CN109190673B - Ground target classification method based on random forest and data rejection - Google Patents

Ground target classification method based on random forest and data rejection Download PDF

Info

Publication number
CN109190673B
CN109190673B CN201810874485.5A CN201810874485A CN109190673B CN 109190673 B CN109190673 B CN 109190673B CN 201810874485 A CN201810874485 A CN 201810874485A CN 109190673 B CN109190673 B CN 109190673B
Authority
CN
China
Prior art keywords
value
frequency
maximum value
echo signal
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.)
Active
Application number
CN201810874485.5A
Other languages
Chinese (zh)
Other versions
CN109190673A (en
Inventor
杜兰
李泉
高勇
何浩男
任科
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd
Original Assignee
Xidian University
Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xidian University, Xian Cetc Xidian University Radar Technology Collaborative Innovation Research Institute Co Ltd filed Critical Xidian University
Priority to CN201810874485.5A priority Critical patent/CN109190673B/en
Publication of CN109190673A publication Critical patent/CN109190673A/en
Application granted granted Critical
Publication of CN109190673B publication Critical patent/CN109190673B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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 ground target classification method based on random forest and data rejection, which comprises the following implementation steps: (1) preprocessing a training sample set; (2) extracting a training feature matrix; (3) training a random forest classifier; (4) pretreating a test sample; (5) extracting a test feature vector; (6) calculating a ground target output probability vector; (7) judging whether the test sample refuses to be judged; (8) if the judgment is refused, the test sample is used as an echo signal without the target micro-motion characteristic; (9) and if the judgment is not rejected, outputting the category corresponding to the maximum value in the probability vector as the ground target classification result of the test sample. The invention rejects clutter, deception jamming and target signals without micromotion, improves the classification recognition rate of ground moving targets, and simultaneously adopts a classifier with parallel processing capability to improve the performance of the method in the aspect of real-time property.

Description

Ground target classification method based on random forest and data rejection
Technical Field
The invention belongs to the technical field of communication, and further relates to a ground target classification method based on random forest and data rejection in the technical field of radar signal processing. The invention can classify different vehicle targets and human body targets moving on the ground in real time in the environment with clutter and deceptive trunk.
Background
In radar ground target classification, radar returns may contain clutter, deceptive interference, and ground target signals without micromotion. For clutter, deceptive interference, the classification method should reject the classification; for the target signal without the micro motion, the classification method cannot effectively and correctly classify because the effective micro motion features required by classification cannot be extracted. How to reasonably and effectively remove clutter, deception jamming and target signals without micromotion is one of the difficulties in ground target classification. Meanwhile, the existing ground target classification method of the radar still needs longer processing time, and the real-time performance needs to be improved.
A moving vehicle object classification method is proposed in the patent document 'a moving vehicle object classification method and system' (patent application No.: CN201510738784.2, publication No.: CN105403872A) applied by Beijing antenna electrical measurement research. The method comprises the following specific steps: the method comprises the steps of firstly, obtaining an original Doppler spectrum of a target vehicle, and performing clutter suppression on the original Doppler spectrum to obtain a target Doppler spectrum after clutter suppression; secondly, carrying out speed normalization on the target Doppler spectrum to obtain a normalized Doppler spectrum; thirdly, respectively calculating the distances between the normalized Doppler spectrum and a wheeled vehicle target Doppler spectrum template and between the normalized Doppler spectrum and a tracked vehicle target Doppler spectrum template according to the normalized Doppler spectrum, and respectively obtaining a wheeled distance and a tracked distance; and fourthly, comparing the wheel type distance with the track distance, and judging that the target vehicle is a wheel type vehicle or a track vehicle according to a comparison result. Although the method can process radar signals with high signal-to-noise ratio and identify the tracked vehicle and the wheeled vehicle, the method still has the defects that: because the method only directly compares the distances between the observation Doppler spectrum and the Doppler spectrum template of the wheeled vehicle and the crawler Doppler spectrum template, clutter and deceptive interference can still be judged as a vehicle target in the environment with clutter and deceptive interference; the ground target signals without micromotion are randomly classified, which causes misjudgment of the classification method and reduces the recognition rate.
The university of western electronic technology proposes a ground target classification method based on robustness time-frequency characteristics in the patent document "stability-time-frequency-characteristic-based ground target classification method" (patent application number: cn201510475477.x, publication number: CN 105044701A). The method comprises the following specific steps: normalizing the energy of the acquired high signal-to-noise ratio signal to obtain a training signal; secondly, extracting 3-dimensional time-frequency characteristics from the time-frequency spectrum of the training signal and training a support vector machine classifier; thirdly, normalizing the energy of the collected test signal to obtain a test signal; fourthly, extracting 3-dimensional time-frequency characteristics from the time-frequency spectrum of the test signal; and fifthly, sending the 3-dimensional time-frequency characteristics of the test signals into a trained support vector machine classifier to obtain a classification result. Although the method can process radar signals of ground targets and classify the radar signals, the method still has the following defects: because the method adopts the support vector machine classifier, and the support vector machine is a classifier with a serial structure, the support vector machine classifier needs more processing time, which can cause the performance of the method in the aspect of real-time performance to be poor.
Disclosure of Invention
The invention aims to provide a ground target classification method based on random forest and data rejection aiming at the defects in the prior art.
The idea for realizing the purpose of the invention is that while the ground target narrow-band radar echo signals with micro-motion are correctly classified, the ground target radar echo signals with clutter, deceptive jamming and no micro-motion are rejected, and different vehicle targets and human body targets moving on the ground are effectively classified in real time in the environment with clutter and deceptive jamming. Meanwhile, the classification method of the invention needs shorter processing time and improves the performance in the aspect of real-time performance.
The method comprises the following specific steps:
(1) and preprocessing the training sample set.
(1a) At least 1000 echo signals with micro Doppler effect in frequency spectrum are randomly selected from narrow-band radar echo signals of different ground targets to form a training sample set.
(1b) And performing clutter suppression on the echo signals in the training sample set by using a regional CLEAN method.
(1c) And performing noise suppression on echo signals in the training sample set after clutter suppression by using a global CLEAN method to obtain a preprocessed training sample set.
(2) And extracting a training feature matrix.
(2a) And 7 features of echo signals in the preprocessed training sample set are extracted by adopting a feature extraction method to form a training feature matrix.
(2b) And according to a normalized training formula, performing normalization processing on elements of the training feature matrix to obtain a normalized training feature matrix.
(3) And training a random forest classifier.
And inputting the normalized training feature matrix into a random forest classifier for training until the number of decision trees subjected to parallel processing in the random forest classifier is more than 500, and stopping training to obtain a trained random forest classifier with parallel processing capability.
(4) The test sample is pre-treated.
(4a) And taking an echo signal received by the narrow-band radar in real time as a test sample.
(4b) And performing clutter suppression on the echo signal of the test sample by using a regional CLEAN method.
(4c) And performing noise suppression on the echo signal of the test sample after clutter suppression by using a global CLEAN method to obtain a preprocessed test sample.
(5) And extracting the test feature vector.
(5a) And 7 features of echo signals in the preprocessed test sample are extracted by a feature extraction method to form a test feature vector.
(5b) And according to a normalization test formula, performing normalization processing on elements of the test characteristic vector to obtain a normalized test characteristic vector.
(6) And calculating a ground target output probability vector.
(6a) And inputting the normalized test feature vector into a trained random forest classifier with parallel processing capability to obtain a judgment result of each decision tree in the random forest classifier.
(6b) And calculating the output probability of each type of ground target according to an output probability formula, and forming the output probability of all types of ground targets into a ground target output probability vector.
(7) Judging whether elements of the ground target output probability vector are all smaller than a threshold value, if so, executing the step (8); otherwise, step (9) is performed.
(8) And (4) taking the test sample as an echo signal without the micro-motion characteristic of the ground target, and executing the step (4) after rejecting the test sample.
(9) And taking the category corresponding to the maximum value in the ground target output probability vector as the ground target classification result of the test sample.
Compared with the prior art, the invention has the following advantages:
firstly, judging whether elements of a ground target output probability vector are all smaller than a threshold value, if so, taking a test sample as an echo signal without ground target micro-motion characteristics, refusing to judge the test sample, and then re-collecting the test sample; otherwise, the corresponding category of the maximum value in the ground target output probability vector is used as the ground target classification result of the test sample, so that the problem of low recognition rate caused by the fact that the clutter and the deceptive interference are still judged as the ground target and the ground target signals without micromotion are randomly classified in the environment with the clutter and the deceptive interference in the prior art is solved, and the method has the advantages of reducing misjudgment and improving the classification recognition rate.
Secondly, the invention utilizes the trained random forest classifier with parallel processing capability to classify, thereby overcoming the problem that the ground target classification method is poorer in real-time performance due to the fact that the support vector machine classifier which needs more processing time is utilized to classify in the prior art, ensuring that the processing time needed by the invention is shorter, and the performance in real-time performance is improved.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The implementation steps of the present invention are further described below with reference to fig. 1.
Step 1, preprocessing a training sample set.
At least 1000 echo signals with micro Doppler effect in frequency spectrum are randomly selected from narrow-band radar echo signals of different ground targets to form a training sample set.
And performing clutter suppression on the echo signals in the training sample set by using a regional CLEAN method.
The method for area CLEAN comprises the following specific steps:
and step 1, estimating the ground clutter energy in the radar echo according to the radar parameters.
And 2, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal, and taking a clutter frequency spectrum range as a clutter region.
And 3, searching the maximum value of the clutter region, the phase corresponding to the maximum value of the clutter region, the amplitude corresponding to the maximum value of the clutter region and the Doppler frequency corresponding to the maximum value of the clutter region in the Doppler spectrum.
And 4, reconstructing a time domain signal corresponding to the maximum value of the clutter region according to the following formula:
Figure BDA0001752922660000041
wherein s (t) represents the signal amplitude of the reconstructed clutter region maximum value corresponding to the time t in the time domain signal, R represents the amplitude corresponding to the clutter region maximum value, K represents the number of discrete Fourier transform points, exp represents the exponential operation with natural number as the base, j represents the imaginary unit, pi represents the circumference ratio, d represents the Doppler frequency corresponding to the clutter region maximum value,
Figure BDA0001752922660000042
the phase corresponding to the maximum value of the clutter region is represented.
And 5, subtracting the time domain signal corresponding to the maximum value of the reconstructed clutter region from the echo signal to obtain a processed echo signal.
And 6, calculating the energy of the processed echo signal in the clutter region.
Step 7, judging whether the energy of the processed echo signal in the clutter area is smaller than the ground clutter energy, if so, obtaining the echo signal after clutter suppression; otherwise, executing step 2.
And performing noise suppression on echo signals in the training sample set after clutter suppression by using a global CLEAN method to obtain a preprocessed training sample set.
The specific steps of the global CLEAN method are as follows:
and step 1, estimating noise energy in radar echo according to radar parameters.
And 2, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal.
And 3, searching the maximum value of the Doppler spectrum, the phase corresponding to the maximum value of the Doppler spectrum, the amplitude corresponding to the maximum value of the Doppler spectrum and the Doppler frequency corresponding to the maximum value of the Doppler spectrum in the Doppler spectrum.
And 4, reconstructing a time domain signal corresponding to the maximum value of the Doppler spectrum according to the following formula:
Figure BDA0001752922660000051
wherein z (t) represents the signal amplitude at time t in the time domain signal corresponding to the maximum value of the reconstructed doppler spectrum, Y represents the amplitude corresponding to the maximum value of the doppler spectrum, m represents the doppler frequency corresponding to the maximum value of the doppler spectrum, and θ represents the phase corresponding to the maximum value of the doppler spectrum.
And 5, subtracting the time domain signal corresponding to the maximum value of the reconstructed Doppler spectrum from the echo signal to obtain a processed echo signal.
And 6, calculating the energy of the processed echo signal.
Step 7, judging whether the energy of the processed echo signal is greater than the noise energy, if so, executing the step 2; otherwise, executing step 8.
And 8, subtracting the processed echo signal from the echo signal subjected to clutter suppression to obtain the echo signal subjected to noise suppression.
And 2, extracting a training feature matrix.
And 7 features of echo signals in the preprocessed training sample set are extracted by adopting a feature extraction method to form a training feature matrix.
The specific steps of the feature extraction method are as follows:
step 1, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal.
And 2, searching the maximum value of the Doppler spectrum, the phase corresponding to the maximum value of the Doppler spectrum, the amplitude corresponding to the maximum value of the Doppler spectrum and the Doppler frequency corresponding to the maximum value of the Doppler spectrum in the Doppler spectrum.
And 3, reconstructing a time domain signal corresponding to the maximum value of the Doppler spectrum.
And 4, subtracting the time domain signal corresponding to the maximum value of the reconstructed Doppler spectrum from the echo signal to obtain a residual echo signal.
And 5, performing discrete Fourier transform on the residual echo signals to obtain Doppler spectrums of the residual echo signals.
And 6, performing three discrete Fourier transforms on the residual echo signals by utilizing the steps 1 to 5 to obtain Doppler spectrums of the three residual echo signals.
And 7, calculating the value of the first characteristic according to the following formula:
Figure BDA0001752922660000061
wherein D is1A value representing the first characteristic, n the number of frequency points in the Doppler spectrum, U2(n) denotes the amplitude, U, of the nth frequency point in the Doppler spectrum of the second residual echo signal0(n) represents the amplitude of the nth frequency point in the doppler spectrum of the echo signal.
And 8, calculating the value of the second characteristic according to the following formula:
Figure BDA0001752922660000062
wherein D is2A value representing a second characteristic, U3(n) represents the amplitude of the nth frequency point in the doppler spectrum of the third residual echo signal.
And 9, performing linear normalization on the echo signals with the square sum of 1.
And step 10, performing sliding window with the step length of 1 on the normalized echo signal by adopting a window which is rounded downwards and is 0.5 time of the echo signal length to obtain a sliding window matrix.
And 11, performing autocorrelation operation on the sliding window matrix to obtain an autocorrelation matrix.
And 12, performing eigenvalue decomposition on the autocorrelation matrix to obtain an eigenvalue sequence.
And step 13, sequencing the characteristic value sequences according to a non-increasing sequence to obtain a characteristic spectrum.
Step 14, calculating the value of the third feature according to the following formula:
Figure BDA0001752922660000071
wherein D is3A value representing a third feature, x represents the number of alternatives for the feature value in the feature spectrum,
Figure BDA0001752922660000072
representing the value of x corresponding to the minimum value of the element satisfying the requirement of being greater than 0.98, r representing the number of the characteristic value in the characteristic spectrum, λrRepresenting the r-th eigenvalue in the characteristic spectrum, M representing the signal length,
Figure BDA0001752922660000073
denotes the rounding-down operation, b denotes the number of the characteristic values in the characteristic spectrum, λbRepresenting the b-th eigenvalue in the characteristic spectrum.
Step 15, calculating a value of the fourth feature according to the following formula:
Figure BDA0001752922660000074
wherein D is4The value representing the fourth feature, ln (-) represents the log operation.
And 16, performing short-time Fourier transform on the time domain echo signal to obtain an amplitude spectrum.
Step 17, normalizing the amplitude spectrum according to the following formula:
Figure BDA0001752922660000075
wherein S (f, a) represents a normalized amplitude of frequency f and time a,
Figure BDA0001752922660000076
representing the amplitude of frequency F and time a in the amplitude spectrum, F representing the frequency maximum, F representing the frequency in the amplitude spectrum, A representing the time domain maximum, a tableTime in the amplitude spectrum is shown.
And step 18, calculating time-frequency entropy according to the following formula:
Figure BDA0001752922660000077
wherein e (f) represents the time-frequency entropy with frequency f.
Step 19, calculating a time-frequency entropy mean value according to the following formula:
Figure BDA0001752922660000078
wherein,
Figure BDA0001752922660000081
representing the time-frequency entropy mean.
Step 20, calculating a value of the fifth feature according to the following formula:
Figure BDA0001752922660000082
wherein D is5A value representing a fifth characteristic is determined,
Figure BDA0001752922660000083
indicating the frequency f corresponding to the maximum value.
Step 21, the value of the fifth characteristic is within the range from the maximum value of the frequency
Figure BDA0001752922660000084
And
Figure BDA0001752922660000085
the frequency of the condition closest to the value of the fifth feature is taken as the upper limit of the subject frequency.
Step 22, the frequency minimum value is within the range of the value of the fifth characteristic
Figure BDA0001752922660000086
And
Figure BDA0001752922660000087
the frequency of the condition closest to the value of the fifth feature is taken as the lower limit of the subject frequency.
Step 23, calculating a value of the sixth feature according to the following formula:
Figure BDA0001752922660000088
wherein D is6Denotes the value of the sixth feature, w denotes the upper frequency limit of the body, and q denotes the lower frequency limit of the body.
Step 24, the value of the fifth characteristic is within the range from the maximum value of the frequency
Figure BDA0001752922660000089
And
Figure BDA00017529226600000810
the frequency of the condition closest to the value of the fifth feature is taken as the upper limit doppler frequency.
Step 25, the frequency minimum value is within the range of the value of the fifth characteristic, and the requirement is met
Figure BDA00017529226600000811
And
Figure BDA00017529226600000812
the frequency closest to the value of the fifth feature under the condition is taken as the doppler lower limit frequency.
And 26, subtracting the Doppler lower limit frequency from the Doppler upper limit frequency to obtain a seventh characteristic value.
And according to a normalized training formula, performing normalization processing on elements of the training feature matrix to obtain a normalized training feature matrix.
The normalized training formula is as follows:
Figure BDA00017529226600000813
wherein,
Figure BDA0001752922660000091
a j normalized feature, o, representing the i echo signal in the normalized training feature matrixi,jRepresents the jth feature of the ith echo signal in the training feature matrix, min represents the minimum value operation, max represents the maximum value operation, ojAnd a vector composed of j-th features representing all echo signals in the training feature matrix.
And 3, training a random forest classifier.
And inputting the normalized training feature matrix into a random forest classifier for training until the number of decision trees subjected to parallel processing in the random forest classifier is more than 500, and stopping training to obtain a trained random forest classifier with parallel processing capability.
And 4, preprocessing the test sample.
And taking an echo signal received by the narrow-band radar in real time as a test sample.
And performing clutter suppression on the echo signal of the test sample by using a regional CLEAN method.
The specific steps of the zone CLEAN method are as follows.
And step 1, estimating the ground clutter energy in the radar echo according to the radar parameters.
And 2, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal, and taking a clutter frequency spectrum range as a clutter region.
And 3, searching the maximum value of the clutter region, the phase corresponding to the maximum value of the clutter region, the amplitude corresponding to the maximum value of the clutter region and the Doppler frequency corresponding to the maximum value of the clutter region in the Doppler spectrum.
And 4, reconstructing a time domain signal corresponding to the maximum value of the clutter region according to the following formula:
Figure BDA0001752922660000092
wherein s (t) represents the signal amplitude of the reconstructed clutter region maximum value corresponding to the time t in the time domain signal, R represents the amplitude corresponding to the clutter region maximum value, K represents the number of discrete Fourier transform points, exp represents the exponential operation with natural number as the base, j represents the imaginary unit, pi represents the circumference ratio, d represents the Doppler frequency corresponding to the clutter region maximum value,
Figure BDA0001752922660000093
the phase corresponding to the maximum value of the clutter region is represented.
And 5, subtracting the time domain signal corresponding to the maximum value of the reconstructed clutter region from the echo signal to obtain a processed echo signal.
And 6, calculating the energy of the processed echo signal in the clutter region.
Step 7, judging whether the energy of the processed echo signal in the clutter area is smaller than the ground clutter energy, if so, obtaining the echo signal after clutter suppression; otherwise, executing step 2.
And performing noise suppression on the echo signal of the test sample after clutter suppression by using a global CLEAN method to obtain a preprocessed test sample.
The specific steps of the global CLEAN method are as follows:
and step 1, estimating noise energy in radar echo according to radar parameters.
And 2, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal.
Step 3, searching the maximum value of the Doppler spectrum, the phase corresponding to the maximum value of the Doppler spectrum, the amplitude corresponding to the maximum value of the Doppler spectrum and the Doppler frequency corresponding to the maximum value of the Doppler spectrum in the Doppler spectrum; .
And 4, reconstructing a time domain signal corresponding to the maximum value of the Doppler spectrum according to the following formula:
Figure BDA0001752922660000101
wherein z (t) represents the signal amplitude at time t in the time domain signal corresponding to the maximum value of the reconstructed doppler spectrum, Y represents the amplitude corresponding to the maximum value of the doppler spectrum, m represents the doppler frequency corresponding to the maximum value of the doppler spectrum, and θ represents the phase corresponding to the maximum value of the doppler spectrum.
And 5, subtracting the time domain signal corresponding to the maximum value of the reconstructed Doppler spectrum from the echo signal to obtain a processed echo signal.
And 6, calculating the energy of the processed echo signal.
Step 7, judging whether the energy of the processed echo signal is greater than the noise energy, if so, executing the step 2; otherwise, executing step 8.
And 8, subtracting the processed echo signal from the echo signal subjected to clutter suppression to obtain the echo signal subjected to noise suppression.
And 5, extracting the test feature vector.
And 7 features of echo signals in the preprocessed test sample are extracted by a feature extraction method to form a test feature vector.
The specific steps of the feature extraction method are as follows:
step 1, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal.
And 2, searching the maximum value of the Doppler spectrum, the phase corresponding to the maximum value of the Doppler spectrum, the amplitude corresponding to the maximum value of the Doppler spectrum and the Doppler frequency corresponding to the maximum value of the Doppler spectrum in the Doppler spectrum.
And 3, reconstructing a time domain signal corresponding to the maximum value of the Doppler spectrum.
And 4, subtracting the time domain signal corresponding to the maximum value of the reconstructed Doppler spectrum from the echo signal to obtain a residual echo signal.
And 5, performing discrete Fourier transform on the residual echo signals to obtain Doppler spectrums of the residual echo signals.
And 6, performing three discrete Fourier transforms on the residual echo signals by utilizing the first step to the fifth step to obtain Doppler spectrums of the three residual echo signals.
And 7, calculating the value of the first characteristic according to the following formula:
Figure BDA0001752922660000111
wherein D is1A value representing the first characteristic, n the number of frequency points in the Doppler spectrum, U2(n) denotes the amplitude, U, of the nth frequency point in the Doppler spectrum of the second residual echo signal0(n) represents the amplitude of the nth frequency point in the doppler spectrum of the echo signal.
And 8, calculating the value of the second characteristic according to the following formula:
Figure BDA0001752922660000112
wherein D is2A value representing a second characteristic, U3(n) represents the amplitude of the nth frequency point in the doppler spectrum of the third residual echo signal.
And 9, performing linear normalization on the echo signals with the square sum of 1.
And step 10, performing sliding window with the step length of 1 on the normalized echo signal by adopting a window which is rounded downwards and is 0.5 time of the echo signal length to obtain a sliding window matrix.
And 11, performing autocorrelation operation on the sliding window matrix to obtain an autocorrelation matrix.
And 12, performing eigenvalue decomposition on the autocorrelation matrix to obtain an eigenvalue sequence.
And step 13, sequencing the characteristic value sequences according to a non-increasing sequence to obtain a characteristic spectrum.
Step 14, calculating the value of the third feature according to the following formula:
Figure BDA0001752922660000113
wherein D is3A value representing a third feature, x represents the number of alternatives for the feature value in the feature spectrum,
Figure BDA0001752922660000121
representing the value of x corresponding to the minimum value of the element satisfying the requirement of being greater than 0.98, r representing the number of the characteristic value in the characteristic spectrum, λrRepresenting the r-th eigenvalue in the characteristic spectrum, M representing the signal length,
Figure BDA0001752922660000122
denotes the rounding-down operation, b denotes the number of the characteristic values in the characteristic spectrum, λbRepresenting the b-th eigenvalue in the characteristic spectrum.
Step 15, calculating a value of the fourth feature according to the following formula:
Figure BDA0001752922660000123
wherein D is4The value representing the fourth feature, ln (-) represents the log operation.
And 16, performing short-time Fourier transform on the time domain echo signal to obtain an amplitude spectrum.
Step 17, normalizing the amplitude spectrum according to the following formula:
Figure BDA0001752922660000124
wherein S (f, a) represents a normalized amplitude of frequency f and time a,
Figure BDA0001752922660000125
representing the amplitude of the amplitude spectrum with frequency F and time a, F representing the frequency maximum, F representing the frequency in the amplitude spectrum, a representing the time domain maximum, a representing the time in the amplitude spectrum.
And step 18, calculating time-frequency entropy according to the following formula:
Figure BDA0001752922660000126
wherein e (f) represents the time-frequency entropy with frequency f.
Step 19, calculating a time-frequency entropy mean value according to the following formula:
Figure BDA0001752922660000127
wherein,
Figure BDA0001752922660000128
representing the time-frequency entropy mean.
Step 20, calculating a value of the fifth feature according to the following formula:
Figure BDA0001752922660000131
wherein D is5A value representing a fifth characteristic is determined,
Figure BDA0001752922660000132
indicating the frequency f corresponding to the maximum value.
Step 21, the value of the fifth characteristic is within the range from the maximum value of the frequency
Figure BDA0001752922660000133
And
Figure BDA0001752922660000134
the frequency of the condition closest to the value of the fifth feature is taken as the upper limit of the subject frequency.
Step 22, the frequency minimum value is within the range of the value of the fifth characteristic
Figure BDA0001752922660000135
And
Figure BDA0001752922660000136
conditional from a fifth featureThe frequency with the most recent value serves as the lower limit of the subject frequency.
Step 23, calculating a value of the sixth feature according to the following formula:
Figure BDA0001752922660000137
wherein D is6Denotes the value of the sixth feature, w denotes the upper frequency limit of the body, and q denotes the lower frequency limit of the body.
Step 24, the value of the fifth characteristic is within the range from the maximum value of the frequency
Figure BDA0001752922660000138
And
Figure BDA0001752922660000139
the frequency of the condition closest to the value of the fifth feature is taken as the upper limit doppler frequency.
Step 25, the frequency minimum value is within the range of the value of the fifth characteristic, and the requirement is met
Figure BDA00017529226600001310
And
Figure BDA00017529226600001311
the frequency closest to the value of the fifth feature under the condition is taken as the doppler lower limit frequency.
And 26, subtracting the Doppler lower limit frequency from the Doppler upper limit frequency to obtain a seventh characteristic value.
And according to a normalization test formula, performing normalization processing on elements of the test characteristic vector to obtain a normalized test characteristic vector.
The normalized test formula is as follows:
Figure BDA00017529226600001312
wherein,
Figure BDA00017529226600001313
representing the jth normalized feature, p, in the normalized test feature vectorjRepresenting the jth feature in the test feature vector.
And 6, calculating a ground target output probability vector.
And inputting the normalized test feature vector into a trained random forest classifier with parallel processing capability to obtain a judgment result of each decision tree in the random forest classifier.
And calculating the output probability of each type of ground target according to an output probability formula, and forming the output probability of all types of ground targets into a ground target output probability vector.
The output probability formula is as follows:
Figure BDA0001752922660000141
wherein J (c) represents the output probability of the class c ground target class, v represents the number of decision trees in the random forest classifier, Σ represents summation operation, l represents the number of decision trees in the random forest classifier, G (-) represents an indicative function, hlAnd representing the judgment result of the first decision tree in the random forest classifier.
Step 7, judging whether elements of the ground target output probability vector are all smaller than a threshold value, if so, executing step 8; otherwise, step 9 is executed.
The threshold value is a limit value which meets the requirements of false alarm rate and recognition rate of each type of ground target according to the working characteristic curve of a subject; and taking the limiting value of the ground target of the type corresponding to the element in the ground target output probability vector as a threshold value.
And 8, taking the test sample as an echo signal without the micro-motion characteristic of the ground target, and executing the step 4 after rejecting the test sample.
And 9, taking the category corresponding to the maximum value in the ground target output probability vector as the ground target classification result of the test sample.
The effect of the present invention will be further described with reference to simulation experiments.
1. Simulation experiment conditions are as follows:
the parameters of the narrow-band radar in the simulation experiment of the invention are as follows: repetition frequency 2131Hz, pulse accumulation number 128; the software platform is as follows: windows 10 operating system and Matlab R2017 a.
2. Simulation experiment contents:
the simulation experiment content of the invention is to utilize narrow-band radar to collect echo signals of three ground targets, namely a wheeled vehicle, a tracked vehicle and a human. The motion directions of the wheeled vehicle and the tracked vehicle comprise approaching radar, being away from the radar, turning around and turning, and the motion directions of people comprise approaching radar and being away from the radar. Randomly selecting echo signals with micro Doppler effect in 6000 frequency spectrums to form a training sample set, wherein the number of each type of samples is not less than 600; and the test sample comprises clutter, deceptive interference, a ground target echo signal without micro motion and a ground target radar echo signal with micro motion. In a simulation experiment, the method and the conventional classification method are adopted to classify 30279 test samples of three types of ground targets respectively. The conventional classification method is different from the present invention in that the conventional classification method uses a support vector machine classifier and does not reject the classification result.
3. And (3) simulation result analysis:
in order to evaluate the method and the conventional classification method, the test recognition rate of each type of ground target of the simulation experiment is respectively calculated according to the following formula:
Figure BDA0001752922660000151
wherein gamma (c) represents the test recognition rate of the class c ground target,
Figure BDA0001752922660000152
the number of the test samples for correctly classifying the class c ground target is represented, and ξ (c) represents the total number of the non-rejection test samples of the class c ground target.
The average recognition rate is the average of the test recognition rates of all classes of ground targets.
The identification rates of radar echo test samples in the environments with clutter and deceptive jamming by the two methods of the simulation experiment are listed as follows:
TABLE 1 results of conventional classification methods and test identification comparison List of the present invention
Wheeled vehicle Crawler vehicle Human being Average recognition rate
Conventional classification method 62.42％ 62.56％ 70.68％ 65.22％
The method of the invention 88.65％ 84.08％ 92.23％ 88.32％
As can be seen from the average recognition rate in table 1, the method of the present invention has significant advantages over the conventional classification method in the presence of clutter and deceptive interference. Specifically, compared with the conventional classification method, when the method is used for classifying the ground targets, the test recognition rate of the three types of ground targets is increased by over 21 percent, so that the overall recognition performance is greatly improved. Because the method and the conventional method adopt the same clutter suppression, noise suppression and feature extraction methods, the performance improvement is mainly brought by data rejection, namely the data rejection can effectively remove clutter, deceptive interference and radar echo test samples without micro-moving targets, and the conventional method can receive the clutter, the deceptive interference and the radar echo test samples without the micro-moving targets and carry out random classification, which can cause the identification rate and the average identification rate of each type of ground targets to be reduced. In table 1, compared with the conventional classification method, the test recognition rates of different moving vehicle targets and human body targets are improved, and the performance improvement amplitudes are similar, which indicates that the data rejection is not limited to the categories of the ground targets, and the classification recognition rates of all categories in the test sample can be integrally improved.

Claims (8)

1. A ground target classification method based on random forest and data rejection is characterized in that whether elements of a ground target output probability vector are all smaller than a threshold value or not is judged, and a trained random forest classifier with parallel processing capacity is provided; the method comprises the following specific steps:
(1) preprocessing a training sample set:
(1a) randomly selecting at least 1000 echo signals with micro Doppler effect in a frequency spectrum from narrow-band radar echo signals of different ground targets to form a training sample set;
(1b) performing clutter suppression on echo signals in the training sample set by using a regional CLEAN method;
(1c) performing noise suppression on echo signals in the training sample set after clutter suppression by using a global CLEAN method to obtain a preprocessed training sample set;
(2) extracting a training feature matrix:
(2a) extracting 7 characteristics of echo signals in the preprocessed training sample set by adopting a characteristic extraction method to form a training characteristic matrix;
(2b) according to a normalized training formula, carrying out normalization processing on elements of a training feature matrix to obtain a normalized training feature matrix;
(3) training a random forest classifier:
inputting the normalized training feature matrix into a random forest classifier for training until the number of decision trees subjected to parallel processing in the random forest classifier is more than 500, and stopping training to obtain a trained random forest classifier with parallel processing capacity;
(4) pretreating a test sample:
(4a) taking an echo signal received by a narrow-band radar in real time as a test sample;
(4b) performing clutter suppression on an echo signal of a test sample by using a regional CLEAN method;
(4c) performing noise suppression on the echo signal of the test sample after clutter suppression by using a global CLEAN method to obtain a preprocessed test sample;
(5) extracting a test feature vector:
(5a) extracting 7 characteristics of echo signals in the preprocessed test sample by adopting a characteristic extraction method to form a test characteristic vector;
(5b) according to a normalization test formula, carrying out normalization processing on elements of the test characteristic vector to obtain a normalized test characteristic vector;
(6) calculating a ground target output probability vector:
(6a) inputting the normalized test feature vector into a trained random forest classifier with parallel processing capacity to obtain a judgment result of each decision tree in the random forest classifier;
(6b) calculating the output probability of each type of ground target according to an output probability formula, and forming the output probabilities of all types of ground targets into a ground target output probability vector;
(7) judging whether elements of the ground target output probability vector are all smaller than a threshold value, if so, executing the step (8); otherwise, executing step (9);
(8) taking the test sample as an echo signal without the micro-motion characteristic of the ground target, and executing the step (4) after rejecting the test sample;
(9) and taking the category corresponding to the maximum value in the ground target output probability vector as the ground target classification result of the test sample.
2. The method for classifying ground targets based on random forests and data rejections as claimed in claim 1, wherein the specific steps of the regional CLEAN method in the steps (1b) and (4b) are as follows:
the method comprises the steps of firstly, estimating ground clutter energy in radar echoes according to radar parameters;
secondly, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal, and taking a clutter frequency spectrum range as a clutter region;
thirdly, searching the maximum value of the clutter region, the phase corresponding to the maximum value of the clutter region, the amplitude corresponding to the maximum value of the clutter region and the Doppler frequency corresponding to the maximum value of the clutter region in the Doppler spectrum;
fourthly, reconstructing a time domain signal corresponding to the maximum value of the clutter region according to the following formula:
Figure FDA0003125326950000021
wherein s (t) represents the signal amplitude of the reconstructed clutter region maximum value corresponding to the time t in the time domain signal, R represents the amplitude corresponding to the clutter region maximum value, K represents the point number of discrete Fourier transform, exp represents the exponential operation with natural number as the base, j represents the imaginary number unit, pi represents the circumference ratio, d represents the Doppler frequency corresponding to the clutter region maximum value, and theta represents the phase corresponding to the clutter region maximum value;
fifthly, subtracting a time domain signal corresponding to the maximum value of the reconstructed clutter region from the echo signal to obtain a processed echo signal;
sixthly, calculating the energy of the processed echo signal in the clutter area;
seventhly, judging whether the energy of the processed echo signal in the clutter area is smaller than the ground clutter energy, if so, obtaining the echo signal after clutter suppression; otherwise, the second step is executed.
3. The ground target classification method based on random forest and data rejection as claimed in claim 2, wherein the global CLEAN method in step (1c) and step (4c) comprises the following specific steps:
firstly, estimating noise energy in radar echo according to radar parameters;
secondly, performing discrete Fourier transform on the echo signal to obtain a Doppler spectrum of the echo signal;
thirdly, searching the maximum value of the Doppler spectrum, the phase corresponding to the maximum value of the Doppler spectrum, the amplitude corresponding to the maximum value of the Doppler spectrum and the Doppler frequency corresponding to the maximum value of the Doppler spectrum in the Doppler spectrum;
fourthly, reconstructing a time domain signal corresponding to the maximum value of the Doppler spectrum according to the following formula:
Figure FDA0003125326950000031
wherein z (t) represents the signal amplitude at the time t in the time domain signal corresponding to the maximum value of the reconstructed Doppler spectrum, Y represents the amplitude corresponding to the maximum value of the Doppler spectrum, m represents the Doppler frequency corresponding to the maximum value of the Doppler spectrum, and theta represents the phase corresponding to the maximum value of the Doppler spectrum;
fifthly, subtracting a time domain signal corresponding to the maximum value of the reconstructed Doppler spectrum from the echo signal to obtain a processed echo signal;
sixthly, calculating the energy of the processed echo signal;
seventhly, judging whether the energy of the processed echo signal is greater than the noise energy, and if so, executing the second step; otherwise, executing the eighth step;
and eighthly, subtracting the processed echo signal from the echo signal subjected to clutter suppression to obtain the echo signal subjected to noise suppression.
4. The method for classifying ground targets based on random forests and data rejection as claimed in claim 2, wherein the specific steps of the feature extraction method in the steps (2a) and (5a) are as follows:
firstly, performing discrete Fourier transform on an echo signal to obtain a Doppler spectrum of the echo signal;
secondly, searching the maximum value of the Doppler spectrum, the phase corresponding to the maximum value of the Doppler spectrum, the amplitude corresponding to the maximum value of the Doppler spectrum and the Doppler frequency corresponding to the maximum value of the Doppler spectrum in the Doppler spectrum;
thirdly, reconstructing a time domain signal corresponding to the maximum value of the Doppler spectrum;
fourthly, subtracting a time domain signal corresponding to the maximum value of the reconstructed Doppler spectrum from the echo signal to obtain a residual echo signal;
fifthly, performing discrete Fourier transform on the residual echo signals to obtain Doppler spectrums of the residual echo signals;
sixthly, performing three discrete Fourier transforms on the residual echo signals by utilizing the first step to the fifth step to obtain Doppler spectrums of the three residual echo signals;
the seventh step, according to the following formula, calculates the value of the first characteristic:
Figure FDA0003125326950000041
wherein D is1A value representing the first characteristic, n the number of frequency points in the Doppler spectrum, U2(n) denotes the amplitude, U, of the nth frequency point in the Doppler spectrum of the second residual echo signal0(n) represents the amplitude of the nth frequency point in the doppler spectrum of the echo signal;
eighth, calculating a value of the second feature according to the following formula:
Figure FDA0003125326950000042
wherein D is2A value representing a second characteristic, U3(n) represents the amplitude of the nth frequency point in the doppler spectrum of the third residual echo signal;
the ninth step, carry on the square sum to the echo signal for the linear normalization of 1;
tenth, a window with 0.5 time echo signal length after being rounded downwards is adopted for the normalized echo signals to perform sliding window with the step length of 1, and a sliding window matrix is obtained;
the eleventh step, carrying out autocorrelation operation on the sliding window matrix to obtain an autocorrelation matrix;
performing eigenvalue decomposition on the autocorrelation matrix to obtain an eigenvalue sequence;
step thirteen, sorting the characteristic value sequence according to a non-increasing order to obtain a characteristic spectrum;
fourteenth, the value of the third feature is calculated according to the following formula:
Figure FDA0003125326950000051
wherein D is3A value representing a third feature, x represents the number of alternatives for the feature value in the feature spectrum,
Figure FDA0003125326950000052
representing the value of x corresponding to the minimum value of the element satisfying the requirement of being greater than 0.98, r representing the number of the characteristic value in the characteristic spectrum, λrRepresenting the r-th eigenvalue in the characteristic spectrum, M representing the signal length,
Figure FDA0003125326950000053
denotes the rounding-down operation, b denotes the number of the characteristic values in the characteristic spectrum, λbRepresenting the b-th characteristic value in the characteristic spectrum;
a fifteenth step of calculating a value of the fourth feature according to the following equation:
Figure FDA0003125326950000054
wherein D is4A value representing a fourth feature, ln (-) represents a log operation;
sixthly, performing short-time Fourier transform on the time domain echo signal to obtain an amplitude spectrum;
seventeenth, normalizing the magnitude spectrum according to the following formula:
Figure FDA0003125326950000055
wherein S (f, a) represents a normalized amplitude of frequency f and time a,
Figure FDA0003125326950000056
representing the amplitude with frequency F and time a in the amplitude spectrum, wherein F represents the maximum value of the frequency, F represents the frequency in the amplitude spectrum, A represents the maximum value of the time domain, and a represents the time in the amplitude spectrum;
eighteenth, calculating the time-frequency entropy according to the following formula:
Figure FDA0003125326950000057
wherein E (f) represents the time-frequency entropy with frequency f;
the nineteenth step, calculating the time-frequency entropy mean value according to the following formula:
Figure FDA0003125326950000058
wherein,
Figure FDA0003125326950000061
representing a time-frequency entropy mean value;
twentieth, calculating a value of the fifth feature according to the following formula:
Figure FDA0003125326950000062
wherein D is5A value representing a fifth characteristic is determined,
Figure FDA0003125326950000063
represents the frequency f corresponding to the maximum value;
a twentieth step of obtaining a frequency maximum value within a range from the value of the fifth feature
Figure FDA0003125326950000064
And
Figure FDA0003125326950000065
the frequency of the condition closest to the value of the fifth feature is taken as the upper limit of the subject frequency;
a twentieth step of satisfying the condition from the frequency minimum value to a value of the fifth feature
Figure FDA0003125326950000066
And
Figure FDA0003125326950000067
the frequency of the condition closest to the value of the fifth feature is taken as a lower limit of the main frequency;
a twenty-third step of calculating a value of the sixth feature according to the following formula:
Figure FDA0003125326950000068
wherein D is6A value representing a sixth characteristic, w represents a body frequency upper limit, and q represents a body frequency lower limit;
a twenty-fourth step of converting the value at the fifth feature to a frequencyWithin a maximum value range of
Figure FDA0003125326950000069
And
Figure FDA00031253269500000610
the frequency of the condition closest to the value of the fifth feature is taken as the upper limit frequency of Doppler;
twenty-fifth step, within the range from the frequency minimum value to the value of the fifth characteristic, satisfying
Figure FDA00031253269500000611
And
Figure FDA00031253269500000612
the frequency closest to the value of the fifth feature under the condition is taken as the lower limit frequency of Doppler;
and twenty-sixth step, subtracting the Doppler lower limit frequency from the Doppler upper limit frequency to obtain a seventh characteristic value.
5. The method for classifying ground targets based on random forests and data rejections as claimed in claim 1, wherein the normalized training formula in step (2b) is as follows:
Figure FDA0003125326950000071
wherein,
Figure FDA0003125326950000072
a j normalized feature, o, representing the i echo signal in the normalized training feature matrixi,jRepresents the jth feature of the ith echo signal in the training feature matrix, min represents the minimum value operation, max represents the maximum value operation, ojAnd a vector composed of j-th features representing all echo signals in the training feature matrix.
6. The method for classifying ground targets based on random forests and data rejections as claimed in claim 5, wherein the normalized test formula in step (5b) is as follows:
Figure FDA0003125326950000073
wherein,
Figure FDA0003125326950000074
representing the jth normalized feature, p, in the normalized test feature vectorjRepresenting the jth feature in the test feature vector.
7. The method for classifying ground targets based on random forests and data rejections as claimed in claim 1, wherein the output probability formula in step (6b) is as follows:
Figure FDA0003125326950000075
wherein J (c) represents the output probability of the class c ground target class, v represents the number of decision trees in the random forest classifier, Σ represents summation operation, l represents the number of decision trees in the random forest classifier, G (-) represents an indicative function, hlAnd representing the judgment result of the first decision tree in the random forest classifier.
8. The method for classifying ground targets based on random forests and data rejection as claimed in claim 1, wherein the threshold in step (7) is that for each type of ground target, a limit value meeting the requirements of false alarm rate and recognition rate of the type of ground target is searched according to the working characteristic curve of the subject; and taking the limiting value of the ground target of the type corresponding to the element in the ground target output probability vector as a threshold value.
CN201810874485.5A 2018-08-03 2018-08-03 Ground target classification method based on random forest and data rejection Active CN109190673B (en)

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 CN109190673A (en) 2019-01-11
CN109190673B true 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)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
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
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
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
CN113221631B (en) * 2021-03-22 2023-02-10 西安电子科技大学 Sequence pulse anti-interference target detection method based on convolutional neural network
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
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

Citations (6)

* Cited by examiner, † Cited by third party
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
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

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150301167A1 (en) * 2009-12-18 2015-10-22 Christopher Gary Sentelle Detection of movable objects

Patent Citations (6)

* Cited by examiner, † Cited by third party
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
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)

* Cited by examiner, † Cited by third party
Title
Classification of Ground Clutter and Anomalous Propagation Using Dual-Polarization Weather Radar;Miguel Angel Rico-Ramirez等;《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》;20080731;第46卷(第7期);1892-1904 *
Developments in Target Micro-Doppler Signatures Analysis:Radar Imaging, Ultrasound and Through-the-Wall Radar;Carmine Clemente等;《https://core.ac.uk/download/pdf/19542409.pdf》;20141231;1-22 *
Noise-Robust Classification of Ground Moving Targets Based on Time-Frequency Feature From Micro-Doppler Signature;Lan Du等;《IEEE Sensors Journal》;20140328;第14卷(第8期);2672-2682 *
基于微多普勒特征的地面目标分类;李彦兵等;《电子与信息学报》;20101215;第32卷(第12期);2848-2853 *
基于混合概率主成分分析的HRRP特征提取;李彬等;《***工程与电子技术》;20161115;第39卷(第01期);1-7 *
窄带雷达飞机目标特征提取与分类器设计研究;费大勇;《中国优秀硕士学位论文全文数据库_信息科技辑》;20180415(第04期);I136-2431 *

Also Published As

Publication number Publication date
CN109190673A (en) 2019-01-11

Similar Documents

Publication Publication Date Title
CN109190673B (en) Ground target classification method based on random forest and data rejection
CN108564005B (en) Human body falling identification method based on convolutional neural network
CN106778610B (en) Intra-pulse modulation identification method based on time-frequency image characteristics
CN112882009B (en) Radar micro Doppler target identification method based on amplitude and phase dual-channel network
CN101470194B (en) Torpedo target recognition method
CN106330385A (en) Interference type identification method
CN111488801A (en) Ship classification method based on vibration noise identification
CN111428631B (en) Visual identification and sorting method for unmanned aerial vehicle flight control signals
CN113640768B (en) Low-resolution radar target identification method based on wavelet transformation
CN111175718A (en) Time-frequency domain combined ground radar automatic target identification method and system
CN101587186A (en) Characteristic extraction method of radar in-pulse modulation signals
CN110514975A (en) A kind of shelf depreciation genre classification methods of GIS device
CN104680183B (en) SAR target discrimination methods based on scattering point and the centers K oneclass classification device
Yang et al. Ship recognition via its radiated sound: The fractal based approaches
CN109164429B (en) Multi-fractal characteristic aircraft target classification method based on principal component analysis
CN107015207A (en) Active pressing jamming classifying identification method based on FRFT domains peak value discrete feature
CN105158749A (en) High-frequency radar sea-clutter amplitude statistical distribution test method
CN112731331B (en) Micro-motion target noise steady identification method based on signal-to-noise ratio adaptive network
CN103605120B (en) The orderly fusion detection method of range extension target Intrinsic Mode local singular value
CN101644768A (en) Torpedo target recognition method based on cepstrum analysis
CN115061094B (en) Radar target recognition method based on neural network and SVM
Raja Abdullah et al. Improvement in detection with forward scattering radar
DE102012106790A1 (en) Method for determining scattering center of object by processing of digital data, particularly for automotive driver assistance applications, involves roughly estimating relevant characteristics of object by processing of digital data
CN105718856A (en) EEMD-based vehicle micro-tremor signal extraction and classification method
Sobolewski et al. Automatic modulation recognition techniques based on cyclostationary and multifractal features for distinguishing LFM, PWM and PPM waveforms used in radar systems as example of artificial intelligence implementation in test

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