CN113567953A - Full-waveform laser echo signal classification method based on SIFT visual word bag - Google Patents
Full-waveform laser echo signal classification method based on SIFT visual word bag Download PDFInfo
- Publication number
- CN113567953A CN113567953A CN202110859165.4A CN202110859165A CN113567953A CN 113567953 A CN113567953 A CN 113567953A CN 202110859165 A CN202110859165 A CN 202110859165A CN 113567953 A CN113567953 A CN 113567953A
- Authority
- CN
- China
- Prior art keywords
- echo signal
- sift
- target
- feature
- fringe pattern
- 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
- 238000000034 method Methods 0.000 title claims abstract description 41
- 230000000007 visual effect Effects 0.000 title claims abstract description 36
- 239000013598 vector Substances 0.000 claims abstract description 44
- 238000012706 support-vector machine Methods 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims description 39
- 230000003287 optical effect Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 abstract description 8
- 238000005070 sampling Methods 0.000 description 7
- 238000001514 detection method Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 241000282320 Panthera leo Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- 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
- G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Optical Radar Systems And Details Thereof (AREA)
Abstract
The invention provides a full-waveform laser echo signal classification method based on SIFT visual word bag, which comprises the following steps: an SIFT visual word bag is established in advance; controlling a streak tube laser radar to scan a target area row by row, and collecting an original echo signal of each row of the scanned target area; extracting all SIFT feature points of the original echo signal, and quantizing all SIFT feature points by using the SIFT visual word bag to obtain a target feature vector; and inputting the target characteristic vector into a support vector machine model in real time, and carrying out ground feature classification on the original echo signal. The method of the invention directly classifies according to the local characteristics of the original echo signal fringe pattern, can simplify the classification process and improve the classification accuracy.
Description
Technical Field
The invention relates to the technical field of laser radars, in particular to a full-waveform laser echo signal classification method based on SIFT visual word bag.
Background
Compared with the traditional microwave radar, the laser radar has the characteristics of high precision, high resolution, high detection sensitivity, high confidentiality, small volume, light weight and convenience for airborne and shipborne. In addition, due to different working mechanisms, compared with a traditional microwave radar signal, a laser pulse emitted by the laser radar has stronger anti-interference capability and richer acquired data information, so that the laser radar has higher detection and identification capabilities.
The array laser radar based on the streak tube has the advantages of high imaging frame frequency, full waveform sampling, high detection sensitivity and range resolution, and has wide prospects in the aspects of airborne radar mapping and military. Currently, the rapid development of the radar with the new system has enabled the data acquisition rate to be improved by orders of magnitude. However, in some cases, we are only interested in a portion of the data and do not need to process all the data in the measurement range. Taking the forestry department as an example, the forestry department only cares about vegetation information in a measurement area, information such as buildings, roads and the like belongs to interference information, useful information can be only screened and processed when data are processed, and the data are required to be classified in the process, so that the processing of mass data becomes one of bottlenecks in the development of laser radars.
Disclosure of Invention
The invention aims to provide a full-waveform laser echo signal classification method based on SIFT visual word bag, which can solve at least one technical problem mentioned above. The specific scheme is as follows:
the invention provides a full-waveform laser echo signal classification method based on SIFT visual word bag, which comprises the following steps:
an SIFT visual word bag is established in advance;
controlling a streak tube laser radar to scan a target area row by row, and collecting an original echo signal of each row of the scanned target area;
extracting all SIFT feature points in the original echo signal, and quantizing all SIFT feature points by using the SIFT visual word bag to obtain a target feature vector of the original echo signal;
inputting the target characteristic vector of the original echo signal into a support vector machine model in real time, and classifying the ground features of the original echo signal; the support vector machine model is trained by using a plurality of groups of training data, wherein the plurality of groups of training data comprise first type training data and second type training data; each set of data in the first class of training data comprises: a first echo signal of a target ground object type, a tag used for identifying the first echo signal and a first feature vector used for characterizing the first echo signal; each set of data in the second class of training data includes: a second echo signal that does not belong to the target surface feature type, a tag to identify the second echo signal, and a second feature vector to characterize the second echo signal.
Optionally, the pre-established SIFT visual bag of words includes:
selecting a plurality of echo signals of different ground object types, and respectively extracting SIFT feature points of each echo signal;
and (3) collecting the SIFT feature points of all the echo signals, merging SIFT feature points with similar word senses by using a K-Means algorithm, and constructing an SIFT visual word bag containing K words.
Optionally, the extracting all SIFT feature points in the original echo signal includes:
detecting a plurality of extreme points of the original echo signal in a Gaussian scale space;
removing unstable points from the plurality of extreme points to obtain key points;
determining a principal direction of each of the keypoints;
and generating descriptors of the key points to obtain SIFT feature points.
Optionally, the quantizing all the SIFT feature points by using the SIFT visual word bag to obtain a target feature vector, including:
replacing the SIFT feature points with word approximations in the SIFT visual word bag;
and counting the occurrence times of each word in the SIFT feature points 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 in the SIFT feature points.
Optionally, the controlling the streak tube laser radar to scan the target area column by column and acquire an original echo signal of each column of the scanned target area includes:
controlling the streak tube laser radar to scan a target area column by column along a flight direction perpendicular to the airplane;
receiving optical signals returned by ground targets in each row of scanning target areas at different moments, and converting the optical signals into photoelectron pulses in the streak tube;
and linearly deflecting the photoelectron pulses to spread the photoelectron pulses at different moments on a fluorescent screen according to a time sequence so as to form an original echo signal of each column of the scanning target area.
Optionally, the target feature vector is input into a support vector machine model in real time, and the original echo signal is subjected to ground feature classification; the support vector machine model is trained by using a plurality of groups of training data, wherein the plurality of groups of training data comprise a first class of training data and a second class of training data, and the method comprises the following steps:
selecting a first echo signal of a target ground object type and a second echo signal which does not belong to the target ground object type as training data; respectively allocating a first label and a second label to the first echo signal and the second echo signal;
extracting a first feature vector of the first echo signal based on the SIFT visual word bag, and combining the first feature vector and the first label to form a first feature set; extracting a second feature vector of the second echo signal, and combining the second feature vector and a second label to form a second feature set;
establishing a support vector machine model to be trained, training the support vector machine model to be trained by utilizing the first feature set and the second feature set, and optimizing parameters of the support vector machine model to be trained by utilizing a training result;
and inputting a pre-classified target characteristic vector of an original echo signal into a trained support vector machine model in real time, and outputting the classification type of the original echo signal.
Optionally, the first echo signal may be any one of a stripe image in a plain stripe image, a building stripe image, and a tree stripe image.
Optionally, when the first echo signal is a plain fringe pattern, the second echo signal is a non-plain fringe pattern, where the non-plain fringe pattern includes one or more of a building fringe pattern and a tree fringe pattern; alternatively, the first and second electrodes may be,
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 comprises one or more of a plain fringe pattern and a tree fringe pattern; alternatively, the first and second electrodes may be,
and 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 comprises one or more of a building fringe pattern and a plain fringe pattern.
Optionally, the method further includes: and if the classification type of the original echo signal is the target ground object type, storing the original echo signal in a target folder, wherein the target folder is used for storing a plurality of echo signals of the same ground object type.
Optionally, the method further includes: and performing point cloud inversion on the plurality of original echo signals in the target folder to generate point cloud data or images of the target ground object type.
Compared with the prior art, the scheme of the embodiment of the invention at least has the following beneficial effects:
the full-waveform laser echo signal classification method provided by the invention has the advantages that the full-waveform acquisition is carried out on the original echo signal of the ground target through the airborne laser radar of the streak tube, and the classification is directly carried out according to the local characteristics of the streak image of the original echo signal, so that the classification process is simplified; and classifying the original echo signals by adopting a support vector machine model, wherein the data volume based on each waveform echo signal can reach 1000 times of that of a non-waveform sampling signal, so that a large amount of information is acquired, and the accuracy rate of classification is improved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort. In the drawings:
fig. 1 is a flowchart illustrating a full waveform laser echo signal classification method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating the imaging principle of a streak tube lidar according to an embodiment of the present invention;
3a, 3b, 3c respectively show schematic diagrams of original echo signals acquired in an embodiment of the present invention;
FIG. 4 illustrates scanning of a rectangular area by the streak tube lidar in accordance with embodiments of the present invention;
fig. 5 is a flowchart illustrating a method for extracting all SIFT feature points in the original echo signal according to an embodiment of the present invention;
fig. 6 shows a visual bag-of-words histogram with a bag-of-words size of 80 in an embodiment of the invention.
FIG. 7 is a flow chart of a method of training a support vector machine model according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail with reference to the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the examples of the present invention and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, and "a plurality" typically includes at least two.
It should be understood that the term "and/or" as used herein is merely one type of association that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.
It should be understood that although the terms first, second, third, etc. may be used to describe embodiments of the present invention, they should not be limited to these terms.
It is also noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that an article or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such article or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in the article or device in which the element is included.
The echo signal images acquired by the full-waveform laser echo signal classification method provided by the invention contain abundant local features, and the echo signal images can be classified by extracting the local features of different terrains. The SIFT features have good robustness and are widely used in the aspect of pattern recognition, so that the SIFT algorithm is adopted to extract the SIFT features from the echo signal image to realize classification extraction.
The SIFT features are proposed by Lowe in 2004, maintain certain stability to view angle, affine change and noise, and have good invariance in rotation, scale scaling and brightness change, and the characteristics of SIFT make the SIFT features become very stable local features.
Alternative embodiments of the present invention are described in detail below with reference to the accompanying drawings.
Fig. 1 shows a flowchart of a full waveform laser echo signal classification method according to an embodiment of the present invention. As shown in fig. 1, the method includes:
s10, establishing SIFT visual word bag in advance;
the method specifically comprises the following steps: selecting a plurality of echo signals of different ground object types, and respectively extracting SIFT feature points of each echo signal; the echo signal can be selected according to actual requirements, for example, a river is extracted, and then the echo signal of the river is selected. In this embodiment, 1000 plain fringe patterns, 1000 building fringe patterns and 1000 tree fringe patterns are respectively selected to perform SIFT feature point extraction.
And (3) gathering the SIFT feature points of all echo signals of different ground object types, combining the SIFT feature points with similar word senses by using a K-Means algorithm, and constructing an SIFT visual word bag containing K words. Specifically, all SIFT feature points are clustered into K clusters by using a K-means algorithm, so that the clusters have higher similarity and lower inter-cluster similarity, and an SIFT visual bag-of-words model is constructed. The more perfect the SIFT visual word bag model is established, the higher the accuracy of subsequent classification.
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 scanning target area;
in this embodiment, step S20 includes:
controlling the streak tube laser radar to emit scanning beams to a target area, and scanning column by column along a flight direction perpendicular to an airplane;
receiving optical signals returned by ground targets in each row of scanning target areas at different moments, and converting the optical signals into photoelectron pulses in a streak tube;
and linearly deflecting the photoelectron pulse generated by the laser echo to spread photoelectrons at different moments on a fluorescent screen according to a time sequence to form an original echo signal of each column of ground targets. In each column of scanning area, there will be a plurality of ground targets, and a plurality of optical signals will be returned, and these optical signals are converted into a plurality of photoelectron arrangements to generate an original echo signal, that is, each column of scanning in the target area will generate a scanning image, i.e. the original echo signal.
Specifically, the detection principle of the laser radar detection of the streak tube is as follows:
as shown in fig. 2, after the laser pulse is emitted to the surface of the object by the streak tube lidar, part of the optical signal is returned; the partial optical signals are shaped into linear beams through the slits and then focused on a photocathode of the streak tube by a focusing lens; the photocathode generates a photoelectric effect, and photons are converted into photoelectrons and amplified; then, the photoelectron pulse enters a deflection system, and the deflection system linearly deflects the photoelectrons returned at different moments in a certain dimension, so that the photoelectrons at different moments are spread on a fluorescent screen in the dimension according to a time sequence, and a fringe image corresponding to each row of concave regions is obtained, such as the original echo signals shown in fig. 3a to 3 c.
The target area is a preset ground area, can be a circle, a sector, a rectangle or other shapes, and is an area which is scanned by taking the streak tube laser radar as a center according to a preset mode. In this embodiment, fig. 4 shows a scanning manner of the streak tube laser radar to a rectangular region, and the streak tube laser radar is preset to scan in a direction perpendicular to a flight direction of the aircraft, so that large-scale mapping can be realized by combining with the motion of the aircraft.
The invention utilizes the stripe laser radar to carry out full-waveform acquisition on the echo signals, the data volume of the full-waveform echo signals of each ground target can reach 1000 times of that of non-waveform sampling signals, and about 30G of data volume can be generated in 1 minute, so that more characteristic details of the ground targets are obtained, the obtained ground characteristics are richer, and the classification is more accurately facilitated.
S30, extracting all SIFT feature points of the original echo signal, and quantizing all SIFT feature points by using the SIFT visual word bag to obtain a target feature vector;
as shown in fig. 5, the extracting all SIFT feature points in the original echo signal includes:
s31, detecting a plurality of extreme points of the original echo signal in a Gaussian scale space;
because the image has characteristic points under different scales, the initial detection of extreme points is carried out in a Gaussian scale space. The gaussian scale space of an image can be obtained by its convolution with a gaussian kernel:
L(x,y,σ)＝G(x,y,σ)×I(x,y)；
where 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 a gaussian kernel function as follows:
wherein, σ is a scale space factor, and the larger the value of σ is, the more blurred the image is, the larger the corresponding scale is. Performing Gaussian blur of different scales on the image, and then performing down-sampling to construct a Gaussian pyramid; after the Gaussian pyramid is constructed, the adjacent Gaussian pyramids are subtracted to obtain a Difference of Gaussian (DOG). DOG is defined as:
D(x,y,σ)＝[G(x,y,κσ)-G(x,y,σ)]×I(x,y)＝L(x,y,κσ)-L(x,y,σ)；
wherein k is a scale factor of adjacent scale spaces; when a pixel point is larger (or smaller) than all the adjacent points of the image domain and the scale domain, the point is an extreme point.
S32, removing unstable extreme points from the extreme points to obtain key points;
local extreme points of the DOG are obtained by searching in a discrete space, and a plurality of extreme points which are not in a true sense exist, so that the extreme points can be found by performing curve fitting on a DOG function in a scale space, and the essence of the step is to remove points with very asymmetric local curvature of the DOG. Specifically, taylor expansion is performed on the gaussian difference function, and the formula is as follows:
extreme points are determined through the formula, extreme points with absolute values smaller than 0.03 are abandoned, and in order to enhance the stability and the anti-noise performance of key points, two-dimensional Hessian is adopted to remove key points with low contrast and unstable edge influence points.
S33, determining the main direction of each key point;
specifically, a reference direction is allocated to each extreme point by calculating the modulus and the direction of the gradient of the pixels in the extreme point field, so that the descriptor has rotational invariance, and the modulus formula of the gradient is as follows:
where m (x, y) refers to a modulus of the gradient at pixel (x, y), and L represents a scale at which the SIFT feature point is located.
The direction formula of the gradient is as follows:
where θ (x, y) is the direction at pixel (x, y).
And S34, generating descriptors of the key points to obtain SIFT feature points. The main idea of the SIFT feature point descriptor is to describe the central pixel by using the gradient direction of the neighborhood pixels.
Because the number of extracted SIFT features of each image is inconsistent, the use and classification accuracy of the support vector machine model can be influenced, for any image, the basic elements in the image can be extracted through the SIFT visual bag-of-words model, the occurrence frequency of the basic elements in the image is counted, the counted result can be used as the input information of the support vector machine model, and the classification accuracy can be improved through the SIFT visual bag-of-words model. Specifically, the quantizing all the SIFT feature points by using the SIFT visual word bag to obtain a target feature vector includes:
replacing the SIFT feature points with word approximations in the SIFT visual word bag;
and counting the occurrence times of each word in the SIFT feature points to obtain a plurality of target feature vectors of the original echo signal, wherein the number of the target feature vectors is the number of different words appearing in the SIFT feature points. In the present embodiment, the plurality of target feature vectors are characterized in the form of a histogram, and fig. 6 shows a visual bag-of-words histogram with a bag-of-words size of 80 as an example.
S40, inputting the target feature vector of the original echo signal into a support vector machine model in real time, and classifying the ground features of the original echo signal; the support vector machine model is trained by using a plurality of groups of training data, wherein the plurality of groups of training data comprise first type training data and second type training data;
each set of data in the first class of training data comprises: a first echo signal of a target ground object type, a tag used for identifying the first echo signal and a first feature vector used for characterizing the first echo signal; each set of data in the second class of training data includes: a second echo signal that does not belong to the target surface feature type, a tag to identify the second echo signal, and a second feature vector to characterize the second echo signal.
Due to the technical secrecy reason, the commercial single-point scanning laser radar does not provide original laser echo signals, and only provides well-processed point cloud data, so that the existing classification technology is only limited to the stage based on the point cloud data. Because the original point cloud data is a large number of three-dimensional discrete points which are irregularly distributed, and each laser foot point has no morphological relation, filtering is needed to remove ground point cloud before classification, and then the ground feature characteristics are analyzed for classification; however, the laser point cloud data only contains a small amount of information such as position information, reflection intensity information, echo frequency information and the like, and during filtering and feature analysis, accurate classification of single features is difficult to achieve. The invention directly classifies the original echo signals by using the support vector machine model without converting the original echo signals into point cloud data, thereby simplifying the classification process.
The support vector machine is a binary model based on the maximum sample point interval, and is a linear machine learning method which is provided through a large amount of researches on the basis of statistics. Nonlinear classification can be performed again through the kernel function. In the support vector machine, the classification performance difference of different hyperplanes is extremely large, the support vector machine aims to search the hyperplane with the optimal classification generalization performance, and the hyperplane is searched by adopting the idea of the maximum classification interval, namely the classification interval is ensured to be maximum while the classification plane is required to correctly separate two types of samples. In the invention, a support vector machine model (C-SVC) with a penalty factor C is selected, a radial basis kernel function (RBF) and an ovr decision function are selected in a classifier, and different penalty factors are selected for analysis; and then, each acquired original echo signal is input into the support vector machine model in real time for classification, and classification is not required to be carried out after all the original echo signals in the target area are received, so that the time cost in the classification process is saved.
In this embodiment, as shown in fig. 7, the training process of the support vector machine is as follows:
s41, selecting a first echo signal of a target ground object type and a second echo signal which does not belong to the target ground object type as training data; respectively allocating a first label and a second label to the first echo signal and the second echo signal;
in this step, a ground area of a known target ground object type may be selected for scanning to obtain the first echo signal, and another ground area not belonging to the target ground object type may be selected for scanning to obtain the second echo signal.
The first echo signal can be any one of a stripe image in a plain stripe image, a building stripe image and a tree stripe image. Of course, the method is not limited to the illustrated stripe images, and any existing stripe images can be used for training, specifically, the selection is performed according to actual requirements, for example, if a river image needs to be extracted, a river stripe image can be selected for training.
When the first echo signal is a plain fringe pattern, the second echo signal is a non-plain fringe pattern, and specifically can 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, and may specifically 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, and may specifically be one or more of a building fringe pattern and a plain fringe pattern.
The first label and the second label can be numbers, letters or other symbols as long as different ground object types are distinguished. For example, if the first echo signal is selected as a plain fringe pattern, the echo signal is coded as 1, the second echo signal is selected as a building fringe pattern and a tree fringe pattern, and the two kinds of fringe patterns are respectively coded as 2 and 3 to respectively represent the plain, the building and the tree;
s42, extracting a first feature vector of the first echo signal based on SIFT visual word bag, and combining the first feature vector and the first label to form a first feature set; extracting a second feature vector of the second echo signal, and combining the second feature vector and a second label to form a second feature set;
it should be noted that, in this step, the first feature vector and the second feature vector are extracted by the SIFT visual word bag in the same way as the target feature vector of the original echo signal.
S43, establishing a support vector machine model to be trained, training the support vector machine model to be trained by using the first feature set and the second feature set, and optimizing parameters of the support vector machine model to be trained by using a training result;
and S44, inputting the target characteristic vector of one 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.
As other optional embodiments, the method further comprises:
and S50, if the classification type of the original echo signal is the target feature type, storing the original echo signal in a target folder, wherein the target folder is used for storing a plurality of echo signals of the same feature type.
In this embodiment, a plurality of different folders may be established in the terminal in advance, and each folder is used for storing an echo signal of one type of ground object. And when the output original echo signal belongs to a certain ground object type, storing the echo signal in a corresponding folder.
As other optional embodiments, the method further comprises:
and S60, performing point cloud inversion on the original echo signals in the target folder to generate point cloud data or images of the type of the ground object corresponding to the folder. For example, if an image of a building in the target area is to be generated, a target folder for storing the building is selected to directly perform point cloud inversion, which is convenient for a user to quickly image.
According to the full-waveform laser echo signal classification method provided by the embodiment of the invention, full-waveform sampling is carried out on original echo signals generated by a ground target through an airborne laser radar of a streak tube, and classification is directly carried out according to local characteristics of a streak chart of the original echo signals, so that the complicated process of converting all the original echo signals into a point cloud data chart is reduced; in addition, the data volume of each echo signal based on waveform sampling can reach 1000 times of the data volume of non-waveform sampling, and a large amount of characteristic information of ground targets is obtained, so that more accurate classification is realized, and the classification accuracy rate is more than 95%.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A full-waveform laser echo signal classification method based on SIFT visual word bag is characterized by comprising the following steps:
an SIFT visual word bag is established in advance;
controlling a streak tube laser radar to scan a target area row by row, and collecting an original echo signal of each row of the scanned target area;
extracting all SIFT feature points in the original echo signal, and quantizing all SIFT feature points by using the SIFT visual word bag to obtain a target feature vector of the original echo signal;
inputting the target characteristic vector of the original echo signal into a support vector machine model in real time, and classifying the ground features of the original echo signal; the support vector machine model is trained by using a plurality of groups of training data, wherein the plurality of groups of training data comprise first type training data and second type training data; each set of data in the first class of training data comprises: a first echo signal of a target ground object type, a tag used for identifying the first echo signal and a first feature vector used for characterizing the first echo signal; each set of data in the second class of training data includes: a second echo signal that does not belong to the target surface feature type, a tag to identify the second echo signal, and a second feature vector to characterize the second echo signal.
2. The method of claim 1, wherein pre-establishing SIFT visual bags of words comprises:
selecting a plurality of echo signals of different ground object types, and respectively extracting SIFT feature points of each echo signal;
and (3) collecting the SIFT feature points of all the echo signals, merging SIFT feature points with similar word senses by using a K-Means algorithm, and constructing an SIFT visual word bag containing K words.
3. The method of claim 1, wherein the extracting all SIFT feature points in the original echo signal comprises:
detecting a plurality of extreme points of the original echo signal in a Gaussian scale space;
removing unstable extreme points from the plurality of extreme points to obtain key points;
determining a principal direction of each of the keypoints;
and generating descriptors of the key points to obtain SIFT feature points.
4. The method of claim 1, wherein the quantizing all the SIFT feature points using the SIFT visual bag of words to obtain a target feature vector comprises:
replacing the SIFT feature points with word approximations in the SIFT visual word bag;
and counting the occurrence times of each word in the SIFT feature points 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 appearing in the SIFT feature points.
5. The method of claim 1, wherein controlling the streak tube lidar to scan the target area column by column and to collect one raw echo signal for each column of the scanned target area comprises:
controlling the streak tube laser radar to scan a target area column by column along a flight direction perpendicular to the airplane;
receiving optical signals returned by ground targets in each row of scanning target areas at different moments, and converting the optical signals into photoelectron pulses in the streak tube;
and linearly deflecting the photoelectron pulses to spread the photoelectron pulses at different moments on a fluorescent screen according to a time sequence so as to form an original echo signal of each column of the scanning target area.
6. The method according to claim 1, wherein the target feature vector of an original echo signal is input into a support vector machine model in real time to classify the ground feature of the original echo signal; the support vector machine model is trained by using a plurality of groups of training data, wherein the plurality of groups of training data comprise a first class of training data and a second class of training data, and the method comprises the following steps:
selecting a first echo signal of a target ground object type and a second echo signal which does not belong to the target ground object type as training data; respectively allocating a first label and a second label to the first echo signal and the second echo signal;
extracting a first feature vector of the first echo signal based on the SIFT visual word bag, and combining the first feature vector and the first label to form a first feature set; extracting a second feature vector of the second echo signal, and combining the second feature vector and a second label to form a second feature set;
establishing a support vector machine model to be trained, training the support vector machine model to be trained by utilizing the first feature set and the second feature set, and optimizing parameters of the support vector machine model to be trained by utilizing a training result;
and inputting a pre-classified target characteristic vector of an original echo signal into a trained support vector machine model in real time, and outputting the classification type of the original echo signal.
7. The method according to claim 1 or 6, wherein the first echo signal is a fringe image selected from a plain fringe image, a building fringe image, and a tree fringe image.
8. The method of claim 7, wherein when the first echo signal is a plain fringe pattern, the second echo signal is a non-plain fringe pattern, the non-plain fringe pattern comprising one or more of a building fringe pattern, a tree fringe pattern; alternatively, the first and second electrodes may be,
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 comprises one or more of a plain fringe pattern and a tree fringe pattern; alternatively, the first and second electrodes may be,
and 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 comprises one or more of a building fringe pattern and a plain fringe pattern.
9. The method of claim 1 or 6, further comprising:
and if the classification type of the original echo signal is the target ground object type, storing the original echo signal in a target folder, wherein the target folder is used for storing a plurality of echo signals of the same ground object type.
10. The method of claim 9, further comprising:
and performing point cloud inversion on the plurality of original echo signals in the target folder to generate point cloud data or images of the target ground object type.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110859165.4A CN113567953A (en) | 2021-07-28 | 2021-07-28 | Full-waveform laser echo signal classification method based on SIFT visual word bag |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110859165.4A CN113567953A (en) | 2021-07-28 | 2021-07-28 | Full-waveform laser echo signal classification method based on SIFT visual word bag |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113567953A true CN113567953A (en) | 2021-10-29 |
Family
ID=78168588
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110859165.4A Pending CN113567953A (en) | 2021-07-28 | 2021-07-28 | Full-waveform laser echo signal classification method based on SIFT visual word bag |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113567953A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102622607A (en) * | 2012-02-24 | 2012-08-01 | 河海大学 | Remote sensing image classification method based on multi-feature fusion |
| CN103645480A (en) * | 2013-12-04 | 2014-03-19 | 北京理工大学 | Geographic and geomorphic characteristic construction method based on laser radar and image data fusion |
| CN106919920A (en) * | 2017-03-06 | 2017-07-04 | 重庆邮电大学 | Scene recognition method based on convolution feature and spatial vision bag of words |
| CN107368807A (en) * | 2017-07-20 | 2017-11-21 | 东南大学 | A kind of monitor video vehicle type classification method of view-based access control model bag of words |
| WO2021008202A1 (en) * | 2019-07-16 | 2021-01-21 | 浙江大学 | Method for kernel support vector machine target classification based on millimeter-wave radar point cloud features |
-
2021
- 2021-07-28 CN CN202110859165.4A patent/CN113567953A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102622607A (en) * | 2012-02-24 | 2012-08-01 | 河海大学 | Remote sensing image classification method based on multi-feature fusion |
| CN103645480A (en) * | 2013-12-04 | 2014-03-19 | 北京理工大学 | Geographic and geomorphic characteristic construction method based on laser radar and image data fusion |
| CN106919920A (en) * | 2017-03-06 | 2017-07-04 | 重庆邮电大学 | Scene recognition method based on convolution feature and spatial vision bag of words |
| CN107368807A (en) * | 2017-07-20 | 2017-11-21 | 东南大学 | A kind of monitor video vehicle type classification method of view-based access control model bag of words |
| WO2021008202A1 (en) * | 2019-07-16 | 2021-01-21 | 浙江大学 | Method for kernel support vector machine target classification based on millimeter-wave radar point cloud features |
Non-Patent Citations (2)
| Title |
|---|
| 徐涛: "基于决策树的激光雷达回波数据分类技术研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, pages 5 - 6 * |
| 费旋珈: "基于MRF模型的北冰洋海冰SAR图像分类研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, pages 19 - 42 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xia et al. | DOTA: A large-scale dataset for object detection in aerial images | |
| Shi et al. | Study on modeling method of forest tree image recognition based on CCD and theodolite | |
| Liu et al. | Multiscale morphological compressed change vector analysis for unsupervised multiple change detection | |
| Gao et al. | Automatic change detection in synthetic aperture radar images based on PCANet | |
| Weir et al. | Spacenet mvoi: A multi-view overhead imagery dataset | |
| Lu et al. | A survey of image classification methods and techniques for improving classification performance | |
| CN109871902B (en) | SAR small sample identification method based on super-resolution countermeasure generation cascade network | |
| Hu et al. | Automatic plankton image recognition with co-occurrence matrices and support vector machine | |
| CN105404886B (en) | Characteristic model generation method and characteristic model generating means | |
| CN111898688B (en) | Airborne LiDAR data tree classification method based on three-dimensional deep learning | |
| CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
| CN109711446A (en) | A kind of terrain classification method and device based on multispectral image and SAR image | |
| WO2013088175A1 (en) | Image processing method | |
| CN108257151B (en) | PCANet image change detection method based on significance analysis | |
| CN113298137B (en) | Hyperspectral image classification method based on local similarity data gravitation | |
| CN108564092A (en) | Sunflower disease recognition method based on SIFT feature extraction algorithm | |
| CN108898065A (en) | Candidate regions quickly screen and the depth network Ship Target Detection method of dimension self-adaption | |
| CN109902618A (en) | A kind of sea ship recognition methods and device | |
| CN112907520A (en) | Single tree crown detection method based on end-to-end deep learning method | |
| CN106485239A (en) | One kind is using one-class support vector machines detection river mesh calibration method | |
| Larabi et al. | High-resolution optical remote sensing imagery change detection through deep transfer learning | |
| JPH08508128A (en) | Image classification method and apparatus using distribution map | |
| CN113567953A (en) | Full-waveform laser echo signal classification method based on SIFT visual word bag | |
| Liu et al. | Semi-automatic extraction and mapping of farmlands based on high-resolution remote sensing images | |
| Guo et al. | A novel method of ship detection from spaceborne optical image based on spatial pyramid matching |
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 |