signal-to-noise ratio increases, which indicates that Wnr-Gauss has a better advantage in the case of low signal-to-noise level than other combined decomposition methods.
Table 3 below shows the mean value and standard deviation of the peak position errors of the gaussian decomposition and different combined decomposition methods, and it is found that the Lucy-Gauss peak position error is the smallest (0.136), which indicates that the ranging accuracy is more accurate than that of other combined decomposition methods. And secondly, the Reg-Gauss combined decomposition method also has smaller range error and standard deviation. The mean of the Blind-Gauss errors is slightly lower than the Gauss decomposition, but with a smaller standard deviation. Wnr-Gauss has the largest mean error value and the smallest standard deviation error, which indicates that the distance measurement precision is low.
TABLE 3 comparison of peak position error mean and standard deviation of different combination decomposition methods for simulation waveforms
For a specific scene such as an urban area and a forest area, we further assume that the number of peaks is 2-4 and 4-6 respectively. Table 4 below shows the quantitative index comparison between the gaussian decomposition and different combined decomposition methods, where the combined decomposition method has a higher detection rate of at least 6.5% in forest zones than the direct gaussian decomposition; the forest region is at least 15.8 percent and can reach 20.8 percent at most, which shows that the combined decomposition method has stronger distinguishing capability on different vertical structures aiming at more complex waveforms.
TABLE 4 comparison of simulated waveform decomposition results of different combination decomposition methods and direct Gaussian decomposition in urban and forest regions
3. True waveform
3.1 test scenario selection
As shown in fig. 11, generally, complex waveforms are generally generated from diverse ground features, surface types, forest zones, or the like. In order to test a deconvolution method and verify a combined decomposition method, a point cloud data set which has a land area with various ground object types and high-precision point cloud data and ICESat/GLAS data in space synchronously distributed and open sources is selected to be obtained from a laser radar project in Utah, and the data obtaining time is 10 months in 2013, 18 months to 5 months and 31 days in 2014. (http:// opentopo. sdsc. edu/datasetmetadataotcollectionID ═ OT.122014.26912.1)
3.2 ICESat/GLAS data
In this experiment, a total of 82 footpoint samples were used for comparison. The data of the transmitted pulse and the echo waveform come from a global altimetry data product GLAH01 of L1A, the parameters (Gaussian component amplitude, position and width) related to the waveform through Gaussian decomposition come from product parameter data GLAH05, and the exact number of Gaussian components on land comes from the parameters of the GLAH14 product. Different data product parameters are matched by the ID of each pulse.
3.3, referencing airborne point cloud data
High resolution (0.5m or 1m) airborne point cloud data is distributed in Utah, U.S. with horizontal and vertical references UTM Zone 12N, NAD83(2011) [ EPSG:26912], NAVD88(GEOID12A) [ EPSG:5703], and point cloud density of 11.93pts/m2, respectively. In this experiment, the intensity value attributes of the point clouds are used to represent the reflectivity changes of various ground objects in the foot points.
3.4 Experimental results and discussion
The actual reference scattering cross section obtained after randomly selecting one of the 82 sample echo waveforms and applying different deconvolution methods respectively is shown in fig. 12 below. As can be seen from the figure, Blind deconvolution (Deconv-Blind) well maintains the amplitude level of the original waveform, while the noise amplitude is also enhanced, but the effective peak can be screened by the noise threshold level. The other three deconvolution methods, wiener filter (Deconv-Wnr), regularization filter (Deconv-Reg), Lucy-Richardson RL deconvolution (Deconv-Lucy), were similar in magnitude level.
The distribution of the number of gaussian peaks after the 82 GLAH01 echo waveforms are respectively subjected to gaussian decomposition, deconvolution and gaussian combination decomposition is shown in fig. 13, and as can be seen from fig. 13, compared with the distribution of the number of real peaks, the detection phenomena are less in different decomposition methods. But the combined decomposition is obviously improved compared with the Gaussian decomposition, especially on the distribution of the number of complex echo waveforms (5-6 Gaussian components). In the combined decomposition method, 2 echoes without effective Gaussian components and some single-wave peak echoes exist in the Lcuy-Gauss combination, and the analysis reason is that after RL deconvolution, parameters of adjacent peaks do not meet the constraint condition in the combined decomposition optimization algorithm and are screened out.
Table 5 below shows the mean value of the peak numbers of the gaussian decomposition and different combined decomposition methods, which can show that the combined decomposition is significantly improved over the number of peak components of the gaussian decomposition, especially 2.4878 number of blind deconvolution and gaussian combined decomposition average detection components.
TABLE 5 simulation waveform different combination decomposition method peak value number average value
3.5 verification of blind convolution-Gaussian combined decomposition result
Both fig. 13 and table 5 above illustrate that the method based on the Blind-Gauss combined decomposition has higher peak detection rate than the GLAH05 product parameter and the other three combined decompositions. In order to verify whether the number of multi-detection peak values of the combination method is reasonable, a waveform with large difference of the number of decomposed waveform components is extracted for qualitative and quantitative comparison analysis: in fig. 14(a), 5 waveform components are the result of Blind-Gauss combined decomposition, and in fig. 14(b), the actual distribution of features in the laser foot points obviously has 5 different height differences, as indicated by blue numbers 1-5, and the number of decomposition by GLAH05 and other combined decomposition methods is 2-4 (fig. 14(c) - (f)). From the fitting effect, the Blind-Gauss fitting effect is better, and the fitting deviation of other decomposition methods is relatively larger. Assuming that the last peak, component1, represents the ground information, several different combined decomposition methods and the GLAH05 method differ in ground position determination and separation of the following different height-difference superposition waves.
When the first peak (component5) in the Blind-Gauss decomposition result is represented by the first highest flat roof (as marked by 1 in the figure) and the last peak (component1) is represented by the ground as marked by 5 in the figure, the waveforms reflected by the planes and slopes marked by 2, 3 and 4 in the figure with small height difference are overlapped and broadened, and are relatively consistent with the waveform decomposition result. The highest flat roof information can be detected by other decomposition methods, only 2 peak values exist in the GLAH05, the ground information is lost when planes and slopes marked by 2, 3 and 4 in the graph are mixed, and compared with the GLAH05, the three combined decomposition methods of Lucy-Gauss, Wnr-Gauss and Reg-Gauss have the advantage that superposed components are separated, but ground components are not separated. Fig. 14(g) is a scattering cross section obtained by simulation based on high-precision airborne point cloud data, and it can be seen that there are also 5 distinct peak values as indicated by the numbers 1-5 in the figure. FIG. 14(h) shows TINs constructed from airborne point clouds, with significant height differences also visible in the plots and corresponding elevation legends.
On the assumption of a relative time axis, the last peak value represents ground information, and the relative ground object height obtained by different deconvolution and Gaussian combined decomposition inversion, the relative ground object height obtained by GLAH05 product parameter inversion and a simulated scattering cross section are calculated through the graph 14 to obtain result comparison; as can be seen by comparison, 4 ground object heights relative to the ground are obtained by inverting 5 peaks obtained by Blind-Gauss decomposition, except for the difference of the first height difference of 0.851m, the other three height differences are very close to the inversion result of the scattering cross section, and the combination decomposition results of other kinds and the GLAH05 have insufficient peak decomposition and larger difference of the height errors of the inverted relative ground objects.
4. Conclusion
The embodiment provides a ground object space information acquisition method combining deconvolution and Gaussian decomposition based on respective advantages of the deconvolution and Gaussian decomposition methods, aiming at the problem that the resolution and the ranging precision are affected due to the fact that superimposed waveforms or weak waves exist in satellite laser height measurement waveform data and are difficult to distinguish and detect. The method can be analyzed through a large number of simulation waveform experiments: (1) compared with a direct Gaussian decomposition method, the deconvolution and Gaussian decomposition combination method can obviously improve the detection rate of the Gaussian peak value, and particularly, the wiener filter, the blind deconvolution method and the Gaussian combination method have better effect than two deconvolution methods. (2) In the real waveform decomposition, the blind deconvolution and Gaussian combination decomposition Gaussian component mean value is the largest, and the visual fitting effect is better. In addition, compared with the relative ground object height information extracted from the reference scattering cross section obtained by the GLAH05 product parameter comparison analysis, the Google earth image and the airborne point cloud simulation, the wave form fitting precision is higher based on the Blind-Gauss combined decomposition, and the peak value decomposition number is more in line with the actual vertical distribution of the ground object. The method introduced by the embodiment can be applied to satellite-borne laser height measurement waveform decomposition of complex waveforms in cities and towns or forest areas and the like, and provides a new processing method for accurate satellite-borne laser height measurement waveform decomposition processing.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (10)
1. A surface feature space information acquisition method based on deconvolution and Gaussian decomposition is characterized by comprising the following steps:
s1: acquiring original waveform data of a satellite laser height measurement system;
s2: preprocessing original waveform data, wherein the preprocessing comprises estimation of background noise and removal of random noise;
s3: deconvoluting the preprocessed waveform data, and extracting a first waveform parameter which accords with a preset first constraint condition from the deconvoluted waveform data;
s4: according to the first waveform parameter, carrying out Gaussian decomposition on the waveform data after deconvolution, and extracting a second waveform parameter which accords with a preset second constraint condition;
s5: and acquiring specific ground object space information according to the second waveform parameters.
2. The method for acquiring ground object space information based on deconvolution and gaussian decomposition of claim 1, wherein in step S3, the preprocessed waveform data is deconvolved by using a wiener filter, a regularization filter, a riches-richardson method or a blind deconvolution method.
3. The method for acquiring the ground feature space information based on deconvolution and Gaussian decomposition as claimed in claim 2, wherein the blind deconvolution method uses a blind deconvolution term based on a maximum expectation algorithm.
4. The method for acquiring the ground feature space information based on the deconvolution and the gaussian decomposition according to claim 2, wherein the iteration times of the two iterative solution methods of the lucy-richardson method and the blind deconvolution are 10 times.
5. The method according to claim 1, wherein the first waveform parameter and the second waveform parameter each include a peak number, a peak amplitude, a peak position, and a peak width, and the peak position is a time corresponding to a peak.
6. The method for acquiring feature space information based on deconvolution and gaussian decomposition of claim 5, wherein in step S4, gaussian decomposition is performed through a gaussian model, and a calculation expression of the gaussian model is as follows:
wherein w (t) is an echo waveform, NlevelIs the noise level, AmIs the amplitude position of the mth Gaussian component, tmIs the peak position of the mth gaussian component,
is the half width of the mth Gaussian component, and N is the number of Gaussian components.
7. The method as claimed in claim 6, wherein the Gaussian decomposition further comprises fitting and optimizing the Gaussian-decomposed waveform by using a Levenseger-Marquardt method.
8. The method as claimed in claim 5, wherein the first constraint condition includes that a peak amplitude is greater than a noise threshold and an adjacent peak position interval is greater than 1ns, the estimation of the background noise is specifically to calculate a mean value and a variance value of the background noise according to the multiple segments of original waveform data, and the noise threshold is obtained according to the mean value and the variance value of the background noise.
9. The method according to claim 8, wherein the first constraint condition further includes sorting the peak values according to peak areas, and taking the first six peak values, wherein the peak areas are obtained by multiplying peak amplitudes by peak half-widths.
10. The method as claimed in claim 5, wherein the second constraint condition includes that a peak amplitude is greater than a noise threshold, a peak position is greater than a waveform data start time and less than a waveform data end time, a peak half width is between (2.5ns,300ns), and an interval between adjacent peak positions is greater than 1ns, the estimation of the background noise is specifically to calculate a mean value and a variance value of the background noise according to a plurality of segments of original waveform data, and the noise threshold is obtained according to the mean value and the variance value of the background noise.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911156330.9A CN111077532A (en) | 2019-11-22 | 2019-11-22 | Surface feature space information acquisition method based on deconvolution and Gaussian decomposition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911156330.9A CN111077532A (en) | 2019-11-22 | 2019-11-22 | Surface feature space information acquisition method based on deconvolution and Gaussian decomposition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111077532A true CN111077532A (en) | 2020-04-28 |
Family
ID=70311732
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911156330.9A Pending CN111077532A (en) | 2019-11-22 | 2019-11-22 | Surface feature space information acquisition method based on deconvolution and Gaussian decomposition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111077532A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114070408A (en) * | 2021-11-16 | 2022-02-18 | 哈尔滨工程大学 | Cross-ice medium acoustic communication waveform design method |
| CN116719086A (en) * | 2023-05-11 | 2023-09-08 | 中国科学院精密测量科学与技术创新研究院 | Sparse seabed four-component data high-resolution imaging method based on point spread function |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2267480A1 (en) * | 2009-06-08 | 2010-12-29 | Honeywell International Inc. | Systems and methods for Gaussian decomposition of weather radar data for communication |
| CN102445685A (en) * | 2011-09-28 | 2012-05-09 | 赖旭东 | Small spot radar signal decomposition method |
| CN102538673A (en) * | 2011-12-19 | 2012-07-04 | 中国林业科学研究院资源信息研究所 | Method for extracting forest stand vertical structure parameters on basis of airborne small-facula LiDAR (light detection and ranging) waveform data |
| CN105068063A (en) * | 2015-07-16 | 2015-11-18 | 同济大学 | Ground object characteristic obtaining method based on echo full-waveform decomposition |
| CN105676205A (en) * | 2016-01-27 | 2016-06-15 | 武汉大学 | Airborne LiDAR waveform data Gaussian decomposition method |
| CN109709527A (en) * | 2019-01-14 | 2019-05-03 | 上海海洋大学 | The Gauss wave crest method of Gauss Decomposition in a kind of Full wave shape laser-measured height echo signal |
-
2019
- 2019-11-22 CN CN201911156330.9A patent/CN111077532A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2267480A1 (en) * | 2009-06-08 | 2010-12-29 | Honeywell International Inc. | Systems and methods for Gaussian decomposition of weather radar data for communication |
| CN102445685A (en) * | 2011-09-28 | 2012-05-09 | 赖旭东 | Small spot radar signal decomposition method |
| CN102538673A (en) * | 2011-12-19 | 2012-07-04 | 中国林业科学研究院资源信息研究所 | Method for extracting forest stand vertical structure parameters on basis of airborne small-facula LiDAR (light detection and ranging) waveform data |
| CN105068063A (en) * | 2015-07-16 | 2015-11-18 | 同济大学 | Ground object characteristic obtaining method based on echo full-waveform decomposition |
| CN105676205A (en) * | 2016-01-27 | 2016-06-15 | 武汉大学 | Airborne LiDAR waveform data Gaussian decomposition method |
| CN109709527A (en) * | 2019-01-14 | 2019-05-03 | 上海海洋大学 | The Gauss wave crest method of Gauss Decomposition in a kind of Full wave shape laser-measured height echo signal |
Non-Patent Citations (7)
| Title |
|---|
| TAN ZHOU ET AL，: "Gold – A novel deconvolution algorithm with optimization for waveform LiDAR processing", 《ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING》 * |
| 刘世杰等: "一种改进的高分辨率卫星影像数字表面模型生成方法", 《同济大学学报（自然科学版）》 * |
| 刘诏等: "星载雷达全波形数据分解方法", 《第五届高分辨率对地观测学术年会》 * |
| 李国元等: "一种改进的卫星激光测高全波形高斯分解方法", 《测绘科学》 * |
| 段乙好等: "一种用于机载激光雷达波形分解的小波变换与广义高斯模型组合法", 《激光与光电子学进展》 * |
| 王滨辉等: "全波形激光雷达的波形优化分解算法", 《测绘学报》 * |
| 黄冬梅等: "ＧＬＡＳ全波形数据的高斯分解与高斯小波基分解对比分析", 《激光与光电子学进展》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114070408A (en) * | 2021-11-16 | 2022-02-18 | 哈尔滨工程大学 | Cross-ice medium acoustic communication waveform design method |
| CN114070408B (en) * | 2021-11-16 | 2023-09-19 | 哈尔滨工程大学 | Method for designing acoustic communication waveform of ice-crossing medium |
| CN116719086A (en) * | 2023-05-11 | 2023-09-08 | 中国科学院精密测量科学与技术创新研究院 | Sparse seabed four-component data high-resolution imaging method based on point spread function |
| CN116719086B (en) * | 2023-05-11 | 2024-03-12 | 中国科学院精密测量科学与技术创新研究院 | Sparse seabed four-component data high-resolution imaging method based on point spread function |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR101007662B1 (en) | Radar Signals Clustering Method using Frequency Modulation Characteristics and Combination Characteristics of Signals, and System for Receiving and Processing Radar Signals using the same | |
| Hilliard | Selection of a clutter rejection algorithm for real-time target detection from an airborne platform | |
| Magruder et al. | Noise filtering techniques for photon-counting ladar data | |
| CN106154247B (en) | A kind of multiple dimensioned Full wave shape laser radar data optimizes decomposition method | |
| US8659747B2 (en) | Determining thresholds to filter noise in GmAPD LADAR data | |
| CN105160648B (en) | Radar target and Shadow segmentation method based on small echo and constant false alarm rate | |
| CN111665517B (en) | Density statistics-based single photon laser height finding data denoising method and device | |
| CN111077532A (en) | Surface feature space information acquisition method based on deconvolution and Gaussian decomposition | |
| Wu et al. | A robust signal preprocessing chain for small-footprint waveform lidar | |
| CN112986964B (en) | Photon counting laser point cloud self-adaptive denoising method based on noise neighborhood density | |
| Zhang et al. | A combined deconvolution and Gaussian decomposition approach for overlapped peak position extraction from large-footprint satellite laser altimeter waveforms | |
| CN115984778A (en) | Multi-feature optimization based method for rapidly and dynamically monitoring Sentinel-1 data in flood | |
| CN111257905B (en) | Slice self-adaptive filtering algorithm based on single photon laser point cloud density segmentation | |
| CN113281716A (en) | Photon counting laser radar data denoising method | |
| CN112666527B (en) | Weather radar filtering method fusing live precipitation data | |
| CN106022217A (en) | Civil airport runway area detection method free from supervision multistage classification | |
| Chapple et al. | Stochastic model-based processing for detection of small targets in non-gaussian natural imagery | |
| CN112989940A (en) | Raft culture area extraction method based on high-resolution three-satellite SAR image | |
| Kampa et al. | An adaptive multiscale filter for segmenting vegetation in ALSM data | |
| CN114839602B (en) | Radar signal statistical feature extraction method and device based on electromagnetic data clustering | |
| CN110837088A (en) | Data denoising method for spaceborne laser altimeter | |
| CN115980689A (en) | Point cloud detection-based radiation source signal sorting method, device, equipment and medium | |
| CN115327539A (en) | Road guardrail detection method and device | |
| Stevens et al. | Quantitative data quality metrics for 3D laser radar systems | |
| Neuenschwander et al. | Signal processing techniques for feature extraction and classification using small-footprint full-waveform airborne LIDAR |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200428 |
|
| RJ01 | Rejection of invention patent application after publication |