CN112462355B - Intelligent sea target detection method based on time-frequency three-feature extraction - Google Patents
Intelligent sea target detection method based on time-frequency three-feature extraction Download PDFInfo
- Publication number
- CN112462355B CN112462355B CN202011251746.1A CN202011251746A CN112462355B CN 112462355 B CN112462355 B CN 112462355B CN 202011251746 A CN202011251746 A CN 202011251746A CN 112462355 B CN112462355 B CN 112462355B
- Authority
- CN
- China
- Prior art keywords
- time
- frequency
- target
- clutter
- kurtosis
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- 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
- G01S13/02—Systems using reflection of radio waves, e.g. primary radar systems; Analogous systems
-
- 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
- G01S7/418—Theoretical aspects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Optimization (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Algebra (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides an intelligent detection method for a sea target based on time-frequency three-feature extraction, which adopts a short-time Fourier analysis method to map the mixed echo data of the target and strong clutter into a two-dimensional plane of time and frequency, and realizes fine feature extraction and stable target detection by means of the time-frequency characteristic difference of the target and the clutter. The invention utilizes the difference of target clutter and combines a machine learning method to realize the reliable detection of the target under the background of strong sea clutter. The problem that targets and clutter are overlapped in one-dimensional domains such as time domain or frequency domain and are difficult to separate is effectively solved, robust detection of a slow moving target is achieved in a time-frequency domain, and the method has higher accuracy and detection efficiency.
Description
Technical Field
The invention relates to the technical field of radars, in particular to a detection method based on time-frequency three characteristics, which is suitable for detecting a slow moving target under a complex clutter background.
Background
The detection research of the ground/sea clutter background target is from simple to complex, the dimension of signal processing is continuously expanded, and the development process of single domain processing from time, frequency and space is carried out, two-dimensional processing from time, frequency and space is carried out, and the development process of multi-domain processing from space, time and frequency is carried out. In terms of radar offshore target detection technology, fixed threshold detection was proposed in the 40 s of the 20 th century, however, with increasing complexity of the background environment, in order to meet the requirements of radar on false alarm rate control, constant false alarm rate detection technology is proposed and gradually applied to radar equipment. In the 50 s of the 20 th century, a transform domain detection method represented by short-time Fourier transform appears, and then a representation domain processing technology is gradually developed on the basis of compressed sensing and sparse representation. In the 90 s of the 20 th century, nonlinear science represented by fractal and chaos began to be applied to target detection of sea radars and became a very active and important branch.
The method for detecting the target of the sea radar can be roughly summarized into the following ways: (1) The time-frequency analysis clutter suppression method is used for selecting a proper time-frequency analysis method according to the motion state of a moving target to be detected, converting the moving target to a corresponding time-frequency domain, and detecting and judging according to the amplitude value or the time-varying rule of a time-frequency curve of the moving target; (2) The micro Doppler signal characteristic extraction technology of the moving target takes micro characteristics of the target as research targets, and micro Doppler reflects Doppler change characteristics, thereby providing a new path for radar target characteristic extraction and identification; (3) The moving target signal fraction domain coherent accumulation processing technology (fractional Fourier transform, FRFT) can accumulate uniformly accelerated moving target signals, and is suitable for detection and parameter estimation of low observable moving targets; (4) According to the moving target high-resolution sparse time-frequency representation domain processing technology, an SFRFT (Short-time fractional Fourier transform) method is provided by combining the advantages of FRFT (fractional Fourier transform) by using a sparse Fourier transform method, the background noise and clutter are suppressed while the high-resolution signal spectrum characteristics are obtained, and the operand in large data volume processing is remarkably reduced compared with an FRFT method; (5) The multi-means moving target information sensing and fusion technology utilizes an acoustic, optical, electric and magnetic multi-sensor target comprehensive monitoring system to sense moving target information through five stages of distributed detection fusion, position fusion, attribute fusion, situation assessment and threat assessment; (6) The intelligent learning and recognition technology for the moving target features based on deep learning describes the moving feature parameters and the moving state of the target through mathematical modeling, and the recognition accuracy is improved by training a model through intelligent learning ideas such as deep learning.
In general, the trend of radar toward sea exploration technology can be divided into two aspects: on the one hand, multidimensional information is comprehensively utilized, high-dimensional refined information is fused and utilized, echo signals can be more refined and described, and the performance of detection, estimation, identification, evaluation and decision making is improved. On the other hand, radar intelligent target detection is characterized in that in the process of feature selection and extraction, the artificial intelligent method can acquire deep-layer essential feature information of the target, and a more abstract high-level representation attribute category or feature is formed by combining low-level features so as to find out the fine feature representation of data, thereby being beneficial to improving the recognition accuracy of the target and clutter.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an intelligent detection method for a sea target based on time-frequency three-feature extraction. In order to solve the problem of insufficient identification accuracy of the existing single-domain-based slow target detection method, the invention adopts a short-time Fourier analysis method to map the mixed echo data of the target and the strong clutter into a two-dimensional plane of time and frequency, and realizes refined feature extraction and stable target detection by means of the time-frequency characteristic difference of the target and the clutter. The invention provides an intelligent detection method for sea targets based on time-frequency three-feature extraction on the basis of fully analyzing time-frequency features of measured data, and realizes reliable detection of targets under the background of strong sea clutter by utilizing the difference of target clutter and combining a machine learning method.
The technical scheme adopted by the invention for solving the technical problems comprises the following specific steps:
step one: time-frequency two-dimensional transformation;
let the radar echo signal to be tested be denoted x (n), its short-time fourier transform be denoted as:
where n and k are the discrete time samples and the discrete frequency samples; h is a window function; m is the window length; n is the number of Fourier transform points;
step two: extracting time-frequency characteristics;
firstAccumulating the spectral function values in each time slice along the time axis of the short-time Fourier transform to calculate the spectral function value T_E of the nth time slice n :
The total short-time Fourier transform spectrum function value E of the time-frequency plane is as follows:
on the basis, the time-frequency characteristics are extracted as follows:
(1) Time-frequency kurtosis T kurtosis
Time-frequency kurtosis T kurtosis Expressed as:
wherein m is T Sum sigma T Respectively the feature vectors T_E 1 ,T_E 2 ,……,T_E n ]E (·) represents the function of the signal mean;
(2) Time-frequency offset T skeness
Time-frequency offset T skeness Expressed as:
(3) Time-frequency aggregation degree v
The ratio of four-mode to two-mode of X (n, k) places a large amount of signal energy in a small region of the time-frequency domain:
the high value of the time-frequency aggregation degree v indicates that the signal has better frequency aggregation degree in a time-frequency plane;
step three: model training
Dividing the data set into a training set and a testing set, respectively calculating time-frequency kurtosis, time-frequency deviation and frequency aggregation according to the data set and the testing set, sending the training set into a machine learning network for model training, and determining network optimal parameters by iterative calculation of the model through an algorithm; and finally, testing the current network by adopting test data, outputting a predicted result of the test data by the network, comparing the predicted result with an actual result, and counting the target detection accuracy, wherein when the error of the target detection accuracy and the expected accuracy is within a preset range, the machine learning network training is successful.
And h is a window function, and a Hamming window is taken.
In the third step, a support vector machine classification method (Support Vector Machine Classification, SVM) is adopted, the trained SVM model is used as a decision device for target detection output, whether a target exists in a current unit to be detected is judged, and robust identification of the target and clutter is achieved.
The method has the advantages that the method utilizes the difference of the refined characteristics of clutter and target signals in the time-frequency domain, can effectively solve the problem that targets and clutter are difficult to separate due to high overlapping in one-dimensional domains such as the time domain or the frequency domain, and realizes the robust detection of slow moving targets in the time-frequency domain. Compared with the traditional target identification method based on single domain features, the method can extract the refined features of the target, and the algorithm verification is carried out by using the disclosed actually measured sea clutter data, so that the classification accuracy of the target and the clutter reaches 98.92%.
The invention creatively provides the extraction and analysis of radar target characteristics from the time-frequency domain, and training and learning of the classification model are carried out through the SVM, so that the intelligent detection of the low-speed small target under the sea clutter background is finally realized. Compared with the traditional radar target detection technology (such as CAFR) and the existing radar target detection technology based on single domain features, the identification method has higher accuracy and detection efficiency.
Drawings
Fig. 1 is a short-time fourier transform diagram of (a) a clutter unit, and fig. 1 (b) is a short-time fourier transform diagram of a target unit.
Fig. 2 is a flowchart of the sea target detection technique based on SVM time-frequency feature extraction of the present invention.
Fig. 3 is a schematic diagram of a matrix of classification characteristic parameters.
Fig. 4 (a) is a three-dimensional distribution contrast diagram of target and clutter unit features, fig. 4 (b) is a time-frequency analysis result, and fig. 4 (c) is a frequency-domain analysis result.
Detailed Description
The invention will be further described with reference to the drawings and examples.
Step one: time-frequency two-dimensional transformation;
let the radar echo signal to be tested be denoted x (n), its short-time fourier transform be denoted as:
where n and k are the discrete time samples and the discrete frequency samples; h is a window function, and a Hamming window is taken; m is the window length; n is the number of Fourier transform points;
step two: extracting time-frequency characteristics;
firstly, accumulating the spectral function values in each time slice along the time axis of short-time Fourier transform to calculate the spectral function value T_E of the nth time slice n :
The total short-time Fourier transform spectrum function value E of the time-frequency plane is as follows:
on the basis, the time-frequency characteristics are extracted as follows:
(1) Time-frequency kurtosis T kurtosis
Kurtosis is a characteristic parameter describing a probability density curve and reflects the peak value of the curve at the mean value position; from probabilityOn the density distribution diagram, the sharpness of the peak is visually displayed by kurtosis, the kurtosis value of the test sample point needs to be compared with the normal distribution, and according to the correlation theory of probability and statistics, when the kurtosis value is more than 3, the sample peak is steeper than the normal distribution peak, and vice versa. Time-frequency kurtosis T kurtosis Expressed as:
wherein m is T Sum sigma T Respectively the feature vectors T_E 1 ,T_E 2 ,……,T_E n ]E (·) represents the function of the signal mean;
(2) Time-frequency offset T skeness
The deviation represents the characteristic number of the asymmetry degree of the probability distribution density curve relative to the average value, and the deviation is larger than zero and represents the deviation right side of the statistical data distribution, and the deviation degree is higher as the deviation value is larger; similarly, the deviation is smaller than zero, which represents that the distribution of the statistical data is on the left side, and the deviation degree is higher as the deviation value is smaller; the deviation equal to zero represents a more uniform distribution of statistical data, and the time-frequency deviation T skeness Expressed as:
(3) Time-frequency aggregation degree v
The ratio of four-mode to two-mode of X (n, k) places a large amount of signal energy in a small region of the time-frequency domain:
the high value of the time-frequency aggregation degree v indicates that the signal has better frequency aggregation degree in a time-frequency plane;
step three: model training
Dividing the data set into a training set and a testing set, respectively calculating time-frequency kurtosis, time-frequency deviation and frequency aggregation according to the data set and the testing set, sending the training set into a machine learning network for model training, and determining network optimal parameters by iterative calculation of the model through an algorithm; and finally, testing the current network by adopting test data, outputting a predicted result of the test data by the network, comparing the predicted result with an actual result, and counting the target detection accuracy, wherein when the error of the target detection accuracy and the expected accuracy is within a preset range, the machine learning network training is successful.
The invention adopts a support vector machine classification method (Support Vector Machine Classification, SVM), takes the trained SVM model as a decision device for target detection output, judges whether a target exists in a current unit to be detected, realizes the robust identification of the target and clutter, and a processing algorithm flow chart is shown in figure 2.
The following example process refers to fig. 2.
This example uses IPIX published data that provides echo data for 14 range bins, each of which acquired 131072 pulse echo signals. For the data, the sea target detection technology based on time-frequency three-feature extraction specifically comprises the following steps:
step one: time-frequency two-dimensional transformation
The echo data of the 14 distance units are respectively subjected to short-time Fourier transform, wherein the echo signal length is selected to be 2048, and the window length is set to be 192.
Step two: time-frequency feature extraction
In order to obtain a sufficient time-frequency three-feature data set and a test sample set, sliding window segmentation is carried out on each distance unit data to carry out short-time Fourier transform, and a training set and a test set of short-time Fourier transform of a 1 st distance unit are taken as an example, three features are obtained as follows:
by analogy, calculating 1 group of time-frequency three characteristics by sliding 50 discrete sampling points each time, wherein in the experiment, 2500 time-frequency three characteristic samples are collected in total by a 1 st distance unit; then, the 14 units are traversed in turn, and finally, a 14×2500 three-feature matrix can be constructed, and the arrangement manner of the three-feature matrix is shown in fig. 3 (each black square represents the value of the feature parameter corresponding to the segment).
Step three: SVM model training
Combining prior information, the main energy of the target is in the 7 th distance unit, three characteristics of the 6 th to 8 th distance units can be selected to be marked as target characteristics in consideration of the expansion of the target, then the test set and the training set randomly select 100 sample points from clutter units 1 to 4 and clutter units 10 to 14 respectively, 100 sample points are randomly selected from target units 6 to 8 respectively, 1200 sample points are altogether, and [ -1,1] normalization and parameter optimization preprocessing are carried out.
Step four: data testing
And inputting the preprocessed training set into an SVM for classification training, and training to obtain a classification model. And sending the preprocessed test set into a classification model for testing to obtain a classification result, and completing detection and identification of the target.
The time-frequency characteristic is used to distinguish between target and background sea clutter, and for comparison, the same distance cell is tested in three unipolar modes (HH/VV/HV), where 'H' represents horizontal polarization and 'V' represents vertical polarization. And (3) extracting the data of each single polarization according to the step (1) and the step (2) to obtain a characteristic matrix of each polarization mode, and taking a training set and a testing set from the characteristic matrix and sending the training set and the testing set into the SVM for training and testing. The distinction of objects and clutter can be considered a classification problem. The results of the measurements at the different polarizations are shown in table 1.
TABLE 1 detection results of different polarizations
Polarization mode | Normalized accuracy | Normalized false alarm rate |
HH polarization | 96.0833% | 0.6667% |
VV polarization | 98.25% | 0.3333% |
HV polarization | 98.9167% | 0% |
As shown in Table 1, the number of sample points in the test set under single polarization is 1200, the highest recognition accuracy is HV polarization, the accuracy is 98.92%, and the false alarm rate is 0%. The ideal recognition accuracy is achieved, the false alarm rate is well controlled, and the target recognition function is realized.
In order to test the influence of feature dimensions on target recognition, different dual-polarized features are taken from the same distance unit under HH polarization for testing. The data of each test method are extracted according to the step 1 and the step 2 to obtain a dual-polarized feature matrix, and a training set and a test set are taken and sent into the SVM for training and testing. Under the same condition, extracting the characteristic matrix according to the step 1 and the step 2, taking a training set and a testing set from the characteristic matrix, sending the training set and the testing set into an SVM for training and testing, and calculating the normalized recognition accuracy of the tripolar characteristic for comparison and reference. The results of the tests under the different test methods are shown in Table 2.
Table 2 test results of different test methods under HH polarization
As can be seen from Table 2, the three-feature recognition accuracy is improved compared with the two-feature recognition accuracy in different test methods, and the recognition accuracy of the pure target unit is greater than that of the target unit combined with the target extension unit. Because the SVM classification problem is essentially a problem of searching an optimal classification hyperplane to meet the maximum classification interval, the higher the dimension of the features is, the larger the optimal classification interval is after the features are mapped to a high-dimensional space, and the identification accuracy is improved to a certain extent; however, the improvement is not infinite, the increase of the feature dimension causes the increase of the information redundancy and the calculation complexity, and the recognition accuracy exhibits a marginal decreasing effect. Therefore, the method adopts the tripolar characteristic as the classification characteristic combination under the premise of considering the actual application background.
The simulation results are shown in fig. 4. The results of frequency domain analysis, time domain analysis and time frequency analysis feature extraction target recognition are shown in the figure. As can be seen from the figure, as the target and the clutter are highly overlapped in the time domain and the frequency domain, the effective extraction of the target and the clutter is difficult to realize by adopting the traditional one-dimensional time domain, the traditional one-dimensional frequency domain and other processing methods, the clutter features are smaller than the target feature values in the time-frequency domain, and the features are linearly separable in the high-dimensional space; for the monopole target detection method based on time-frequency domain analysis, effective recognition of targets and clutter is realized, and higher recognition accuracy is achieved.
In conclusion, the method has good target detection performance, and the radar can detect the small targets on the sea surface at a low speed.
Claims (3)
1. The intelligent detection method for the sea target based on time-frequency three-feature extraction is characterized by comprising the following steps of:
step one: time-frequency two-dimensional transformation;
let the radar echo signal to be tested be denoted x (n), its short-time fourier transform be denoted as:
where n and k are the discrete time samples and the discrete frequency samples; h is a window function; m is the window length; n is the number of Fourier transform points;
step two: extracting time-frequency characteristics;
firstly, accumulating the spectral function values in each time slice along the time axis of short-time Fourier transform to calculate the spectral function value T_E of the nth time slice n :
The total short-time Fourier transform spectrum function value E of the time-frequency plane is as follows:
on the basis, the time-frequency characteristics are extracted as follows:
(1) Time-frequency kurtosis T kurtosis
Time-frequency kurtosis T kurtosis Expressed as:
wherein m is T Sum sigma T Respectively the feature vectors T_E 1 ,T_E 2 ,……,T_E n ]E (·) represents the function of the signal mean;
(2) Time-frequency offset T skeness
Time-frequency offset T skeness Expressed as:
(3) Time-frequency aggregation degree v
The ratio of four-mode to two-mode of X (n, k) places a large amount of signal energy in a small region of the time-frequency domain:
the high value of the time-frequency aggregation degree v indicates that the signal has better frequency aggregation degree in a time-frequency plane;
step three: model training
Dividing the data set into a training set and a testing set, respectively calculating time-frequency kurtosis, time-frequency deviation and frequency aggregation according to the data set and the testing set, sending the training set into a machine learning network for model training, and determining network optimal parameters by iterative calculation of the model through an algorithm; and finally, testing the current network by adopting test data, outputting a predicted result of the test data by the network, comparing the predicted result with an actual result, and counting the target detection accuracy, wherein when the error of the target detection accuracy and the expected accuracy is within a preset range, the machine learning network training is successful.
2. An intelligent detection method for a sea target based on time-frequency three-feature extraction by using the method disclosed in claim 1, which is characterized by comprising the following steps:
and h is a window function, and a Hamming window is taken.
3. An intelligent detection method for a sea target based on time-frequency three-feature extraction by using the method disclosed in claim 1, which is characterized by comprising the following steps:
in the third step, a support vector machine classification method is adopted, a trained SVM model is used as a decision device for target detection output, whether a target exists in a current unit to be detected is judged, and robust identification of the target and clutter is achieved.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011251746.1A CN112462355B (en) | 2020-11-11 | 2020-11-11 | Intelligent sea target detection method based on time-frequency three-feature extraction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011251746.1A CN112462355B (en) | 2020-11-11 | 2020-11-11 | Intelligent sea target detection method based on time-frequency three-feature extraction |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112462355A CN112462355A (en) | 2021-03-09 |
CN112462355B true CN112462355B (en) | 2023-07-14 |
Family
ID=74825419
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011251746.1A Active CN112462355B (en) | 2020-11-11 | 2020-11-11 | Intelligent sea target detection method based on time-frequency three-feature extraction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112462355B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114019236A (en) * | 2021-10-12 | 2022-02-08 | 中国科学院空天信息创新研究院 | Power grid harmonic single-channel aliasing target signal detection method and device |
CN114429585A (en) * | 2021-12-26 | 2022-05-03 | 西北工业大学 | Sea surface target multi-feature intelligent identification method |
CN114488058B (en) * | 2022-01-07 | 2024-05-10 | 武汉大学 | Micro clutter removal method based on STFT spectrogram sliding window cancellation |
CN114638262B (en) * | 2022-03-09 | 2024-02-27 | 西北工业大学 | Radar sea clutter intelligent suppression method based on time-frequency two-dimensional feature learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN106526568A (en) * | 2016-12-29 | 2017-03-22 | 中国人民解放军海军航空工程学院 | Radar moving target detection method based on short-time sparse fractional Fourier transform (ST-SFRFT) |
CN109901129A (en) * | 2019-03-06 | 2019-06-18 | 中国人民解放军海军航空大学 | Object detection method and system in a kind of sea clutter |
WO2020042708A1 (en) * | 2018-08-31 | 2020-03-05 | 大象声科(深圳)科技有限公司 | Time-frequency masking and deep neural network-based sound source direction estimation method |
-
2020
- 2020-11-11 CN CN202011251746.1A patent/CN112462355B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN106526568A (en) * | 2016-12-29 | 2017-03-22 | 中国人民解放军海军航空工程学院 | Radar moving target detection method based on short-time sparse fractional Fourier transform (ST-SFRFT) |
WO2020042708A1 (en) * | 2018-08-31 | 2020-03-05 | 大象声科(深圳)科技有限公司 | Time-frequency masking and deep neural network-based sound source direction estimation method |
CN109901129A (en) * | 2019-03-06 | 2019-06-18 | 中国人民解放军海军航空大学 | Object detection method and system in a kind of sea clutter |
Non-Patent Citations (1)
Title |
---|
基于深度自编码多维特征融合的慢动目标检测;张文涛;许治国;郑霖;杨超;;计算机工程(第05期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN112462355A (en) | 2021-03-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112462355B (en) | Intelligent sea target detection method based on time-frequency three-feature extraction | |
CN105137498B (en) | A kind of the underground objects detection identifying system and method for feature based fusion | |
CN106896306B (en) | A kind of GIS oscillatory surge pressure test signal antinoise method | |
Yan et al. | Robust target detection within sea clutter based on graphs | |
CN111580064B (en) | Sea surface small target detection method based on multi-domain and multi-dimensional feature fusion | |
Tang et al. | Feature parameters extraction of GIS partial discharge signal with multifractal detrended fluctuation analysis | |
CN106569193B (en) | The small targets detection in sea clutter method filtered based on anterior-posterior to income reference particle | |
CN103941254A (en) | Soil physical property classification recognition method and device based on geological radar | |
CN103777189A (en) | Radar weak target detecting method based on information geometry multiple autoregressive model | |
Li et al. | Radar signal recognition algorithm based on entropy theory | |
Ramirez Jr et al. | Machine learning for seismic signal processing: Phase classification on a manifold | |
Ningbo et al. | High-dimensional feature extraction of sea clutter and target signal for intelligent maritime monitoring network | |
CN109342828A (en) | A kind of lightening pulse signal detecting method based on frequency domain constant false alarm | |
CN107132518B (en) | A kind of range extension target detection method based on rarefaction representation and time-frequency characteristics | |
CN107861108B (en) | A kind of method for detecting targets at sea based on power spectrum box counting dimension | |
CN116466408B (en) | Artificial neural network superbedrock identification method based on aeromagnetic data | |
CN103149555B (en) | Self-adaptive moving target detection method capable of combining polarized classification and power grouping | |
CN106019250B (en) | Based on angle scintillations relay type decoy discrimination method | |
CN106338722A (en) | High-resolution radar one-dimension range profile target identification method based on multi-order samples | |
CN114594463A (en) | Sea surface small target feature detection method based on combined convex hull | |
CN113093135B (en) | Target detection method and device based on F norm normalization distance | |
CN113030892B (en) | Sea surface small target detection method based on high-dimensional feature domain gradient lifting tree | |
CN114415118B (en) | Sea surface target detection method based on two-dimensional fractal characteristics | |
Shen et al. | Fractal detector design and application in maritime target detection | |
CN116224271B (en) | Sea surface weak target detection method based on polarized scattering characteristics |
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 |