Dual-threshold lidar cloud detection algorithm based on cluster analysis

Abstract

The present invention discloses a dual-threshold laser radar cloud detection algorithm based on cluster analysis, including: preprocessing laser radar echo signals; using the layer peak-to-layer bottom ratio threshold to screen the preprocessed laser radar echo signals and extract cloud signals; using the background noise threshold to extract the missed thin cloud signals from the preprocessed laser radar echo signals; using the differential zero crossing method to obtain the differential zero points of each cloud layer and inversely calculate the cloud top and cloud bottom heights; and using the ISODATA clustering algorithm to cluster the relative humidity and vertical height of the differential zero points of each cloud layer, clustering samples with similar relative humidity characteristics within the same height range into one category, obtaining multiple differential zero point sets after classification, and distinguishing the cloud structure to obtain cloud information. The present invention effectively eliminates the interference of aerosol signals and can accurately screen out cloud signals, thereby realizing accurate processing of cloud layers.

Description

Technical Field

[0001] The invention relates to a dual-threshold laser radar cloud detection algorithm based on cluster analysis, and belongs to the technical field of laser radar detection. Background Art

[0002] Clouds cover 60% of the world's area. They participate in large-scale circulation and global water cycle by affecting solar radiation and latent heat release, and play an important role in global climate change research. LiDAR has high detection accuracy and strong sustainability, and is widely used in cloud detection. The laser cloud meter uses LiDAR as its main equipment. When it is working, it emits laser pulses into the atmosphere. By analyzing the received laser echo signal, it can invert cloud base height, cloud top height and other cloud information. Therefore, the correct judgment of cloud signals is extremely critical.

[0003] In view of the real-time, high-efficiency and continuous characteristics of lidar in high-temporal and spatial resolution cloud structure detection, it has gradually become an effective means to detect the spatial and temporal distribution of atmospheric physical parameters, clouds and aerosols, and has been widely used in the continuous monitoring of the spatial distribution of aerosols and clouds in the mid- and low-altitude atmosphere. The differential zero-crossing method is a common inversion method for cloud parameter inversion based on lidar echo signals, but this method is prone to misjudgment of cloud signals when the signal-to-noise ratio is low, affecting the detection accuracy. Under normal circumstances, the density of clouds in the troposphere is much higher than that of aerosols. However, when there is a thin cloud signal similar to the aerosol layer signal or a thick aerosol layer signal similar to the cloud layer, it is difficult to directly extract cloud information through the lidar echo signal. Although there have been many improvements to the algorithm in recent years, such as fitting data, combining the judgment of the neighborhood moments before and after the echo signal, and combining threshold methods. However, there are still the following disadvantages: 1. There is interference from aerosol signals, and correct cloud information cannot be provided; 2. When the background noise in the scene fluctuates greatly, this method will produce large errors; 3. The setting of the threshold is subjective, and it is extremely easy to deviate in practical applications.

[0004] The existence of the above problems seriously restricts the efficiency of lidar in measuring cloud height. Accurate measurement of cloud signals is the key to solving the problem. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a dual-threshold lidar cloud detection algorithm based on cluster analysis, which effectively eliminates the interference of aerosol signals and accurately screens out cloud signals.

[0006] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0007] A dual-threshold lidar cloud detection algorithm based on cluster analysis includes the following steps:

[0008] Step 1: preprocessing the laser radar echo signal;

[0009] Step 2: Use double thresholds to process the preprocessed lidar echo signal: use the layer peak-to-layer bottom ratio threshold to screen the preprocessed lidar echo signal and extract the cloud layer signal; use the background noise threshold to extract the missed thin cloud layer signal from the preprocessed lidar echo signal;

[0010] Step 3: Based on the extracted cloud layer signal and thin cloud layer signal, the differential zero crossing method is used to perform differential processing to obtain the differential zero point of the extracted cloud layer signal, and the cloud top and cloud bottom heights are calculated by inversion;

[0011] Step 4: Use the ISODATA clustering algorithm to perform cluster analysis on the differential zero points of each cloud layer signal according to relative humidity and vertical height, and cluster samples with similar relative humidity characteristics within the same height range into one category, that is, each category of differential zero points can be regarded as the differential zero points of the same layer of cloud, thereby obtaining multiple classified differential zero point sets; the cloud layer structure is judged based on the classified multiple differential zero point sets to obtain the corresponding cloud layer information.

[0012] Furthermore, as a preferred technical solution of the present invention, the step 1 of preprocessing the lidar echo signal includes: using the LMD-ITM algorithm to denoise the initial lidar echo signal, and performing a 5-point sliding average process on the denoised signal.

[0013] Further, as a preferred technical solution of the present invention, in step 2, the preprocessed laser radar echo signal is screened using the layer peak to layer bottom ratio threshold, using the formula:

[0014]

[0015] Among them, X is the layer peak to layer bottom ratio threshold; r m 、r b are the positions of the cloud layer peak and bottom, respectively, P(r b )、P(r m ) are the laser radar echo signal intensities at the cloud layer peak and bottom, respectively.

[0016] Further, as a preferred technical solution of the present invention, in step 2, the background noise threshold is used to extract the missed thin cloud layer signal from the preprocessed lidar echo signal, specifically:

[0017] When the laser radar detection distance r<5km, set the background noise threshold Y=[P(r m)-P(r b )]>3I, where P(r b )、P(r m ) are the laser radar echo signal intensities at the cloud layer peak and bottom, respectively, and I is the background noise intensity;

[0018] When the laser radar detection distance r ≥ 5km, set the daytime background noise threshold Y = [P (r m )-P(r b )]>1.5I, set the background noise threshold at night Y=[P(r m )-P(r b )]>25I.

[0019] Further, as a preferred technical solution of the present invention, in step 4, the differential zero points of each cloud layer signal are clustered and analyzed according to relative humidity and vertical height using the ISODATA clustering algorithm, which specifically includes:

[0020] Construct cloud layer differential zero point set;

[0021] The constructed cloud layer differential zero point set is discriminated and classified, including splitting and merging. In the iterative discrimination process, if the number of zero-point initial cluster centers Nc≤K / 2, K is the expected number of cluster centers, that is, the number of zero-point initial cluster centers is less than or equal to half of the expected number of cluster centers, then the existing clusters are split; if the number of iterative operations is an even number, or Nc≥2K, then the merging process is performed until the maximum number of iterations is reached;

[0022] Then each differential zero point set is the differential zero point of the same cloud layer, and the cloud base and cloud top heights of each cloud layer are the minimum cloud base height and the maximum cloud top height calculated by the differential zero crossing method within the differential zero point set; in this step of calculation, the number of iterations is increased by 1, so far, the cloud layer differential zero point clustering is completed, and multiple differential zero point sets after classification are obtained.

[0023] The present invention adopts the above technical solution to produce the following technical effects:

[0024] The method of the present invention, according to the characteristics of cloud layer signals and aerosol signals, firstly pre-processes the laser radar echo signal so that its zero crossing points are significantly reduced, thereby effectively improving the quality of the laser radar echo signal; and, combines the improved layer peak-to-layer bottom ratio threshold and the background noise threshold with the differential zero crossing method, thereby effectively avoiding the misjudgment of cloud layer signals caused by noise signals and aerosol signals, thereby improving the screening accuracy of the laser radar cloud layer signals; and, for the first time, combines the clustering algorithm with the differential zero crossing method, according to the vertical distribution characteristics of the cloud layer, combined with the vertical height and relative humidity of the signal differential zero point of the cloud layer signal, adopts the ISODATA algorithm to perform cluster analysis on the differential zero points of each cloud layer signal, obtains a plurality of differential zero point sets after classification, adaptively integrates and divides the detected multi-segment cloud layer signals, gets rid of the subjective limitation of the traditional threshold method in the layered processing, and is particularly accurate for the detection of thin cloud layer signals and high-level cloud signals.

[0025] Therefore, the method of the present invention effectively eliminates the interference of aerosol signals and can accurately screen out cloud signals. At the same time, the clustering algorithm is applied to cloud structure analysis for the first time, realizing accurate processing of cloud layers. Experiments were conducted using ARM data. The experimental results show that the algorithm achieves a detection accuracy of 93.62% for low-level clouds, 92.78% for mid-level clouds, and 93.03% for high-level clouds, which is greatly improved compared with traditional algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The figure is a flow chart of the method of the present invention.

[0027] Figure 2 Schematic diagram of comparison between the original laser radar echo signal and the preprocessed laser radar echo signal in the present invention.

[0028] Figure 3 It is a schematic diagram of the basic principle of the differential zero-crossing method in the present invention.

[0029] Figure 4 It is a schematic diagram of the simulated laser radar echo signal and the corresponding algorithm processing results in the present invention.

[0030] Figure 5 It is a schematic diagram of the laser radar echo signal and the cloud layer part amplified signal in the present invention.

[0031] Figure 6 It is a schematic diagram of the inversion result of the improved algorithm in the present invention and the amplified signal of the cloud layer.

[0032] Figure 7 It is a schematic diagram of the detection results of the differential zero-crossing method and the improved algorithm in the present invention. DETAILED DESCRIPTION

[0033] The present invention is further explained below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0034] like Figure 1 As shown, the present invention relates to a dual-threshold lidar cloud detection algorithm based on cluster analysis, comprising the following steps:

[0035] Step 1: First, preprocess the laser radar echo signal.

[0036] The present invention adopts the LMD-ITM algorithm for the preprocessing of the laser radar echo signal. The initial laser radar echo signal is denoised by the LMD-ITM algorithm, and the denoised signal is subjected to a 5-point sliding average process. The laser radar echo signal before and after preprocessing is shown in FIG. Figure 2 As shown. Figure 2 It can be seen that the zero crossing points of the preprocessed lidar echo signal are significantly reduced, which effectively improves the quality of the lidar echo signal.

[0037] Step 2: In order to effectively select cloud signals from interference signals, the present invention uses a double threshold to process the pre-processed laser radar echo signal, as follows:

[0038] 1) Layer peak to layer bottom ratio threshold X: The pre-processed lidar echo signal is screened using the layer peak to layer bottom ratio threshold X to extract the cloud signal. After repeated experiments, the present invention selects the ratio of the layer peak to layer bottom correction signal by Morille et al. as the initial threshold X to distinguish the cloud signal from the thick aerosol signal. The expression is as follows:

[0039]

[0040] Among them, r m 、r b are the positions of the cloud layer peak and bottom, respectively, P(r b )、P(r m ) are respectively the cloud layer peak r m , bottom layer b The laser radar echo signal strength. When the layer peak-to-layer bottom ratio threshold X ≥ 4 in the formula, it can be judged as cloud information. The value of 4 is the best value selected after multiple tests. Otherwise, when X < 4, it can be judged as a thick aerosol signal. Therefore, through the layer peak-to-layer bottom ratio threshold X screening, the obvious cloud signal can be extracted.

[0041] 2) Background noise threshold Y: The background noise threshold is used to extract the missed thin cloud layer signals from the preprocessed lidar echo signal. After being screened by the layer peak-to-layer bottom ratio threshold X, there may be missed valid cloud layer signals in the signal. A more accurate threshold is needed to distinguish the thin cloud layer signals from the aerosol signals. In order to make the judgment of cloud base and cloud peak more accurate, the background noise threshold proposed by Pal et al. is introduced for judgment. Pal believes that only when the difference between the cloud peak in the lidar echo signal and its corresponding cloud bottom intensity is greater than 2 times the background noise intensity, it will be regarded as a valid cloud base and cloud peak. The formula is as follows:

[0042] Y=[P(r m )-P(r b )]>2I (2)

[0043] Wherein, I is the background noise intensity. Considering that there is a certain difference in the background noise intensity during the day and at night, the threshold 2I proposed by Pal et al. is improved here.

[0044] After many experiments, the present invention determines that when the laser radar detection distance r<5km, the background noise threshold Y=[P(r m )-P(r b )]>3I; When the laser radar detection distance r≥5km, set the daytime background noise threshold Y=[P(r m )-P(r b )]>1.5I, set the background noise threshold at night Y=[P(r m )-P(r b )]>25I.

[0045] Therefore, by improving the background noise threshold Y processing on the processed signal, the missed thin cloud layer signal is selected to effectively prevent missed judgment and misjudgment.

[0046] Step 3: Based on the extracted cloud layer signal and thin cloud layer signal, the differential zero crossing method is used to perform differential processing to obtain the differential zero point of the extracted cloud layer signal, and the cloud top and cloud bottom heights are calculated by inversion.

[0047] The basic principle of the differential zero crossing method is as follows Figure 3 As shown, the details are as follows:

[0048] When the laser irradiates the cloud boundary, the atmospheric backscatter echo signal intensity P(r) will increase rapidly because the cloud density is much higher than the aerosol. When the laser passes through the cloud, P(r) will decrease rapidly, and a cloud echo signal will be generated. Figure 3 As shown, curve (a) is the atmospheric backscatter echo signal intensity P(r), and curve (b) is the simulated lidar echo signal, where r b 、rt Respectively represent the cloud base and cloud top positions, r m Indicates the cloud peak position of the cloud layer. Figure 3 It can be seen that a cloud lidar echo signal can be expressed as a set of adjacent peaks and troughs on its corresponding first-order differential signal, and the cloud bottom height r b 、Cloud peak height r m 、Cloud top height r t That is, the zero point where its differential signal intersects the height coordinate axis. Using the differential zero-crossing method to invert cloud parameters, the differential zero point of each cloud layer can be obtained, and the cloud layer information is judged by the zero point on the differential signal. The inversion process has high requirements on the quality of the laser radar echo signal. However, the laser radar will inevitably be polluted by various noises such as sunlight during actual use. In addition, both the aerosol layer and the cloud layer will cause changes in the laser radar echo signal. Using the differential zero-crossing method, it is very easy to misjudge a thick aerosol layer as a cloud layer or a thin cloud layer as an aerosol. In order to reduce the interference caused by the noise signal in the laser radar echo signal, it is necessary to effectively pre-process the original echo signal, especially in the far field where the laser radar echo signal is weak, which easily causes dP / dr to fluctuate around zero. To this end, the present invention uses a dual threshold technology to process the echo signal on the basis of the differential zero-crossing method.

[0049] Step 4: Use the ISODATA clustering algorithm to cluster the differential zero points of each cloud layer signal according to relative humidity and vertical height, and cluster samples with similar relative humidity characteristics within the same height range into one category, that is, each category of differential zero points can be regarded as the differential zero points of the same layer of cloud, thereby obtaining multiple classified differential zero point sets; the cloud layer structure is discriminated according to the classified multiple differential zero point sets to obtain the corresponding cloud layer information, as follows:

[0050] Adjacent cloud layers can be classified as the same layer of clouds based on physical properties such as relative humidity. Therefore, clustering analysis of the relative humidity and height of the differential zero points of each cloud layer signal obtained in step 3 using the ISODATA clustering algorithm can effectively classify clouds with roughly the same relative humidity within the same height range. The ISODATA clustering algorithm can cluster the constructed differential zero point set and group samples with similar relative humidity characteristics into one category, which can be regarded as the same layer of clouds. The present invention sets the distance threshold between every two layers of clouds to 0.05km. If the condition is not met, it is regarded as a layer of clouds. The cloud layering processing process based on the ISODATA clustering algorithm is as follows:

[0051] (1) Construct a cloud layer differential zero point set. Construct a two-dimensional data point set x based on the relative humidity and vertical height detected by the lidar at each zero point. i ={(r i , bi ), i = 1, 2, ..., N}, where r i is the vertical height, b i is the relative humidity, which is the constructed cloud layer differential zero point set. c zero-point initial cluster centers {z1, z2, ...z Nc}, randomly select the initial position. Preselect K = 3 as the expected number of cluster centers, and set the minimum number of zero points in each cluster domain zero point set θ N = 2. The minimum distance between two cluster center zero points θ c =50, if it is less than this θ c , two clusters need to be merged. The maximum number of cluster centers that can be merged in one iteration is 2; set the number of iterations M = 30. Assign N pattern samples to the nearest cluster S j ,if:

[0052] D j =min{||xz i ||, i = 1, 2, ..., N c} (3)

[0053] That is, ||xz j || is the smallest distance, where j represents the jth cluster, x represents a single sample in a cluster, and z j Represented as the jth cluster S j The cluster center of x∈S j , if cluster S j The number of samples in S j <θ N , then cancel the sample subset, and the zero-point initial cluster center N c Minus 1.

[0054] Correct the zero point of each cluster center, where N j represents the jth cluster S j Number of samples:

[0055]

[0056] Calculate the zero point domain B of each cluster j The average distance between the median mode sample and each cluster center:

[0057]

[0058] Calculate the total average distance between all sample zero points and their corresponding cluster center zero points:

[0059]

[0060] (2) Discrimination and classification of multiple cloud layer differential zero point sets constructed. This part mainly includes splitting and merging. During the iterative discrimination process, if the number of zero-point initial cluster centers Nc≤K / 2, K is the expected number of cluster centers, that is, the number of cluster centers is less than or equal to half of the expected number of cluster centers, then the existing clusters are split; if the number of iterative operations is an even number, or Nc≥2K, then the merging process is performed until the maximum number of iterations is reached.

[0061] Splitting process (i): Calculate the standard deviation of the sample distances in the differential zero point set for each cloud layer:

[0062] σ j =(σ 1j , σ 2j , ..., σ nj ) T (7)

[0063] In the formula, σ j is the matrix of standard deviation of sample distances in the cloud layer differential zero point concentration; σ nj is the standard deviation of each specific sample distance in the cloud layer differential zero point set; the components of the vector are:

[0064]

[0065] In the formula, i = 1, 2, ..., n is the dimension of the sample feature vector, j = 1, 2, ..., N c is the number of clusters, N j is the jth cluster zero point set S j The number of samples in x ik is the i-th zero point in the k-th cluster among the expected cluster centers, z ij is the cluster center of the jth cluster among the pre-selected cluster centers.

[0066] Splitting process (ii): Find the standard deviation vector {σ j , j = 1, 2, ..., N c}, with {σ jmax , j = 1, 2, ..., N c}, for example, in any maximum component set {σ jmax , j = 1, 2, ..., N c}, if there is σ jmax >θ S , and satisfies N j >(θ N +1) or Then the zero point cluster center z j Split into two new zero-point cluster centers, the number of zero-point initial cluster centers is Nc Add 1. If the judgment condition is not met, it means that the cluster center z j Preselect two zero points on both sides as new cluster centers and remove the original zero point cluster center z j .

[0067] Merge processing (i): Calculate the distance between all cluster centers:

[0068] D ij =||z i -z j ||, i = 1, 2, ..., N c -1, j = i + 1, ..., N c (9)

[0069] Among them, z i is the zero-point cluster center of the ith cluster, z j is the zero-point cluster center of the j-th cluster;

[0070] Combined processing (ii): Compare D ij With θ c The value of D ij <θ c The values ​​of are arranged in ascending order according to the smallest distance, that is, {D i1j1 , D i2j2 , ..., D jLjL}, where D i1j1 <D i2j2 <...<D iLjL If the minimum distance between two cluster center zero points is θ c =50, if it is less than this θ c , the two clusters need to be merged; if D ij Greater than θ c , it means that the two clusters are independent of each other and do not need to be merged.

[0071] Merge process (iii): merge the distance D ikjk The two zero-point cluster centers Z in the k-th cluster i k and Z j k Merge, and the new zero-point cluster center is:

[0072]

[0073] In the formula, L is the maximum number of pairs of cluster centers that can be merged in one iteration. The two zero-point cluster center vectors to be merged are weighted by the number of samples in their cluster domains, so that the new cluster center z k * is the true mean vector.

[0074] (3) The cloud layer structure is determined based on the differential zero point sets of the classified multiple cloud layer signals to obtain the corresponding cloud layer information. The cloud layer structure is determined for each classified differential zero point set, and each differential zero point set is the differential zero point of the same cloud layer. The cloud base and cloud top heights of each cloud layer are the minimum cloud base height and the maximum cloud top height calculated by the differential zero crossing method in the differential zero point set. In this step of operation, the number of iterations M is increased by 1. At this point, all cloud layer differential zero points are clustered, and finally multiple classified differential zero point sets are obtained.

[0075] Then, the cloud structure is identified based on the classified differential zero point sets to obtain cloud information such as the cloud base, cloud top, and cloud peak position height of each cloud layer.

[0076] Therefore, the method of the present invention can effectively eliminate the interference of aerosol signals and accurately screen out cloud signals, thereby achieving accurate processing of cloud layers.

[0077] In order to verify the applicability of the method of the present invention, simulation experiments and real data tests were carried out respectively.

[0078] Verification Example 1:

[0079] This simulation experiment is as follows Figure 4 As shown, (a) is the simulated low cloud lidar echo signal; (b) is the simulated low cloud signal algorithm processing result; (c) is the simulated high cloud signal; (d) is the simulated high cloud signal algorithm processing result. The simulated low cloud base and cloud top heights are 120m and 644m respectively, and the simulated high cloud base and cloud top heights are 7654m and 8823m respectively. According to the experimental results, for clouds of different heights, the improved differential zero crossing method can effectively extract cloud signals and can well determine the cloud base and cloud top heights.

[0080] Verification Example 2:

[0081] The method proposed in the present invention is used to invert the measured lidar echo data profile. Taking the lidar measured data from the ARM website monitoring station Southern Great Plains (SGP), Lamont, Oklahoma (Lamont, Oklahoma, USA), as an example, the lidar echo signal profile with a monitoring time of 11:30 on April 19, 2021 is selected. It can be seen from the lidar echo signal graph that within the detectable range of the lidar, there are obvious cloud signals in the range of 2.5-4.5km and 4-6.5km. After amplifying the signal at an altitude of 7-10km, it can be found that there are weak cloud signals mixed in the noise signal between 8-10km. The lidar echo signal graph and the local amplified signal are shown in the figure. Figure 5As shown, the improved algorithm proposed in the present invention is used to process the signal, and the processed result and local amplification are shown in Figure 6 shown.

[0082] Figure 5 The cloud base and cloud top heights of the first layer of clouds are 3.12km and 3.94km, the cloud base and cloud top heights of the second layer of clouds are 4.47km and 6.13km, and the cloud base and cloud top heights of the third layer of clouds are 8.48km and 9.50km. For the echo data at the same time at the same station, the traditional differential zero crossing method is used to process the echo signal to obtain the cloud base and cloud top height data.

[0083] It can be seen from the cloud height data obtained by experimental inversion that the improved differential zero-crossing method of the present invention has better performance in the measurement of the cloud base and cloud top heights than the traditional differential zero-crossing method.

[0084] In order to further verify the superiority of this algorithm, the lidar measured data from April 19 to April 20, 2021 at the ARM website monitoring station Southern Great Plains (SGP), Lamont, Oklahoma (Lamont, Oklahoma, USA) were selected for analysis. The lidar echo signal covers a time span of 20 hours from 12:00 on April 19, 2021 to 08:00 on April 20, 2021, during which a total of 667 groups of signal profiles were measured. The detection results of the differential zero crossing method and this algorithm are shown as follows: Figure 7 As shown. Figure 7 In (a), intermittent weak cloud signals can be found in the A area. The breakpoints in the A area are caused by misjudgment. Figure 7 (b) The B block area and Figure 7 By comparing the corresponding areas in (a), we can find that Figure 7 In (a), the original algorithm missed the cloud because the number of points where the cloud signal dP / dr was greater than zero did not meet the threshold requirement. The improved algorithm correctly detected the cloud information in the area by combining the layer peak-to-layer bottom ratio threshold and the background noise threshold. Figure 7 The C block area in (b) is Figure 7By comparing the corresponding areas in (a), it can be found that the original algorithm has poor accuracy and it is difficult to accurately identify the continuously existing cloud information, so the cloud detection rate is low and the continuity is poor, resulting in missed selection and misjudgment. As can be seen from the figure, during the period of 02:00-06:00, the cloud detection results of the two algorithms maintain good consistency. Experiments have shown that the detection accuracy of the cloud layer by the improved algorithm proposed in the present invention is compared with the ARM cloud layer data, achieving a detection accuracy of 93.62% for low-level clouds, 92.78% for middle-level clouds, and 93.03% for high-level clouds.

[0085] In summary, the method of the present invention effectively eliminates the interference of aerosol signals and can accurately screen out cloud signals. At the same time, the clustering algorithm is applied to cloud structure analysis for the first time, achieving accurate processing of cloud layers.

[0086] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A dual-threshold lidar cloud detection algorithm based on cluster analysis, characterized in that: The following steps are involved: Step 1: preprocessing the laser radar echo signal; Step 2: Use double thresholds to process the preprocessed lidar echo signal: use the layer peak-to-layer bottom ratio threshold to screen the preprocessed lidar echo signal and extract the cloud layer signal; use the background noise threshold to extract the missed thin cloud layer signal from the preprocessed lidar echo signal; Step 3: Based on the extracted cloud layer signal and thin cloud layer signal, the differential zero crossing method is used to perform differential processing to obtain the differential zero point of the extracted cloud layer signal, and the cloud top and cloud bottom heights are calculated by inversion; Step 4: Use the ISODATA clustering algorithm to perform cluster analysis on the differential zero points of each cloud layer signal according to relative humidity and vertical height, and cluster samples with similar relative humidity characteristics within the same height range into one category, that is, each category of differential zero points can be regarded as the differential zero points of the same layer of cloud, thereby obtaining multiple classified differential zero point sets; the cloud layer structure is judged based on the classified multiple differential zero point sets to obtain the corresponding cloud layer information.

2. The dual-threshold lidar cloud detection algorithm based on cluster analysis according to claim 1 is characterized in that: The step 1 of preprocessing the laser radar echo signal includes: using the LMD-ITM algorithm to denoise the initial laser radar echo signal, and performing a 5-point sliding average process on the denoised signal.

3. The dual-threshold lidar cloud detection algorithm based on cluster analysis according to claim 1 is characterized in that: In step 2, the preprocessed lidar echo signal is screened using the layer peak to layer bottom ratio threshold, using the formula: Among them, X is the layer peak to layer bottom ratio threshold; r m 、r b are the positions of the cloud layer peak and bottom, respectively, P(r b )、P(r m ) are the laser radar echo signal intensities at the cloud layer peak and bottom, respectively.

4. The dual-threshold lidar cloud detection algorithm based on cluster analysis according to claim 1 is characterized in that: In step 2, the background noise threshold is used to extract the missed thin cloud layer signal from the preprocessed lidar echo signal, specifically: When the laser radar detection distance r<5km, set the background noise threshold Y=[P(r m )-P(r b )]>3I, where P(r b )、P(r m ) are the laser radar echo signal intensities at the cloud layer peak and bottom respectively; I is the background noise intensity; When the laser radar detection distance r≥5km, set the background noise threshold during the day as Y=[P(r m )-P(r b )]>1.5I, set the background noise threshold at night Y=[P(r m )-P(r b )]>25I.

5. The dual-threshold lidar cloud detection algorithm based on cluster analysis according to claim 1 is characterized in that: In step 4, the ISODATA clustering algorithm is used to perform cluster analysis on the differential zero points of each cloud layer signal according to relative humidity and vertical height, specifically including: Construct cloud layer differential zero point set; The constructed cloud layer differential zero point set is discriminated and classified, including splitting and merging. In the iterative discrimination process, if the number of zero-point initial cluster centers Nc≤K / 2, K is the expected number of cluster centers, that is, the number of zero-point initial cluster centers is less than or equal to half of the expected number of cluster centers, then the existing clusters are split; if the number of iterative operations is an even number, or Nc≥2K, then the merging process is performed until the maximum number of iterations is reached; Then each differential zero point set after classification is the differential zero point of the same cloud layer, and the cloud base and cloud top heights of each cloud layer are the minimum cloud base height and the maximum cloud top height calculated by the differential zero crossing method within the differential zero point set; in this step of calculation, the number of iterations is increased by 1, so far, the cloud layer differential zero point clustering is completed, and multiple differential zero point sets after classification are obtained.

Images