CN109190673A - The ground target classification method sentenced is refused based on random forest and data - Google Patents
The ground target classification method sentenced is refused based on random forest and data Download PDFInfo
- Publication number
- CN109190673A CN109190673A CN201810874485.5A CN201810874485A CN109190673A CN 109190673 A CN109190673 A CN 109190673A CN 201810874485 A CN201810874485 A CN 201810874485A CN 109190673 A CN109190673 A CN 109190673A
- Authority
- CN
- China
- Prior art keywords
- signal
- echo
- value
- frequency
- feature
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
-
- 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
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- 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
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/414—Discriminating targets with respect to background clutter
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Electromagnetism (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a kind of ground target classification methods for refusing to sentence based on random forest and data, and implementation step is: (1) pre-processing to training sample set；(2) training characteristics matrix is extracted；(3) training random forest grader；(4) test sample is pre-processed；(5) testing feature vector is extracted；(6) ground target output probability vector is calculated；(7) judge whether test sample refuses to sentence；(8) if refusing to sentence, using test sample as the echo-signal of no target fine motion characteristic；(9) if not refusing to sentence, maximum value corresponds to ground target classification results of the classification as the test sample in output probability vector.The present invention refuses to sentence clutter, Deceiving interference and the echo signal without fine motion, improves the Classification and Identification rate of ground moving object, while using the classifier with parallel processing capability, performance of the improvement method in terms of real-time.
Description
Technical field
The invention belongs to field of communication technology, further relate to one of Radar Signal Processing Technology field be based on
Machine forest and data refuse the ground target classification method sentenced.The present invention can be in the environment done there are clutter, deception formula, over the ground
The different vehicle target and human body target of face movement carry out real-time grading.
Background technique
In the classification of radar ground target, radar return may contain clutter, Deceiving interference and ground appearance without fine motion
Mark signal.For clutter, Deceiving interference, classification method should be refused to classify；And for the echo signal of no fine motion, because not
Effective fine motion feature that classification needs can be extracted, so classification method effectively cannot correctly classify.How rationally effectively to go
Noise wave removing, Deceiving interference and the echo signal without fine motion are one of the problems of ground target classification.Meanwhile the ground of existing radar
There is still a need for the longer processing times for Area Objects classification method, and there are also to be hoisted in terms of real-time.
Its application where Beijing antenna electric measuring study patent document " a kind of moving vehicle objective classification method be
A kind of moving vehicle target is proposed in system " (number of patent application: CN201510738784.2, publication number: CN105403872A)
Classification method.This method comprises the concrete steps that: the first step, obtains the original doppler spectral of target vehicle, and to original Doppler
Spectrum carries out clutter recognition, the target Doppler spectrum after obtaining clutter recognition；Second step composes target Doppler and carries out speed normalizing
Change, obtains normalization doppler spectral；Third step calculates separately normalization doppler spectral and wheeled vehicle according to normalization doppler spectral
Target Doppler composes template, creeper truck vehicle target Doppler spectrum the distance between template, obtain respectively wheeled distance and crawler belt away from
From；4th step is compared wheeled distance with crawler belt distance, determines that target vehicle is wheeled vehicle or crawler belt according to comparison result
Vehicle.Although this method can handle the radar signal of high s/n ratio and identify creeper truck and wheeled vehicle, this method is still had
Shortcoming is: since this method is that how general direct comparative observation doppler spectral and wheeled vehicle doppler spectral template, creeper truck be
The distance for strangling spectrum template, so clutter, Deceiving interference can still be differentiated there are in the environment of clutter, Deceiving interference
At vehicle target；And then can be by random assortment to the ground target signal of no fine motion, this causes the erroneous judgement of classification method, and makes
Discrimination decline.
Xian Electronics Science and Technology University is in patent document " the ground target classification side based on robustness time-frequency characteristics of its application
When proposing one kind in method " (number of patent application: CN201510475477.X, publication number: CN105044701A) based on robustness
The ground target classification method of frequency feature.This method comprises the concrete steps that: the first step, returns to the high s/n ratio signal energy of acquisition
One changes, and obtains training signal；Second step extracts 3 dimension time-frequency characteristics and Training Support Vector Machines point from the time-frequency spectrum of training signal
Class device；Third step obtains test signal to the test signal energy normalization of acquisition；4th step, from the time-frequency spectrum of test signal
In, extract 3 dimension time-frequency characteristics；The 3 dimension time-frequency characteristics for testing signal are sent to trained support vector cassification by the 5th step
In device, classification results are obtained.Although this method can handle the radar signal of ground target and classification, this method is still deposited
Shortcoming be: since this method is using support vector machine classifier, point of the support vector machines sheet as serial structure
Class device, so support vector machine classifier needs the more processing time, this will lead to performance of this method in terms of real-time
It is poor.
Summary of the invention
It is a kind of based on random forest and data it is an object of the invention in view of the deficiency of the prior art, propose
Refuse the ground target classification method sentenced.
Realizing the thinking of the object of the invention is, what is correctly classified to the ground target Narrow-band Radar echo-signal for having fine motion
Meanwhile clutter, Deceiving interference, the ground target radar echo signal without fine motion are carried out refusing to sentence, there are clutter, deception
In the environment of formula interference, real-time grading effectively is carried out to the different vehicle target and human body target of ground motion.Meanwhile this hair
The processing time that bright classification method needs is shorter, and the performance in terms of real-time is promoted.
The specific steps of the present invention are as follows:
(1) training sample set is pre-processed.
(1a) randomly selects at least 1000 and exists in frequency spectrum from the Narrow-band Radar echo-signal of different ground targets
The echo-signal of micro-Doppler effect forms training sample set.
(1b) utilizes region CLEAN method, carries out clutter recognition to the echo-signal that training sample is concentrated.
(1c) utilizes universe CLEAN method, carries out noise suppression to the echo-signal that the training sample after clutter recognition is concentrated
System, obtains pretreated training sample set.
(2) training characteristics matrix is extracted.
(2a) uses feature extracting method, extracts 7 features of the echo-signal that pretreated training sample is concentrated, group
At training characteristics matrix.
(2b) is normalized the element of training characteristics matrix, is normalized according to training formula is normalized
Training characteristics matrix afterwards.
(3) training random forest grader.
Training characteristics Input matrix after normalization is trained into random forest grader, until random forest point
The quantity of the decision tree of parallel processing is greater than 500 in class device, and deconditioning obtains having the trained of parallel processing capability
Random forest grader.
(4) test sample is pre-processed.
(4a) is using an echo-signal of Narrow-band Radar real-time reception as test sample.
(4b) utilizes region CLEAN method, carries out clutter recognition to the echo-signal of test sample.
(4c) utilizes universe CLEAN method, carries out noise suppressed to the echo-signal of the test sample after clutter recognition, obtains
To pretreated test sample.
(5) testing feature vector is extracted.
(5a) uses feature extracting method, extracts 7 features of echo-signal in pretreated test sample, and composition is surveyed
Try feature vector.
(5b) tests formula according to normalization, is normalized, is normalized to the element of testing feature vector
Testing feature vector afterwards.
(6) ground target output probability vector is calculated.
Testing feature vector after normalization is input to the trained random forest with parallel processing capability by (6a)
In classifier, the judgement result of every decision tree in random forest grader is obtained.
(6b) calculates the output probability of every class ground target according to output probability formula, by the ground target of all categories
Output probability form ground target output probability vector.
(7) judge whether the element of ground target output probability vector is respectively less than threshold value, if so, executing step (8)；It is no
Then, step (9) are executed.
(8) it using test sample as the echo-signal of no ground target fine motion characteristic, refuses to execute after sentencing the test sample
Step (4).
(9) maximum value in ground target output probability vector classification is corresponded to as the ground target of the test sample to classify
As a result.
Compared with prior art, the invention has the following advantages that
First, since the present invention judges whether the element of ground target output probability vector is respectively less than threshold value, if so, will
Echo-signal of the test sample as no ground target fine motion characteristic refuses to resurvey test sample after sentencing the test sample；
Otherwise, maximum value in ground target output probability vector is corresponded into ground target classification results of the classification as the test sample,
It overcomes in the prior art there are in the environment of clutter, Deceiving interference, since clutter, Deceiving interference can still be differentiated
At ground target, the ground target signal of no fine motion then can be by random assortment, and the problem for causing discrimination lower, so that this hair
It is bright to have the advantages that can be reduced erroneous judgement and improve Classification and Identification rate.
Second, since the present invention is classified using the trained random forest grader with parallel processing capability,
It overcomes in the prior art since the support vector machine classifier using needs compared with the multiprocessing time is classified, and leads to ground
Objective classification method problem poor in terms of real-time, so that the processing time that the present invention needs is shorter, in terms of real-time
Performance promoted.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Specific embodiment
With reference to the accompanying drawing 1, realization step of the invention is further described.
Step 1, training sample set is pre-processed.
From the Narrow-band Radar echo-signal of different ground targets, randomly selecting at least 1000, there are micro- more in frequency spectrum
The general echo-signal for strangling effect, forms training sample set.
Using region CLEAN method, clutter recognition is carried out to the echo-signal that training sample is concentrated.
Specific step is as follows for the region CLEAN method:
Step 1 estimates land clutter energy in radar return according to radar parameter.
Step 2 does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal, by clutter spectrum model
It encloses as clutter region.
Step 3 searches for the maximum value in clutter region, the corresponding phase of maximum value in clutter region, miscellaneous in doppler spectral
The corresponding amplitude of the maximum value in wave region, the corresponding Doppler frequency of maximum value in clutter region.
Step 4 reconstructs the corresponding time-domain signal of maximum value in clutter region according to the following formula:
Wherein, s (t) indicates that reconstruct clutter maximum regional value corresponds to the signal amplitude of t moment in time-domain signal, and R indicates miscellaneous
The corresponding amplitude of the maximum value in wave region, K indicate the points of discrete Fourier transform, and exp is indicated using natural number as the index at bottom
Operation, j indicate imaginary unit, and π indicates that pi, d indicate the corresponding Doppler frequency of maximum value in clutter region,Indicate miscellaneous
The corresponding phase of the maximum value in wave region.
Step 5 subtracts the corresponding time-domain signal of reconstruct clutter maximum regional value with echo-signal, the echo that obtains that treated
Signal.
Step 6, the energy of echo-signal after calculation processing in clutter region.
Step 7, treated whether energy of the echo-signal in clutter region is less than land clutter energy for judgement, if so,
Echo-signal after to clutter recognition；Otherwise, step 2 is executed.
Using universe CLEAN method, noise suppressed is carried out to the echo-signal that the training sample after clutter recognition is concentrated, is obtained
To pretreated training sample set.
Specific step is as follows for the universe CLEAN method:
Step 1 estimates noise energy in radar return according to radar parameter.
Step 2 does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal.
Step 3 searches for the maximum value of doppler spectral, the corresponding phase of the maximum value of doppler spectral, more in doppler spectral
The corresponding amplitude of maximum value, the corresponding Doppler frequency of the maximum value of doppler spectral of Pu Le spectrum.
Step 4 reconstructs the corresponding time-domain signal of maximum value of doppler spectral according to the following formula:
Wherein, z (t) indicates the signal amplitude of t moment in the corresponding time-domain signal of maximum value of reconstruct doppler spectral, Y table
Show the corresponding amplitude of the maximum value of doppler spectral, m indicates the corresponding Doppler frequency of maximum value of doppler spectral, and how general θ indicate
Strangle the corresponding phase of maximum value of spectrum.
Step 5 subtracts the corresponding time-domain signal of reconstruct doppler spectral maximum value with echo-signal, the echo that obtains that treated
Signal.
Step 6, the echo-signal energy after calculation processing.
Step 7, treated whether echo-signal energy is greater than noise energy for judgement, if so, executing step 2；Otherwise, it holds
Row step 8.
Step 8, the echo with the echo-signal that subtracts that treated of the echo-signal after clutter recognition, after obtaining noise suppressed
Signal.
Step 2, training characteristics matrix is extracted.
Using feature extracting method, 7 features of the echo-signal that pretreated training sample is concentrated, composition instruction are extracted
Practice eigenmatrix.
Specific step is as follows for the feature extracting method:
Step 1 does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal.
Step 2 searches for the maximum value of doppler spectral, the corresponding phase of the maximum value of doppler spectral, more in doppler spectral
The corresponding amplitude of maximum value, the corresponding Doppler frequency of the maximum value of doppler spectral of Pu Le spectrum.
Step 3 reconstructs the corresponding time-domain signal of maximum value of doppler spectral.
Step 4 subtracts the corresponding time-domain signal of reconstruct doppler spectral maximum value with echo-signal, obtains residual echo letter
Number.
Step 5, residual echo signal do discrete Fourier transform, obtain the doppler spectral of residual echo signal.
Step 6 does discrete Fourier transform three times to residual echo signal using step 1 to step 5, obtains three residues
The doppler spectral of echo-signal.
Step 7 calculates the value of first feature according to the following formula:
Wherein, D1Indicate the value of first feature, n indicates the number of Frequency point in doppler spectral, U2(n) it indicates second
The amplitude of n-th of Frequency point, U in the doppler spectral of residual echo signal0(n) it indicates in the doppler spectral of echo-signal n-th
The amplitude of Frequency point.
Step 8 calculates the value of second feature according to the following formula:
Wherein, D2Indicate the value of second feature, U3(n) it indicates in the doppler spectral of third time residual echo signal n-th
The amplitude of Frequency point.
Step 9 carries out the linear normalization that quadratic sum is 1 to echo-signal.
Step 10 walks normalized echo-signal using the window of 0.5 times of echo-signal length after being rounded downwards
A length of 1 sliding window obtains sliding window matrix.
Step 11 carries out autocorrelation operation to sliding window matrix, obtains autocorrelation matrix.
Step 12 carries out Eigenvalues Decomposition to autocorrelation matrix, obtains characteristic value sequence.
Characteristic value sequence is sorted by non-increasing sequence, obtains characteristic spectrum by step 13.
Step 14 calculates the value of third feature according to the following formula:
Wherein, D3Indicating the value of third feature, x indicates the alternative quantity of characteristic value in characteristic spectrum,It indicates to meet the value for being greater than the corresponding x of element minimum value that 0.98 requires, r indicates characteristic value in characteristic spectrum
Number, λrIndicating r-th of characteristic value in characteristic spectrum, M indicates signal length,Indicate that downward floor operation, b indicate characteristic spectrum
The number of middle characteristic value, λbIndicate b-th of characteristic value in characteristic spectrum.
Step 15 calculates the value of the 4th feature according to the following formula:
Wherein, D4Indicate the value of the 4th feature, ln () expression takes log operations.
Step 16 does Short Time Fourier Transform to time domain echo-signal, obtains amplitude spectrum.
Step 17 is according to the following formula normalized amplitude spectrum:
Wherein, S (f, a) indicate frequency be f, the normalization amplitude that the time is a,Indicate that frequency is in amplitude spectrum
F, the time is the amplitude of a, and F indicates maximum frequency, and f indicates the frequency in amplitude spectrum, and A indicates time domain maximum value, and a indicates amplitude
Time in spectrum.
Step 18 calculates time-frequency entropy according to the following formula:
Wherein, E (f) indicates that frequency is the time-frequency entropy of f.
Step 19 calculates time-frequency entropy mean value according to the following formula:
Wherein,Indicate time-frequency entropy mean value.
Step 20 calculates the value of the 5th feature according to the following formula:
Wherein, D5Indicate the value of the 5th feature,Indicate the corresponding frequency f of maximum value.
21st step meets within the scope of the value to maximum frequency of the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as main body upper frequency limit.
22nd step meets within the scope of the value of frequency minima to the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as main body lower-frequency limit.
23rd step calculates the value of the 6th feature according to the following formula:
Wherein, D6Indicate the value of the 6th feature, w indicates main body upper frequency limit, and q indicates main body lower-frequency limit.
24th step meets within the scope of the value to maximum frequency of the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as Doppler's upper limiting frequency.
25th step meets within the scope of the value of frequency minima to the 5th featureWith
Under the conditions of the 5th feature of distance the nearest frequency of value as Doppler's lower frequency limit.
26th step subtracts Doppler's lower frequency limit with Doppler's upper limiting frequency, obtains the value of the 7th feature.
According to training formula is normalized, the element of training characteristics matrix is normalized, after being normalized
Training characteristics matrix.
The normalization training formula is as follows:
Wherein,J-th of normalization characteristic of i-th of echo-signal, o in training characteristics matrix after indicating normalizationi,j
Indicate j-th of feature of i-th of echo-signal in training characteristics matrix, min expression is minimized operation, and max is maximized behaviour
Make, ojIndicate the vector of j-th of feature composition of all echo-signals in training characteristics matrix.
Step 3, training random forest grader.
Training characteristics Input matrix after normalization is trained into random forest grader, until random forest point
The quantity of the decision tree of parallel processing is greater than 500 in class device, and deconditioning obtains having the trained of parallel processing capability
Random forest grader.
Step 4, test sample is pre-processed.
Using an echo-signal of Narrow-band Radar real-time reception as test sample.
Using region CLEAN method, clutter recognition is carried out to the echo-signal of test sample.
Specific step is as follows for the region CLEAN method.
Step 1 estimates land clutter energy in radar return according to radar parameter.
Step 2 does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal, by clutter spectrum model
It encloses as clutter region.
Step 3 searches for the maximum value in clutter region, the corresponding phase of maximum value in clutter region, miscellaneous in doppler spectral
The corresponding amplitude of the maximum value in wave region, the corresponding Doppler frequency of maximum value in clutter region.
Step 4 reconstructs the corresponding time-domain signal of maximum value in clutter region according to the following formula:
Wherein, s (t) indicates that reconstruct clutter maximum regional value corresponds to the signal amplitude of t moment in time-domain signal, and R indicates miscellaneous
The corresponding amplitude of the maximum value in wave region, K indicate the points of discrete Fourier transform, and exp is indicated using natural number as the index at bottom
Operation, j indicate imaginary unit, and π indicates that pi, d indicate the corresponding Doppler frequency of maximum value in clutter region,Indicate miscellaneous
The corresponding phase of the maximum value in wave region.
Step 5 subtracts the corresponding time-domain signal of reconstruct clutter maximum regional value with echo-signal, the echo that obtains that treated
Signal.
Step 6, the energy of echo-signal after calculation processing in clutter region.
Step 7, treated whether energy of the echo-signal in clutter region is less than land clutter energy for judgement, if so,
Echo-signal after to clutter recognition；Otherwise, step 2 is executed.
Using universe CLEAN method, noise suppressed is carried out to the echo-signal of the test sample after clutter recognition, is obtained pre-
Test sample that treated.
Specific step is as follows for the universe CLEAN method:
Step 1 estimates noise energy in radar return according to radar parameter.
Step 2 does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal.
Step 3 searches for the maximum value of doppler spectral, the corresponding phase of the maximum value of doppler spectral, more in doppler spectral
The corresponding amplitude of maximum value, the corresponding Doppler frequency of the maximum value of doppler spectral of Pu Le spectrum；.
Step 4 reconstructs the corresponding time-domain signal of maximum value of doppler spectral according to the following formula:
Wherein, z (t) indicates the signal amplitude of t moment in the corresponding time-domain signal of maximum value of reconstruct doppler spectral, Y table
Show the corresponding amplitude of the maximum value of doppler spectral, m indicates the corresponding Doppler frequency of maximum value of doppler spectral, and how general θ indicate
Strangle the corresponding phase of maximum value of spectrum.
Step 5 subtracts the corresponding time-domain signal of reconstruct doppler spectral maximum value with echo-signal, the echo that obtains that treated
Signal.
Step 6, the echo-signal energy after calculation processing.
Step 7, treated whether echo-signal energy is greater than noise energy for judgement, if so, executing step 2；Otherwise, it holds
Row step 8.
Step 8, the echo with the echo-signal that subtracts that treated of the echo-signal after clutter recognition, after obtaining noise suppressed
Signal.
Step 5, testing feature vector is extracted.
Using feature extracting method, 7 features of echo-signal in pretreated test sample are extracted, composition test is special
Levy vector.
Specific step is as follows for the feature extracting method:
Step 1 does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal.
Step 2 searches for the maximum value of doppler spectral, the corresponding phase of the maximum value of doppler spectral, more in doppler spectral
The corresponding amplitude of maximum value, the corresponding Doppler frequency of the maximum value of doppler spectral of Pu Le spectrum.
Step 3 reconstructs the corresponding time-domain signal of maximum value of doppler spectral.
Step 4 subtracts the corresponding time-domain signal of reconstruct doppler spectral maximum value with echo-signal, obtains residual echo letter
Number.
Step 5, residual echo signal do discrete Fourier transform, obtain the doppler spectral of residual echo signal.
Step 6 does discrete Fourier transform three times to residual echo signal to the 5th step using the first step, obtains three and remains
The doppler spectral of remaining echo-signal.
Step 7 calculates the value of first feature according to the following formula:
Wherein, D1Indicate the value of first feature, n indicates the number of Frequency point in doppler spectral, U2(n) it indicates second
The amplitude of n-th of Frequency point, U in the doppler spectral of residual echo signal0(n) it indicates in the doppler spectral of echo-signal n-th
The amplitude of Frequency point.
Step 8 calculates the value of second feature according to the following formula:
Wherein, D2Indicate the value of second feature, U3(n) it indicates in the doppler spectral of third time residual echo signal n-th
The amplitude of Frequency point.
Step 9 carries out the linear normalization that quadratic sum is 1 to echo-signal.
Step 10 walks normalized echo-signal using the window of 0.5 times of echo-signal length after being rounded downwards
A length of 1 sliding window obtains sliding window matrix.
Step 11 carries out autocorrelation operation to sliding window matrix, obtains autocorrelation matrix.
Step 12 carries out Eigenvalues Decomposition to autocorrelation matrix, obtains characteristic value sequence.
Characteristic value sequence is sorted by non-increasing sequence, obtains characteristic spectrum by step 13.
Step 14 calculates the value of third feature according to the following formula:
Wherein, D3Indicating the value of third feature, x indicates the alternative quantity of characteristic value in characteristic spectrum,It indicates to meet the value for being greater than the corresponding x of element minimum value that 0.98 requires, r indicates characteristic value in characteristic spectrum
Number, λrIndicating r-th of characteristic value in characteristic spectrum, M indicates signal length,Indicate that downward floor operation, b indicate characteristic spectrum
The number of middle characteristic value, λbIndicate b-th of characteristic value in characteristic spectrum.
Step 15 calculates the value of the 4th feature according to the following formula:
Wherein, D4Indicate the value of the 4th feature, ln () expression takes log operations.
Step 16 does Short Time Fourier Transform to time domain echo-signal, obtains amplitude spectrum.
Step 17 is according to the following formula normalized amplitude spectrum:
Wherein, S (f, a) indicate frequency be f, the normalization amplitude that the time is a,Indicate that frequency is in amplitude spectrum
F, the time is the amplitude of a, and F indicates maximum frequency, and f indicates the frequency in amplitude spectrum, and A indicates time domain maximum value, and a indicates amplitude
Time in spectrum.
Step 18 calculates time-frequency entropy according to the following formula:
Wherein, E (f) indicates that frequency is the time-frequency entropy of f.
Step 19 calculates time-frequency entropy mean value according to the following formula:
Wherein,Indicate time-frequency entropy mean value.
Step 20 calculates the value of the 5th feature according to the following formula:
Wherein, D5Indicate the value of the 5th feature,Indicate the corresponding frequency f of maximum value.
21st step meets within the scope of the value to maximum frequency of the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as main body upper frequency limit.
22nd step meets within the scope of the value of frequency minima to the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as main body lower-frequency limit.
23rd step calculates the value of the 6th feature according to the following formula:
Wherein, D6Indicate the value of the 6th feature, w indicates main body upper frequency limit, and q indicates main body lower-frequency limit.
24th step meets within the scope of the value to maximum frequency of the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as Doppler's upper limiting frequency.
25th step meets within the scope of the value of frequency minima to the 5th featureWith
Under the conditions of the 5th feature of distance the nearest frequency of value as Doppler's lower frequency limit.
26th step subtracts Doppler's lower frequency limit with Doppler's upper limiting frequency, obtains the value of the 7th feature.
Formula is tested according to normalization, the element of testing feature vector is normalized, after being normalized
Testing feature vector.
The normalization test formula is as follows:
Wherein,J-th of normalization characteristic in testing feature vector after indicating normalization, pjIndicate test feature to
J-th of feature in amount.
Step 6, ground target output probability vector is calculated.
Testing feature vector after normalization is input to the classification of the trained random forest with parallel processing capability
In device, the judgement result of every decision tree in random forest grader is obtained.
According to output probability formula, the output probability of every class ground target is calculated, by the defeated of the ground target of all categories
Probability forms ground target output probability vector out.
The output probability formula is as follows:
Wherein, J (c) indicates that the output probability of c class ground target classification, v indicate decision tree in random forest grader
Quantity, ∑ indicate sum operation, and l indicates the number of decision tree in random forest grader, and G () indicates indicative function, hlTable
Show the judgement result of the l decision tree in random forest grader.
Step 7, judge whether the element of ground target output probability vector is respectively less than threshold value, if so, executing step 8；It is no
Then, step 9 is executed.
The threshold value refers to, for every class ground target, according to Receiver operating curve, searching meets such ground
The limit value that Area Objects false alarm rate and discrimination require；By such corresponding ground appearance of element in ground target output probability vector
Target limit value is as threshold value.
Step 8, it using test sample as the echo-signal of no ground target fine motion characteristic, refuses to hold after sentencing the test sample
Row step 4.
Step 9, maximum value in ground target output probability vector is corresponded into ground target of the classification as the test sample
Classification results.
Effect of the invention is described further below with reference to emulation experiment.
1. emulation experiment condition:
Narrow-band Radar parameter in emulation experiment of the invention are as follows: repetition 2131Hz, pulse accumulation number 128；Software platform
Are as follows: 10 operating system of Windows and Matlab R2017a.
2. emulation experiment content:
Emulation experiment content of the invention is wheeled vehicle, creeper truck and the people's three classes ground using Narrow-band Radar acquisition movement
The echo-signal of target.Wheeled vehicle, creeper truck the direction of motion include approach radar, far from radar, turn around, turn, the fortune of people
Dynamic direction includes approaching radar, far from radar.Randomly select the echo-signal group in 6000 frequency spectrums there are micro-Doppler effect
At training sample set, the quantity of every class sample is no less than 600；Test sample, comprising clutter, Deceiving interference, without fine motion
Ground return signal, the ground target radar echo signal there are fine motion.In emulation experiment, using the method for the present invention and
Conventional classification method respectively classifies to 30279 test samples of three classes ground target.Conventional classification method uses and this
Invention difference is that conventional classification method uses support vector machine classifier, and classification results are sentenced without refusing.
3. analysis of simulation result:
In order to evaluate the method for the present invention and conventional classification method, according to the following formula, the every of emulation experiment of the present invention is calculated separately
The test discrimination of class ground target:
Wherein, γ (c) indicates the test discrimination of c class ground target,It indicates to the classification of c class ground target just
True test sample number, ξ (c) indicate the non-total number for refusing to sentence test sample of c class ground target.
Average recognition rate is the average value of the test discrimination of the ground target of all categories.
By the two methods of emulation experiment of the present invention there are the radar return test specimens in clutter, Deceiving interference environment
This discrimination is listed as follows:
1 conventional classification method result of table and test discrimination of the invention compare list
| Wheeled vehicle | Creeper truck | People | Average recognition rate | |
| Conventional classification method | 62.42% | 62.56% | 70.68% | 65.22% |
| The method of the present invention | 88.65% | 84.08% | 92.23% | 88.32% |
It can be seen that from the average recognition rate in table 1 there are in clutter, Deceiving interference environment, the method for the present invention phase
It has a clear superiority than conventional classification method.Specifically, relatively conventional classification method, is carried out on a surface target using the present invention
When classification, the test discrimination of three classes ground target rises 21% or more, so that overall recognition performance is substantially improved.Due to this
Inventive method and conventional method use identical clutter recognition, noise suppressed and feature extracting method, so the promotion of performance is main
It is to be refused to sentence bring by data, i.e., data, which refuse to sentence, can effectively remove clutter, Deceiving interference and the radar time without fine motion target
Wave test sample, and conventional method can receive clutter, Deceiving interference and the radar return test sample without fine motion target, go forward side by side
Row random assortment, this will lead to the discrimination of every class ground target and average recognition rate decline.In table 1, relatively conventional classification
The test discrimination of method, different moving vehicle targets and human body target is all promoted, and the amplitude of performance boost is similar,
This illustrates that data refuse to sentence the classification for being not limited to ground target, and the classification that it can integrally promote all categories in test sample is known
Not rate.
Claims (8)
1. a kind of ground target classification method for refusing to sentence based on random forest and data, which is characterized in that judge that ground target is defeated
Whether the element of probability vector is respectively less than threshold value out, the trained random forest grader with parallel processing capability；The party
The specific steps of method include the following:
(1) training sample set is pre-processed:
(1a) from the Narrow-band Radar echo-signal of different ground targets, randomly selecting at least 1000, there are micro- more in frequency spectrum
The general echo-signal for strangling effect, forms training sample set；
(1b) utilizes region CLEAN method, carries out clutter recognition to the echo-signal that training sample is concentrated；
(1c) utilizes universe CLEAN method, carries out noise suppressed to the echo-signal that the training sample after clutter recognition is concentrated, obtains
To pretreated training sample set；
(2) training characteristics matrix is extracted:
(2a) uses feature extracting method, extracts 7 features of the echo-signal that pretreated training sample is concentrated, composition instruction
Practice eigenmatrix；
(2b) is normalized the element of training characteristics matrix, after being normalized according to training formula is normalized
Training characteristics matrix；
(3) training random forest grader:
Training characteristics Input matrix after normalization is trained into random forest grader, until random forest grader
The quantity of the decision tree of middle parallel processing is greater than 500, and deconditioning obtains having the trained random of parallel processing capability
Forest classified device；
(4) test sample is pre-processed:
(4a) is using an echo-signal of Narrow-band Radar real-time reception as test sample；
(4b) utilizes region CLEAN method, carries out clutter recognition to the echo-signal of test sample；
(4c) utilizes universe CLEAN method, carries out noise suppressed to the echo-signal of the test sample after clutter recognition, obtains pre-
Test sample that treated；
(5) testing feature vector is extracted:
(5a) uses feature extracting method, extracts 7 features of echo-signal in pretreated test sample, and composition test is special
Levy vector；
(5b) tests formula according to normalization, the element of testing feature vector is normalized, after being normalized
Testing feature vector；
(6) ground target output probability vector is calculated:
Testing feature vector after normalization is input to the classification of the trained random forest with parallel processing capability by (6a)
In device, the judgement result of every decision tree in random forest grader is obtained；
(6b) calculates the output probability of every class ground target according to output probability formula, by the defeated of the ground target of all categories
Probability forms ground target output probability vector out；
(7) judge whether the element of ground target output probability vector is respectively less than threshold value, if so, executing step (8)；Otherwise, it holds
Row step (9)；
(8) it using test sample as the echo-signal of no ground target fine motion characteristic, refuses to execute step after sentencing the test sample
(4)；
(9) maximum value in ground target output probability vector is corresponded into classification as the ground target of test sample classification knot
Fruit.
2. the ground target classification method according to claim 1 for refusing to sentence based on random forest and data, which is characterized in that
Specific step is as follows for region CLEAN method described in step (1b), step (4b):
The first step estimates land clutter energy in radar return according to radar parameter；
Second step does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal, and clutter spectrum range is made
For clutter region；
Third step searches for maximum value, the corresponding phase of maximum value in clutter region, clutter area in clutter region in doppler spectral
The corresponding amplitude of the maximum value in domain, the corresponding Doppler frequency of maximum value in clutter region；
4th step reconstructs the corresponding time-domain signal of maximum value in clutter region according to the following formula:
Wherein, s (t) indicates that reconstruct clutter maximum regional value corresponds to the signal amplitude of t moment in time-domain signal, and R indicates clutter area
The corresponding amplitude of the maximum value in domain, K indicate the points of discrete Fourier transform, and exp is indicated using natural number as the exponent arithmetic at bottom,
J indicates imaginary unit, and π indicates that pi, d indicate the corresponding Doppler frequency of maximum value in clutter region,Indicate clutter region
The corresponding phase of maximum value；
5th step subtracts the corresponding time-domain signal of reconstruct clutter maximum regional value with echo-signal, the echo letter that obtains that treated
Number；
6th step, the energy of echo-signal after calculation processing in clutter region；
7th step, treated whether energy of the echo-signal in clutter region is less than land clutter energy for judgement, if so, obtaining miscellaneous
Echo-signal after wave inhibition；Otherwise, second step is executed.
3. the ground target classification method according to claim 1 for refusing to sentence based on random forest and data, which is characterized in that
Specific step is as follows for universe CLEAN method described in step (1c), step (4c):
The first step estimates noise energy in radar return according to radar parameter；
Second step does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal；
Third step searches for maximum value, the corresponding phase of the maximum value of doppler spectral, Doppler of doppler spectral in doppler spectral
The corresponding amplitude of the maximum value of spectrum, the corresponding Doppler frequency of the maximum value of doppler spectral；
4th step reconstructs the corresponding time-domain signal of maximum value of doppler spectral according to the following formula:
Wherein, z (t) indicates the signal amplitude of t moment in the corresponding time-domain signal of maximum value of reconstruct doppler spectral, and Y indicates more
The corresponding amplitude of maximum value of Pu Le spectrum, m indicate the corresponding Doppler frequency of maximum value of doppler spectral, and θ indicates doppler spectral
The corresponding phase of maximum value；
5th step subtracts the corresponding time-domain signal of reconstruct doppler spectral maximum value with echo-signal, the echo letter that obtains that treated
Number；
6th step, the echo-signal energy after calculation processing；
7th step, treated whether echo-signal energy is greater than noise energy for judgement, if so, executing second step；Otherwise, it executes
8th step；
8th step, the echo letter with the echo-signal that subtracts that treated of the echo-signal after clutter recognition, after obtaining noise suppressed
Number.
4. the ground target classification method according to claim 1 for refusing to sentence based on random forest and data, which is characterized in that
Specific step is as follows for feature extracting method described in step (2a), step (5a):
The first step does discrete Fourier transform to echo-signal, obtains the doppler spectral of echo-signal；
Second step searches for maximum value, the corresponding phase of the maximum value of doppler spectral, Doppler of doppler spectral in doppler spectral
The corresponding amplitude of the maximum value of spectrum, the corresponding Doppler frequency of the maximum value of doppler spectral；
Third step reconstructs the corresponding time-domain signal of maximum value of doppler spectral；
4th step subtracts the corresponding time-domain signal of reconstruct doppler spectral maximum value with echo-signal, obtains residual echo signal；
5th step, residual echo signal do discrete Fourier transform, obtain the doppler spectral of residual echo signal；
6th step does discrete Fourier transform three times to residual echo signal to the 5th step using the first step, obtains three residues
The doppler spectral of echo-signal；
7th step calculates the value of first feature according to the following formula:
Wherein, D1Indicate the value of first feature, n indicates the number of Frequency point in doppler spectral, U2(n) second of residue is indicated
The amplitude of n-th of Frequency point, U in the doppler spectral of echo-signal0(n) n-th of frequency in the doppler spectral of echo-signal is indicated
The amplitude of point；
8th step calculates the value of second feature according to the following formula:
Wherein, D2Indicate the value of second feature, U3(n) n-th of frequency in the doppler spectral of third time residual echo signal is indicated
The amplitude of point；
9th step carries out the linear normalization that quadratic sum is 1 to echo-signal；
Tenth step uses the window of 0.5 times of echo-signal length after being rounded downwards to carry out step-length as 1 normalized echo-signal
Sliding window, obtain sliding window matrix；
11st step carries out autocorrelation operation to sliding window matrix, obtains autocorrelation matrix；
12nd step carries out Eigenvalues Decomposition to autocorrelation matrix, obtains characteristic value sequence；
Characteristic value sequence is sorted by non-increasing sequence, obtains characteristic spectrum by the 13rd step；
14th step calculates the value of third feature according to the following formula:
Wherein, D3Indicating the value of third feature, x indicates the alternative quantity of characteristic value in characteristic spectrum,
It indicates to meet the value for being greater than the corresponding x of element minimum value that 0.98 requires, r indicates the number of characteristic value in characteristic spectrum, λrIt indicates
R-th of characteristic value in characteristic spectrum, M indicate signal length,Indicate that downward floor operation, b indicate the volume of characteristic value in characteristic spectrum
Number, λbIndicate b-th of characteristic value in characteristic spectrum；
15th step calculates the value of the 4th feature according to the following formula:
Wherein, D4Indicate the value of the 4th feature, ln () expression takes log operations；
16th step does Short Time Fourier Transform to time domain echo-signal, obtains amplitude spectrum；
17th step is according to the following formula normalized amplitude spectrum:
Wherein, S (f, a) indicate frequency be f, the normalization amplitude that the time is a,Indicate that frequency is f, time in amplitude spectrum
For the amplitude of a, F indicates maximum frequency, and f indicates the frequency in amplitude spectrum, and A indicates time domain maximum value, and a is indicated in amplitude spectrum
Time；
18th step calculates time-frequency entropy according to the following formula:
Wherein, E (f) indicates that frequency is the time-frequency entropy of f；
19th step calculates time-frequency entropy mean value according to the following formula:
Wherein,Indicate time-frequency entropy mean value；
20th step calculates the value of the 5th feature according to the following formula:
Wherein, D5Indicate the value of the 5th feature,Indicate the corresponding frequency f of maximum value；
21st step meets within the scope of the value to maximum frequency of the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as main body upper frequency limit；
22nd step meets within the scope of the value of frequency minima to the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as main body lower-frequency limit；
23rd step calculates the value of the 6th feature according to the following formula:
Wherein, D6Indicate the value of the 6th feature, w indicates main body upper frequency limit, and q indicates main body lower-frequency limit；
24th step meets within the scope of the value to maximum frequency of the 5th featureWith
The nearest frequency of the value of the 5th feature of distance of condition is as Doppler's upper limiting frequency；
25th step meets within the scope of the value of frequency minima to the 5th featureWith
Under the conditions of the 5th feature of distance the nearest frequency of value as Doppler's lower frequency limit；
26th step subtracts Doppler's lower frequency limit with Doppler's upper limiting frequency, obtains the value of the 7th feature.
5. the ground target classification method according to claim 1 for refusing to sentence based on random forest and data, which is characterized in that
The training formula of normalization described in step (2b) is as follows:
Wherein,J-th of normalization characteristic of i-th of echo-signal, o in training characteristics matrix after indicating normalizationi,jIt indicates
J-th of feature of i-th of echo-signal in training characteristics matrix, min expression are minimized operation, and max is maximized operation, oj
Indicate the vector of j-th of feature composition of all echo-signals in training characteristics matrix.
6. the ground target classification method according to claim 1 for refusing to sentence based on random forest and data, which is characterized in that
It is as follows that formula is tested in normalization described in step (5b):
Wherein,J-th of normalization characteristic in testing feature vector after indicating normalization, pjIt indicates in testing feature vector
J-th of feature.
7. the ground target classification method according to claim 1 for refusing to sentence based on random forest and data, which is characterized in that
Output probability formula described in step (6b) is as follows:
Wherein, J (c) indicates that the output probability of c class ground target classification, v indicate decision tree quantity in random forest grader,
∑ indicates sum operation, and l indicates the number of decision tree in random forest grader, and G () indicates indicative function, hlIndicate random
The judgement result of the l decision tree in forest classified device.
8. the ground target classification method according to claim 1 for refusing to sentence based on random forest and data, which is characterized in that
Threshold value described in step (7) refers to that, for every class ground target, according to Receiver operating curve, searching meets such
The limit value that ground target false alarm rate and discrimination require；By such corresponding ground of element in ground target output probability vector
The limit value of target is as threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810874485.5A CN109190673B (en) | 2018-08-03 | 2018-08-03 | Ground target classification method based on random forest and data rejection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810874485.5A CN109190673B (en) | 2018-08-03 | 2018-08-03 | Ground target classification method based on random forest and data rejection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109190673A true CN109190673A (en) | 2019-01-11 |
| CN109190673B CN109190673B (en) | 2021-09-03 |
Family
ID=64919959
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810874485.5A Active CN109190673B (en) | 2018-08-03 | 2018-08-03 | Ground target classification method based on random forest and data rejection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109190673B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110318327A (en) * | 2019-06-10 | 2019-10-11 | 长安大学 | A kind of surface evenness prediction technique based on random forest |
| CN112147601A (en) * | 2020-09-03 | 2020-12-29 | 南京信息工程大学 | Sea surface small target detection method based on random forest |
| CN112149524A (en) * | 2020-09-04 | 2020-12-29 | 中国人民解放军海军工程大学 | Radar signal sorting and identifying method and device, detector and storage medium |
| CN112835003A (en) * | 2020-12-31 | 2021-05-25 | 西安电子科技大学 | Radar repetition frequency change steady target identification method based on resampling preprocessing |
| CN112926526A (en) * | 2021-03-30 | 2021-06-08 | 矽典微电子(上海)有限公司 | Parking detection method and system based on millimeter wave radar |
| CN112965038A (en) * | 2021-02-02 | 2021-06-15 | 西安电子工程研究所 | Ground moving target classification method based on trace point data and spectral features |
| CN113221631A (en) * | 2021-03-22 | 2021-08-06 | 西安电子科技大学 | Sequence pulse anti-interference target detection method based on convolutional neural network |
| CN113325380A (en) * | 2021-05-28 | 2021-08-31 | 西安电子科技大学 | Air micro-motion target classification online library building method based on Mondrian forest |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103885043A (en) * | 2014-03-31 | 2014-06-25 | 西安电子科技大学 | Airplane target clutter and noise stable classification method based on generalized matched filtering |
| CN103914703A (en) * | 2014-05-12 | 2014-07-09 | 西安电子科技大学 | Classification and identification method for pedestrian and vehicle micro-motion targets |
| CN104849722A (en) * | 2015-05-08 | 2015-08-19 | 南京林业大学 | Light detection and ranging (LiDAR) waveform comprehensive feature-based individual tree identification method |
| US20150301167A1 (en) * | 2009-12-18 | 2015-10-22 | Christopher Gary Sentelle | Detection of movable objects |
| CN105044701A (en) * | 2015-08-05 | 2015-11-11 | 西安电子科技大学 | Ground object classification method based on robustness time frequency characteristics |
| CN106597400A (en) * | 2016-11-15 | 2017-04-26 | 北京无线电测量研究所 | Ground moving vehicle target classification and recognition method and system based on high-resolution distance image |
| CN107092020A (en) * | 2017-04-19 | 2017-08-25 | 北京大学 | Merge the surface evenness monitoring method of unmanned plane LiDAR and high score image |
-
2018
- 2018-08-03 CN CN201810874485.5A patent/CN109190673B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150301167A1 (en) * | 2009-12-18 | 2015-10-22 | Christopher Gary Sentelle | Detection of movable objects |
| CN103885043A (en) * | 2014-03-31 | 2014-06-25 | 西安电子科技大学 | Airplane target clutter and noise stable classification method based on generalized matched filtering |
| CN103914703A (en) * | 2014-05-12 | 2014-07-09 | 西安电子科技大学 | Classification and identification method for pedestrian and vehicle micro-motion targets |
| CN104849722A (en) * | 2015-05-08 | 2015-08-19 | 南京林业大学 | Light detection and ranging (LiDAR) waveform comprehensive feature-based individual tree identification method |
| CN105044701A (en) * | 2015-08-05 | 2015-11-11 | 西安电子科技大学 | Ground object classification method based on robustness time frequency characteristics |
| CN106597400A (en) * | 2016-11-15 | 2017-04-26 | 北京无线电测量研究所 | Ground moving vehicle target classification and recognition method and system based on high-resolution distance image |
| CN107092020A (en) * | 2017-04-19 | 2017-08-25 | 北京大学 | Merge the surface evenness monitoring method of unmanned plane LiDAR and high score image |
Non-Patent Citations (6)
| Title |
|---|
| CARMINE CLEMENTE等: "Developments in Target Micro-Doppler Signatures Analysis:Radar Imaging, Ultrasound and Through-the-Wall Radar", 《HTTPS://CORE.AC.UK/DOWNLOAD/PDF/19542409.PDF》 * |
| LAN DU等: "Noise-Robust Classification of Ground Moving Targets Based on Time-Frequency Feature From Micro-Doppler Signature", 《IEEE SENSORS JOURNAL》 * |
| MIGUEL ANGEL RICO-RAMIREZ等: "Classification of Ground Clutter and Anomalous Propagation Using Dual-Polarization Weather Radar", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
| 李彦兵等: "基于微多普勒特征的地面目标分类", 《电子与信息学报》 * |
| 李彬等: "基于混合概率主成分分析的HRRP特征提取", 《系统工程与电子技术》 * |
| 费大勇: "窄带雷达飞机目标特征提取与分类器设计研究", 《中国优秀硕士学位论文全文数据库_信息科技辑》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110318327A (en) * | 2019-06-10 | 2019-10-11 | 长安大学 | A kind of surface evenness prediction technique based on random forest |
| CN112147601B (en) * | 2020-09-03 | 2023-05-26 | 南京信息工程大学 | Sea surface small target detection method based on random forest |
| CN112147601A (en) * | 2020-09-03 | 2020-12-29 | 南京信息工程大学 | Sea surface small target detection method based on random forest |
| CN112149524A (en) * | 2020-09-04 | 2020-12-29 | 中国人民解放军海军工程大学 | Radar signal sorting and identifying method and device, detector and storage medium |
| CN112149524B (en) * | 2020-09-04 | 2023-09-26 | 中国人民解放军海军工程大学 | Radar signal sorting and identifying method, device, detector and storage medium |
| CN112835003B (en) * | 2020-12-31 | 2023-06-30 | 西安电子科技大学 | Radar repetition frequency variation steady target recognition method based on resampling preprocessing |
| CN112835003A (en) * | 2020-12-31 | 2021-05-25 | 西安电子科技大学 | Radar repetition frequency change steady target identification method based on resampling preprocessing |
| CN112965038A (en) * | 2021-02-02 | 2021-06-15 | 西安电子工程研究所 | Ground moving target classification method based on trace point data and spectral features |
| CN112965038B (en) * | 2021-02-02 | 2023-07-21 | 西安电子工程研究所 | Ground activity target classification method based on trace point data and frequency spectrum characteristics |
| CN113221631A (en) * | 2021-03-22 | 2021-08-06 | 西安电子科技大学 | Sequence pulse anti-interference target detection method based on convolutional neural network |
| CN112926526A (en) * | 2021-03-30 | 2021-06-08 | 矽典微电子(上海)有限公司 | Parking detection method and system based on millimeter wave radar |
| CN112926526B (en) * | 2021-03-30 | 2023-12-29 | 矽典微电子(上海)有限公司 | Parking detection method and system based on millimeter wave radar |
| CN113325380A (en) * | 2021-05-28 | 2021-08-31 | 西安电子科技大学 | Air micro-motion target classification online library building method based on Mondrian forest |
| CN113325380B (en) * | 2021-05-28 | 2024-04-02 | 西安电子科技大学 | Air micro-motion target classification on-line library building method based on Mondrian forest |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109190673B (en) | 2021-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109190673A (en) | The ground target classification method sentenced is refused based on random forest and data | |
| CN106295684B (en) | A kind of dynamic based on micro-Doppler feature is continuous/discontinuous gesture recognition methods | |
| CN105137498B (en) | A kind of the underground objects detection identifying system and method for feature based fusion | |
| CN106127110B (en) | A kind of human body fine granularity motion recognition method based on UWB radar and optimal SVM | |
| CN102043148B (en) | Ground moving target classification method based on general matched filtering | |
| CN107290741B (en) | Indoor human body posture identification method based on weighted joint distance time-frequency transformation | |
| CN101587186B (en) | Characteristic extraction method of radar in-pulse modulation signals | |
| CN106330385A (en) | Interference type identification method | |
| CN102721952B (en) | Method for steadily classifying ground moving target based on super-resolution Doppler spectrum | |
| CN108459311A (en) | Micro-doppler objective classification method under convolutional neural networks based on Hough transform | |
| CN104215959B (en) | Method for estimating radial initial speeds and radial acceleration of multiple maneuvering targets | |
| CN104636609A (en) | Signal combined denoising method based on empirical mode decomposition (EMD) and wavelet analysis | |
| CN107064894A (en) | A kind of clutter suppression method based on deep learning | |
| CN112580486B (en) | Human behavior classification method based on radar micro-Doppler signal separation | |
| CN109901130A (en) | A kind of rotor wing unmanned aerial vehicle detection and recognition methods converting and improve 2DPCA based on Radon | |
| Padar et al. | Classification of human motion using radar micro-Doppler signatures with hidden Markov models | |
| CN105913073A (en) | SAR image target identification method based on depth increment support vector machine | |
| CN108710114A (en) | Turbulent flow object detection method based on BP neural network multicategory classification | |
| CN112068120A (en) | micro-Doppler time-frequency plane individual soldier squad identification method based on two-dimensional Fourier transform | |
| CN105158749A (en) | High-frequency radar sea-clutter amplitude statistical distribution test method | |
| CN103116740B (en) | A kind of Underwater targets recognition and device thereof | |
| CN103576131B (en) | Range extension target Intrinsic Mode characteristic energy Intelligent Fusion detection method | |
| CN113640768B (en) | Low-resolution radar target identification method based on wavelet transformation | |
| CN103605120B (en) | The orderly fusion detection method of range extension target Intrinsic Mode local singular value | |
| CN106199539A (en) | Ground bounce removal method based on prewhitening filter |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |