A classification method for full waveform laser echo signals based on SIFT visual word bag

Abstract

The present invention provides a full-waveform laser echo signal classification method based on SIFT visual word bag, comprising: pre-establishing SIFT visual word bag; controlling streak tube laser radar to scan the target area column by column, and collecting an original echo signal of each column scanning the target area; extracting all SIFT feature points of the original echo signal, and using the SIFT visual word bag to quantify all SIFT feature points to obtain a target feature vector; inputting the target feature vector into a support vector machine model in real time to classify the original echo signal. The method of the present invention directly classifies according to the local features of the original echo signal stripe pattern, which can simplify the classification process and improve the accuracy of classification.

Description

Technical Field

[0001] The present invention relates to the field of laser radar technology, and in particular to a full-waveform laser echo signal classification method based on SIFT visual word bag. Background Art

[0002] Compared with traditional microwave radar, laser radar has the characteristics of high precision, high resolution, high detection sensitivity, high confidentiality, small size, light weight and convenient for airborne and shipborne. In addition, due to the different working mechanism, the laser pulses emitted by laser radar have stronger anti-interference ability and richer data information than traditional microwave radar signals, which leads to higher detection and recognition capabilities.

[0003] Array lidar based on streak tubes has high imaging frame rate, full waveform sampling, high detection sensitivity and distance resolution, and has broad prospects in airborne radar mapping and military. At present, the rapid development of this new system radar has increased its data acquisition rate by orders of magnitude. However, in some cases, we are only interested in part of the data and do not need to process all the data within the measurement range. Take the forestry department as an example. It only cares about the vegetation information in the measurement area, while the information of buildings, roads, etc. belongs to interference information. When processing data, only useful information can be filtered out. This process requires data classification. Therefore, the processing of massive data has become one of the bottlenecks in the development of lidar. Summary of the invention

[0004] The purpose of the present invention is to provide a full-waveform laser echo signal classification method based on SIFT visual word bag, which can solve at least one of the above-mentioned technical problems. The specific scheme is as follows:

[0005] The present invention provides a full-waveform laser echo signal classification method based on SIFT visual word bag, comprising:

[0006] Pre-build SIFT visual word bag;

[0007] Control the streak tube laser radar to scan the target area column by column, and collect an original echo signal of each column of the scanned target area;

[0008] Extracting all SIFT feature points in the one original echo signal, and quantizing all SIFT feature points using the SIFT visual word bag to obtain a target feature vector of the one original echo signal;

[0009] The target feature vector of the original echo signal is input into a support vector machine model in real time to classify the original echo signal; wherein the support vector machine model is trained using multiple sets of training data, and the multiple sets of training data include a first type of training data and a second type of training data; each set of data in the first type of training data includes: a first echo signal of the target feature type, a label used to identify the first echo signal, and a first feature vector used to characterize the first echo signal; each set of data in the second type of training data includes: a second echo signal that does not belong to the target feature type, a label used to identify the second echo signal, and a second feature vector used to characterize the second echo signal.

[0010] Optionally, the pre-established SIFT visual word bag includes:

[0011] Select multiple echo signals of different ground object types, and extract SIFT feature points of each echo signal respectively;

[0012] The SIFT feature points of all echo signals are collected, and SIFT feature points with similar meanings are merged using the K-Means algorithm to construct a SIFT visual word bag containing K words.

[0013] Optionally, the extracting all SIFT feature points in the one original echo signal includes:

[0014] Detecting multiple extreme value points of the original echo signal in the Gaussian scale space;

[0015] Remove unstable points from the multiple extreme points to obtain key points;

[0016] determining a main direction of each of the key points;

[0017] Generate a descriptor of the key point to obtain a SIFT feature point.

[0018] Optionally, the using the SIFT visual word bag to quantize all the SIFT feature points to obtain a target feature vector includes:

[0019] Approximately replacing the SIFT feature points with words in the SIFT visual word bag;

[0020] The number of occurrences of each word in the SIFT feature point is counted to obtain a target feature vector of the one original echo signal, wherein the number of the target feature vectors is the number of different words that appear in the SIFT feature point.

[0021] Optionally, controlling the streak tube laser radar to scan the target area column by column and collecting an original echo signal of each column of the scanned target area includes:

[0022] Controlling the streak tube laser radar to scan the target area row by row along a direction perpendicular to the flight direction of the aircraft;

[0023] Receiving optical signals returned at different times by ground targets in each scanning target area, and converting the optical signals into photoelectron pulses in the streak tube;

[0024] The photoelectron pulses are linearly deflected so that the photoelectron pulses at different times are spread out on the fluorescent screen in a time sequence to form an original echo signal for each column of the scanning target area.

[0025] Optionally, the target feature vector is input into a support vector machine model in real time to classify the original echo signal; wherein the support vector machine model is trained using multiple sets of training data, and the multiple sets of training data include first-category training data and second-category training data, including:

[0026] Selecting a first echo signal of a target object type and a second echo signal that does not belong to the target object type as training data; assigning a first label and a second label to the first echo signal and the second echo signal respectively;

[0027] Extracting a first feature vector of the first echo signal based on the SIFT visual word bag, and merging the first feature vector with the first label to form a first feature set; and extracting a second feature vector of the second echo signal, and merging the second feature vector with the second label to form a second feature set;

[0028] Establishing a support vector machine model to be trained, and using the first feature set and the second feature set to train the support vector machine model to be trained, and using the training result to optimize the parameters of the support vector machine model to be trained;

[0029] The target feature vector of a pre-classified original echo signal is input into the trained support vector machine model in real time, and the classification type of the original echo signal is output.

[0030] Optionally, the first echo signal may be any one of a plain fringe image, a building fringe image, and a tree fringe image.

[0031] Optionally, when the first echo signal is a plain fringe pattern, the second echo signal is a non-plain fringe pattern, and the non-plain fringe pattern includes one or more of a building fringe pattern and a tree fringe pattern; or,

[0032] When the first echo signal is a building fringe pattern, the second echo signal is a non-building fringe pattern, and the non-building fringe pattern includes one or more of a plain fringe pattern and a tree fringe pattern; or,

[0033] When the first echo signal is a tree fringe pattern, the second echo signal is a non-tree fringe pattern, and the non-tree fringe pattern includes one or more of a building fringe pattern and a plain fringe pattern.

[0034] Optionally, the method further includes: if the classification type of the original echo signal is the target object type, storing the original echo signal in a target folder, wherein the target folder is used to store multiple echo signals of the same object type.

[0035] Optionally, the method further includes: performing point cloud inversion on the multiple original echo signals in the target folder to generate point cloud data or images of the target object type.

[0036] Compared with the prior art, the above solution of the embodiment of the present invention has at least the following beneficial effects:

[0037] The full-waveform laser echo signal classification method provided by the present invention performs full-waveform acquisition of the original echo signal of the ground target through an airborne laser radar of a streak tube, and directly classifies the original echo signal according to the local features of the streak image, thereby simplifying the classification process; a support vector machine model is used to classify the original echo signal, and the data amount based on each waveform echo signal can reach 1000 times that of a non-waveform sampling signal, thereby obtaining a large amount of information, which helps to improve the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings herein are incorporated into and constitute a part of the specification, showing embodiments consistent with the present invention, and together with the specification, are used to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0039] Figure 1 A flow chart of a full-waveform laser echo signal classification method provided by an embodiment of the present invention is shown;

[0040] Figure 2 A schematic diagram showing the imaging principle of a streak tube laser radar provided by an embodiment of the present invention is shown;

[0041] Figure 3a , 3b 3 and 3c are schematic diagrams showing original echo signals obtained in the embodiments of the present invention;

[0042] Figure 4 The scanning method of the streak tube laser radar for a rectangular area in an embodiment of the present invention is shown;

[0043] Figure 5 A flow chart of a method for extracting all SIFT feature points in the original echo signal in an embodiment of the present invention is shown;

[0044] Figure 6 A histogram of a visual word bag with a word bag size of 80 in an embodiment of the present invention is shown.

[0045] Figure 7 A flow chart of a method for training a support vector machine model in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0048] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0049] It should be understood that although the terms first, second, third, etc. may be used to describe the embodiments of the present invention, they should not be limited to these terms.

[0050] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0051] The echo signal image collected by the full waveform laser echo signal classification method provided by the present invention contains rich local features, and the echo signal image can be classified by extracting local features of different terrains. SIFT features have good robustness and are widely used in pattern recognition. Therefore, the present invention uses SIFT algorithm to extract SIFT features from echo signal images to achieve classification extraction.

[0052] The SIFT feature was proposed by Lowe in 2004. The SIFT feature maintains a certain stability to perspective, affine changes and noise, and has good invariance in rotation, scale scaling and brightness changes. These characteristics of SIFT make it a very stable local feature.

[0053] The optional embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0054] Figure 1 FIG. 4 is a flow chart showing a method for classifying a full-waveform laser echo signal provided by an embodiment of the present invention. Figure 1 As shown, the method includes:

[0055] S10, pre-build SIFT visual word bag;

[0056] This step specifically includes: selecting multiple echo signals of different types of objects, and extracting SIFT feature points of each echo signal respectively; wherein the echo signal can be selected according to actual needs, for example, to extract a river, the echo signal of the river is selected. In this embodiment, 1000 plain fringe images, 1000 building fringe images, and 1000 tree fringe images are selected for SIFT feature point extraction.

[0057] The SIFT feature points of all echo signals of different types of objects are collected, and the SIFT feature points with similar meanings are merged using the K-Means algorithm to construct a SIFT visual word bag containing K words. Specifically, the k-means algorithm is used to cluster all SIFT feature points into K clusters so that the similarity within the cluster is high and the similarity between clusters is low, thereby constructing a SIFT visual word bag model. The more perfect the SIFT visual word bag model is, the higher the accuracy of subsequent classification will be.

[0058] S20, controlling the streak tube laser radar to scan the target area column by column, and collecting an original echo signal of each column of the scanned target area;

[0059] In this embodiment, step S20 includes:

[0060] Controlling the streak tube laser radar to emit a scanning beam to a target area and to scan row by row in a direction perpendicular to the flight direction of the aircraft;

[0061] receiving optical signals returned at different times by ground targets in each scanning target area, and converting the optical signals into photoelectron pulses in the streak tube;

[0062] The photoelectron pulse generated by the laser echo is linearly deflected so that the photoelectrons at different times are spread out in time sequence on the fluorescent screen to form an original echo signal for each column of ground targets. There will be multiple ground targets in each column scanning area, and multiple light signals will be returned. The multiple light signals are converted into multiple photoelectrons to generate an original echo signal. In other words, each column scan in the target area will generate a scanned image, that is, an original echo signal.

[0063] Specifically, the detection principle of the streak tube laser radar is as follows:

[0064] like Figure 2 As shown, after the streak tube laser radar emits a laser pulse to the surface of an object, part of the light signal is returned; this part of the light signal is shaped into a linear beam by a slit, and then focused on the photocathode of the streak tube by a focusing lens; the photocathode undergoes a photoelectric effect, converting photons into photoelectrons and amplifying them; then, the photoelectron pulse enters the deflection system, which linearly deflects the photoelectrons returned at different times in a certain dimension, so that the photoelectrons at different times are expanded in time sequence on the dimension on the fluorescent screen, and a stripe image corresponding to each column of concave areas is obtained, as shown in FIG. Figures 3a to 3c The original echo signal is shown.

[0065] The target area is a pre-set ground area, which can be circular, sector-shaped, rectangular or other shapes, and is scanned in a preset manner with the streak tube laser radar as the center. Figure 4 The scanning method of the streak tube laser radar for a rectangular area is shown. The streak tube laser radar is pre-set to scan in a direction perpendicular to the flight direction of the aircraft. Combined with the movement of the aircraft, large-scale mapping can be achieved.

[0066] The present invention utilizes a stripe lidar to collect the full waveform of the echo signal. The data volume of the full waveform echo signal of each ground target can reach 1000 times that of the non-waveform sampling signal, and about 30G of data can be generated in 1 minute, thereby obtaining more characteristic details of the ground target, making the acquired ground features richer and facilitating more accurate classification.

[0067] S30, extracting all SIFT feature points of the one original echo signal, and quantizing all SIFT feature points using the SIFT visual word bag to obtain a target feature vector;

[0068] like Figure 5 As shown, the extracting of all SIFT feature points in the one original echo signal includes:

[0069] S31, detecting multiple extreme value points of the original echo signal in the Gaussian scale space;

[0070] Since images have feature points at different scales, we choose to perform preliminary detection of extreme points in Gaussian scale space. The Gaussian scale space of an image can be obtained by convolving it with a Gaussian kernel:

[0071] L(x,y,σ)=G(x,y,σ)×I(x,y);

[0072] Wherein, I(x, y) represents the pixel position of the image, L(x, y, σ) represents the Gaussian scale space of the image, and G(x, y, σ) is the Gaussian kernel function, which is as follows:

[0073]

[0074] Among them, σ is the scale space factor. The larger its value, the blurrier the image and the larger the corresponding scale. After Gaussian blurring of different scales, the image is downsampled to construct a Gaussian pyramid. After the Gaussian pyramid is constructed, the adjacent Gaussian pyramids are subtracted to obtain the Gaussian difference pyramid (Difference of Gaussian, DOG). DOG is defined as:

[0075] D(x,y,σ)=[G(x,y,κσ)-G(x,y,σ)]×I(x,y)=L(x,y,κσ)-L(x,y,σ);

[0076] Among them, k is the scale factor of the adjacent scale space; when a pixel point is larger (or smaller) than all the adjacent points in its image domain and scale domain, the point is an extreme point.

[0077] S32, removing unstable extreme value points from the multiple extreme value points to obtain key points;

[0078] The local extreme points of DOG are obtained by searching in discrete space. There are many extreme points that are not in the true sense. Therefore, we can find the extreme points by curve fitting through the scale space DOG function. The essence of this step is to remove the points where the local curvature of DoG is very asymmetric. Specifically, the Gaussian difference function is Taylor expanded, and the formula is as follows:

[0079]

[0080] The extreme points are determined by this formula, and the extreme points with absolute values ​​less than 0.03 are discarded. In order to enhance the stability and anti-noise performance of key points, two-dimensional Hessian is used to remove key points with low contrast and unstable edge influence points.

[0081] S33, determining the main direction of each key point;

[0082] Specifically, by calculating the modulus and direction of the gradient of the pixels in the extreme point area, a reference direction is assigned to each extreme point to make the descriptor rotationally invariant. The modulus formula of the gradient is as follows:

[0083]

[0084] Among them, m(x, y) refers to the modulus of the gradient at the pixel (x, y), and L represents the scale of the SIFT feature point.

[0085] The direction formula of the gradient is as follows:

[0086]

[0087] Where θ(x, y) is the direction at pixel (x, y).

[0088] S34, generating a descriptor of the key point to obtain a SIFT feature point. The main idea of ​​the SIFT feature point descriptor is to describe the central pixel using the gradient direction of the neighboring pixels.

[0089] Since the number of SIFT features extracted from each image is inconsistent, it will affect the use of the support vector machine model and the classification accuracy. Therefore, for any image, the basic elements in the image can be extracted through the SIFT visual word bag model, and the frequency of occurrence of these basic elements in the image can be counted. Not only can the statistical results be used as input information of the support vector machine model, but the SIFT visual word bag model can also be used to improve the classification accuracy. Specifically, the use of the SIFT visual word bag to quantize all the SIFT feature points to obtain the target feature vector includes:

[0090] Approximately replacing the SIFT feature points with words in the SIFT visual word bag;

[0091] The number of occurrences of each word in the SIFT feature point is counted to obtain multiple target feature vectors of the original echo signal, wherein the number of the target feature vectors is the number of different words that appear in the SIFT feature point. In this embodiment, the multiple target feature vectors are represented in the form of a histogram. For example, Figure 6 A histogram of the bag-of-visual-words with a bag-of-words size of 80 is shown.

[0092] S40, inputting the target feature vector of the original echo signal into a support vector machine model in real time to classify the original echo signal; wherein the support vector machine model is trained using multiple sets of training data, and the multiple sets of training data include first-class training data and second-class training data;

[0093] Each group of data in the first category of training data includes: a first echo signal of the target object type, a label used to identify the first echo signal, and a first eigenvector used to characterize the first echo signal; each group of data in the second category of training data includes: a second echo signal that does not belong to the target object type, a label used to identify the second echo signal, and a second eigenvector used to characterize the second echo signal.

[0094] Due to the confidentiality of the technology, commercial single-point scanning laser radar does not provide the original laser echo signal, but only provides processed point cloud data, which makes the existing classification technology limited to the stage based on point cloud data. Since the original point cloud data is a large number of irregularly distributed three-dimensional discrete points, there is no morphological relationship between each laser foot point. Therefore, it is necessary to filter and remove the ground point cloud before classification, and then analyze the ground features for classification; however, the laser point cloud data only contains a small amount of information such as position information, reflection intensity information and echo number information. During filtering and feature analysis, it is difficult to achieve accurate classification with a single feature. The present invention uses the support vector machine model to directly classify the original echo signal without converting it into point cloud data, which simplifies the classification process.

[0095] Support vector machine is a binary classification model based on the maximum interval of sample points. It is a linear machine learning method proposed after a lot of research on the basis of statistics. Nonlinear classification can be performed through kernel functions. In support vector machines, the classification performance of different hyperplanes varies greatly. The goal of support vector machines is to find a hyperplane with the best classification generalization performance. The hyperplane is found by adopting the idea of ​​the maximum classification interval, that is, the classification surface is required to correctly separate the two types of samples while ensuring the maximum classification interval. In the present invention, a support vector machine model (C-SVC) with a penalty factor C is selected. In the classifier, we select the radial basis kernel function (RBF) and the ovr decision function, and select different penalty factors for analysis; then each original echo signal obtained is input into the support vector machine model in real time for classification, and there is no need to wait until all the original echo signals in the target area are received before classification, saving the time cost of the classification process.

[0096] In this embodiment, Figure 7 As shown, the training process of the support vector machine is as follows:

[0097] S41, selecting a first echo signal of a target object type and a second echo signal that does not belong to the target object type as training data; assigning a first label and a second label to the first echo signal and the second echo signal respectively;

[0098] In this step, a ground area of ​​a known target object type may be selected for scanning to obtain the first echo signal, and other ground areas that do not belong to the target object type may be selected for scanning to obtain the second echo signal.

[0099] The first echo signal may be any fringe image of a plain fringe image, a building fringe image, or a tree fringe image. Of course, the fringe images listed are not limited to the fringe images listed above. Any existing fringe image may be used for training, and the selection is made according to actual needs. For example, if a river image needs to be extracted, a river fringe image may be selected for training.

[0100] When the first echo signal is a plain fringe pattern, the second echo signal is a non-plain fringe pattern, which may be one or more of a building fringe pattern and a tree fringe pattern; or, when the first echo signal is a building fringe pattern, the second echo signal is a non-building fringe pattern, which may be one or more of a plain fringe pattern and a tree fringe pattern; or, when the first echo signal is a tree fringe pattern, the second echo signal is a non-tree fringe pattern, which may be one or more of a building fringe pattern and a plain fringe pattern.

[0101] The first label and the second label can be numbers, letters or other symbols, as long as they can distinguish different types of land objects. For example, if the first echo signal is selected as a plain fringe pattern, this type of echo signal is encoded as 1, and the second echo signal is selected as a building fringe pattern and a tree fringe pattern, and these two types of fringe patterns are encoded as 2 and 3, respectively, to represent plains, buildings and trees;

[0102] S42, extracting a first feature vector of the first echo signal based on the SIFT visual word bag, and merging the first feature vector with the first label to form a first feature set; and extracting a second feature vector of the second echo signal, and merging the second feature vector with the second label to form a second feature set;

[0103] It should be noted that the method for extracting the first feature vector and the second feature vector mentioned in this step is the same as the method for extracting the target feature vector of the original echo signal, and both are extracted through SIFT visual word bag.

[0104] S43, establishing a support vector machine model to be trained, and using the first feature set and the second feature set to train the support vector machine model to be trained, and using the training result to optimize the parameters of the support vector machine model to be trained;

[0105] S44, inputting a target feature vector of a pre-classified original echo signal into the trained support vector machine model in real time, and outputting the classification type of the original echo signal.

[0106] As another optional implementation, the method further includes:

[0107] S50: If the classification type of the original echo signal is the target object type, the original echo signal is stored in a target folder, where the target folder is used to store multiple echo signals of the same object type.

[0108] In this embodiment, a plurality of different folders may be pre-established in the terminal, each folder being used to store echo signals of a type of ground object. When the output original echo signal belongs to a certain ground object type, the echo signal is stored in the corresponding folder.

[0109] As another optional implementation, the method further includes:

[0110] S60, performing point cloud inversion on the multiple original echo signals in the target folder to generate point cloud data or images of the type of features corresponding to the folder. For example, if you want to generate an image of a building in the target area, you can directly perform point cloud inversion by selecting the target folder storing the building, which is convenient for users to quickly image.

[0111] A full-waveform laser echo signal classification method provided by an embodiment of the present invention performs full-waveform sampling of the original echo signal generated by a ground target through an airborne laser radar of a streak tube, and directly classifies the original echo signal according to the local features of the fringe pattern, thereby reducing the tedious process of converting all the original echo signals into a point cloud data map; in addition, the data volume of each echo signal based on waveform sampling can reach 1000 times the data volume of non-waveform sampling, and a large amount of characteristic information of the ground target is obtained, thereby achieving more accurate classification and making the classification accuracy reach more than 95%.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A full-waveform laser echo signal classification method based on SIFT visual word bag, characterized in that: include: Pre-build SIFT visual word bag; Control the streak tube laser radar to scan the target area column by column, and collect an original echo signal of each column of the scanned target area; Extracting all SIFT feature points in the one original echo signal, and quantizing all SIFT feature points using the SIFT visual word bag to obtain a target feature vector of the one original echo signal; Inputting the target feature vector of the one original echo signal into a support vector machine model in real time to classify the original echo signal; wherein the support vector machine model is trained using multiple sets of training data, and the multiple sets of training data include first-category training data and second-category training data; each set of data in the first-category training data includes: a first echo signal of the target feature type, a label used to identify the first echo signal, and a first feature vector used to characterize the first echo signal; each set of data in the second-category training data includes: a second echo signal that does not belong to the target feature type, a label used to identify the second echo signal, and a second feature vector used to characterize the second echo signal; The pre-established SIFT visual word bag includes: Select multiple echo signals of different ground object types, and extract SIFT feature points of each echo signal respectively; The SIFT feature points of all echo signals are collected, and SIFT feature points with similar meanings are merged using the K-Means algorithm to construct a SIFT visual word bag containing K words; The extracting all SIFT feature points in the original echo signal includes: Detecting multiple extreme value points of the original echo signal in the Gaussian scale space; Remove unstable extreme points from the multiple extreme points to obtain key points; determining a main direction of each of the key points; Generate a descriptor of the key point to obtain a SIFT feature point; The step of quantizing all SIFT feature points using the SIFT visual word bag to obtain a target feature vector includes: Approximately replacing the SIFT feature points with words in the SIFT visual word bag; Counting the number of times each word appears in the SIFT feature point to obtain a target feature vector of the original echo signal, wherein the number of the target feature vectors is the number of different words that appear in the SIFT feature point; The streak tube laser radar is controlled to scan the target area column by column and collect an original echo signal of each column of the scanned target area, including: Controlling the streak tube laser radar to scan the target area row by row along a direction perpendicular to the flight direction of the aircraft; Receiving optical signals returned at different times by ground targets in each scanning target area, and converting the optical signals into photoelectron pulses in the streak tube; Linearly deflecting the photoelectron pulses, so that the photoelectron pulses at different times are spread out on the fluorescent screen in time sequence, forming an original echo signal for each column of the scanned target area; The target feature vector of an original echo signal is input into a support vector machine model in real time to classify the original echo signal; wherein the support vector machine model is trained using multiple sets of training data, and the multiple sets of training data include first-class training data and second-class training data, including: Selecting a first echo signal of a target object type and a second echo signal that does not belong to the target object type as training data; assigning a first label and a second label to the first echo signal and the second echo signal respectively; Extracting a first feature vector of the first echo signal based on the SIFT visual word bag, and merging the first feature vector with the first label to form a first feature set; and extracting a second feature vector of the second echo signal, and merging the second feature vector with the second label to form a second feature set; Establishing a support vector machine model to be trained, and using the first feature set and the second feature set to train the support vector machine model to be trained, and using the training result to optimize the parameters of the support vector machine model to be trained; Inputting a target feature vector of a pre-classified original echo signal into a trained support vector machine model in real time, and outputting the classification type of the original echo signal; The first echo signal may be any one of a plain fringe image, a building fringe image, and a tree fringe image; When the first echo signal is a plain fringe pattern, the second echo signal is a non-plain fringe pattern, and the non-plain fringe pattern includes one or more of a building fringe pattern and a tree fringe pattern; or, When the first echo signal is a building fringe pattern, the second echo signal is a non-building fringe pattern, and the non-building fringe pattern includes one or more of a plain fringe pattern and a tree fringe pattern; or, When the first echo signal is a tree fringe pattern, the second echo signal is a non-tree fringe pattern, and the non-tree fringe pattern includes one or more of a building fringe pattern and a plain fringe pattern; The method further comprises: If the classification type of the original echo signal is the target object type, the original echo signal is stored in a target folder, and the target folder is used to store multiple echo signals of the same object type; The method further comprises: Point cloud inversion is performed on the multiple original echo signals in the target folder to generate point cloud data or images of the target object type.

Images