CN114429585A - Sea surface target multi-feature intelligent identification method - Google Patents
Sea surface target multi-feature intelligent identification method Download PDFInfo
- Publication number
- CN114429585A CN114429585A CN202111606750.XA CN202111606750A CN114429585A CN 114429585 A CN114429585 A CN 114429585A CN 202111606750 A CN202111606750 A CN 202111606750A CN 114429585 A CN114429585 A CN 114429585A
- Authority
- CN
- China
- Prior art keywords
- time
- frequency
- fractal
- target
- polarization
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000010287 polarization Effects 0.000 claims abstract description 51
- 238000001514 detection method Methods 0.000 claims abstract description 32
- 230000009467 reduction Effects 0.000 claims abstract description 16
- 238000009826 distribution Methods 0.000 claims abstract description 11
- 230000002776 aggregation Effects 0.000 claims abstract description 10
- 238000004220 aggregation Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 25
- 238000001228 spectrum Methods 0.000 claims description 17
- 238000012360 testing method Methods 0.000 claims description 12
- 238000013145 classification model Methods 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000012935 Averaging Methods 0.000 claims description 4
- 230000003993 interaction Effects 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 abstract description 18
- 238000005516 engineering process Methods 0.000 abstract description 10
- 238000012800 visualization Methods 0.000 abstract description 4
- 230000000007 visual effect Effects 0.000 abstract description 3
- 238000002592 echocardiography Methods 0.000 abstract description 2
- 238000011946 reduction process Methods 0.000 abstract description 2
- 238000012795 verification Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
Images
Classifications
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a sea surface target multi-feature intelligent identification method, which comprehensively utilizes fractal-time-frequency fourteen-dimensional features, each polarization mode comprises three time-frequency features of time-frequency kurtosis, time-frequency skewness and time-frequency aggregation, visual operation of a dimensionality reduction process and a dimensionality reduction result is realized, a t-distribution random neighbor embedding dimensionality reduction technology is adopted, and finally a support vector machine is combined to realize efficient and accurate learning and identification of target and clutter features. The invention overcomes the problem that the traditional single-domain radar target detection method is difficult to detect due to annihilation of target echoes caused by high coincidence between the target and the clutter domain; the t-SNE dimensionality reduction technology is adopted to realize dimensionality reduction visualization, high-dimensional data is simplified into a visual low-dimensional graph, and the local characteristics of the data are better kept compared with the traditional dimensionality reduction technology; the method has more stable detection performance in a strong clutter environment, algorithm verification is performed by using the disclosed actually-measured sea clutter data set, and the recognition accuracy of the target and the clutter reaches 98.67%.
Description
Technical Field
The invention relates to the technical field of radars, in particular to a target identification method. The method can be used for a shore-based warning radar or a sea searching radar, and realizes the identification of the low-speed sea surface target under the sea clutter background by combining an intelligent method of machine learning.
Background
The radar is generated according to the requirement of long-distance target detection, has the characteristics of all-time, all-weather and easiness in long-distance searching, detecting and tracking targets, plays an extremely important role in air detection and sea detection, is widely used as an important means in modern wars, and also plays an important role in civil detection. The radar has important significance for the detection of weak and small targets in the sea clutter background in the aspects of navigation safety, disaster search and rescue, coast management, homeland safety and the like, and is paid more and more attention.
Different from the earth-to-air detection system, the complex and dynamic working environment of the sea detection system brings a serious challenge to the detection of the sea surface target, and is mainly reflected in that: (1) clutter generation mechanism is abnormally complex, obvious non-Gaussian, non-linear and non-stable three-non complex characteristics are presented, and the difficulty of cognition is increased due to dynamic change; (2) the low-altitude, ultra-low altitude or sea surface low observable targets emerge, echoes of the targets are submerged in strong clutter by means of the natural advantages of the sea clutter, the signal-to-noise-ratio of echo signals is reduced, and the detection and identification difficulty of the targets is greatly increased.
At present, the energy detection method is usually adopted for detecting small sea targets, and differentiation discrimination is carried out on the targets and background clutter by setting an energy threshold. However, for a low observable target with weak target echo energy and highly coincident with a strong clutter in time and frequency domains, it is difficult to realize robust target detection only by a single energy detection means. The time-frequency analysis maps the echo signals to a time-frequency domain space, extracts steady and separable differential description parameters from the time-frequency domain space, and realizes the detection of the target through the difference of the characteristics. But reliable detection under a strong clutter background is difficult to realize only by using single characteristics of a single polarization domain. With the development of nonlinear science, sea clutter is recognized to have many features of chaotic fractal. The fractal features are added to analyze and describe the features of the low observable target from various angles, the stable and separable features are extracted from the time-frequency domain and the fractal domain, the multidimensional joint discrimination feature vector is established, and the refined detection of the target is realized. The multi-polarization fractal-time-frequency domain characteristics can describe the difference of targets and characteristics more comprehensively, more abundantly and more finely, the multi-space multi-scale multi-parameter characteristic description capabilities are fused with each other, a more complete characteristic description system is formed, and the rationality of system detection judgment is improved.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a sea surface target multi-feature intelligent identification method. In order to further improve the target detection performance under the complex covariant environment, the invention provides a sea surface target multi-feature intelligent identification technology which comprehensively utilizes fractal-time-frequency fourteen-dimensional features (namely four polarization modes of HH, HV, VH and VV, wherein H represents horizontal polarization, V represents vertical polarization, each polarization mode comprises three time-frequency features of time-frequency kurtosis, time-frequency skewness and time-frequency aggregation, and comprises two fractal features of AR meter box dimension and intercept, in order to reduce the information redundancy and realize the visualization operation of the dimension reduction process and result, the dimension reduction technology of t-Distributed random Neighbor Embedding (t-SNE) is adopted, and finally the learning and identification of the target and clutter features with high efficiency and accuracy are realized by combining a Support Vector Machine (SVM).
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
the method comprises the following steps: extracting fractal characteristics;
two fractal characteristics are calculated in HH polarization, and the specific steps are as follows:
(1) AR box-counting dimension DAR：
The box-counting dimension of the power spectrum is used for describing the basic characteristics of a fractal set, and the fractal dimension is measured by calculating the number of grids covering the fractal when the side length is delta;
wherein S isuNormalized power spectrum, N, representing a data setδ(Su) Representing the number of bins covering the fractal;
(2) AR spectrum box-dimension curve intercept:
lgN is fitted by least square method by changing the value of deltaδ(Su) The slope of the curve is AR box-counting dimension, and the slope and the intercept of the logarithm-logarithm curve form an AR spectrum joint fractal characteristic parameter;
step two: time-frequency transformation;
four polarization data of HH, HV, VH and VV are sequentially extracted from H, V polarization dual-channel echo signals, and the polarization echo data is expressed as xa(n) performing short-time fourier transform on the echo data to obtain:
the subscript 'a' represents a polarization mode, N belongs to [1, N ] is discrete time sampling, K belongs to [1, K ] is discrete frequency sampling number, h is a window function, m is a window length, and N is a Fourier transform point number;
step three: time-frequency feature extraction
(1) Time-frequency kurtosis Ta,kurtosis：
Kurtosis is used to measure how steep the probability distribution of random variables is;
wherein m isTAnd σTAre respectively a feature vector [ TEa(1),TEa(2),……,TEa(N)]E (-) represents the averaging operation;
(2) time frequency skewness Ta,skeness：
The skewness is used for measuring the asymmetry of random variable probability distribution, the positive or negative of a skewness value represents that statistical data is distributed to the right or left, the greater the absolute value of the skewness value is, the higher the skewness is, and the skewness equal to zero represents that the statistical data are relatively uniformly distributed on the two sides of the average value;
(3) time-frequency concentration Ta,concentration：
The time-frequency aggregation is used for measuring the definition degree of the time resolution and the frequency resolution of the measured area:
step three: preprocessing the characteristic parameters;
dividing a data set into a training set and a test set, respectively calculating two fractal features and three time-frequency features under HH, HV, VH and VV polarization, combining 14 features into a fourteen-dimensional feature vector, and performing t-SNE (space-time interaction) dimension reduction processing on the fourteen-dimensional feature vector;
step four: intelligent target detection
Sending the training set into an SVM for model training, wherein in the SVM training process, the optimal parameters of the classification model are iteratively calculated through an SVM internal parameter optimization algorithm, and the model and the corresponding parameters are output; and finally, sending the test set into an SVM classification model output by training for target detection, and realizing robust recognition of the target and the clutter.
The two fractal characteristics are calculated by adopting two fractal characteristics under VV/HV/VH polarization.
And the window function h takes a Hamming window.
The invention has the advantages that the complementary characteristics of the sea clutter and the target signal in the multi-dimensional domain are utilized, the multi-dimensional joint discrimination characteristic vector is established by carrying out time-frequency transformation and fractal analysis on the echo signal, and the problem that the target echo is difficult to detect due to annihilation of the target echo caused by high coincidence in the target and clutter domains in the traditional single-domain radar target detection method is solved; the t-SNE dimensionality reduction technology is adopted to realize dimensionality reduction visualization, high-dimensional data is simplified into a visual low-dimensional graph, and the local characteristics of the data are better kept compared with the traditional dimensionality reduction technology; in addition, the method also combines multiple polarization characteristics of HH, HV, VH and VV, compared with a single characteristic classification method, the method has more stable detection performance in a strong clutter environment, algorithm verification is carried out by utilizing the disclosed actually-measured sea clutter data set, and the identification accuracy of the target and the clutter reaches 98.67%.
Drawings
Fig. 1 is a graph comparing distributions of targets and clutter in four polarizations of fractal features of the present invention, in which graph (a) of fig. 1 is a graph comparing HH polarization, graph (b) of fig. 1 is a graph comparing HV polarization, graph (c) of fig. 1 is a graph comparing VH polarization, and graph (d) of fig. 1 is a graph comparing VV polarization.
FIG. 2 is a flow chart of an implementation of the sea surface target multi-feature intelligent identification method of the present invention.
FIG. 3 is a time-frequency three-feature comparison of clutter and target cells in four polarizations of the present invention, where (a) in FIG. 3 is a feature comparison diagram in HH polarization, (b) in FIG. 3 is a feature comparison diagram in VV polarization, (c) in FIG. 3 is a feature comparison diagram in HV polarization, (d) in FIG. 3 is a feature comparison diagram in VH polarization,
FIG. 4: and (5) distributing the features in the two-dimensional space after the t-SNE dimension reduction.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
the method comprises the following steps: extracting fractal characteristics;
two fractal characteristics are calculated in HH polarization, and the specific steps are as follows:
(1) AR box-counting dimension DAR：
The box-counting dimension of the power spectrum is used for describing the basic characteristics of a fractal set, and the fractal dimension is measured by calculating the number of grids covering the fractal when the side length is delta;
wherein S isuNormalized power spectrum, N, representing a data setδ(Su) Representing the number of bins covering the fractal;
(2) intercept of AR spectrum box dimension curve
lgN is fitted by least square method by changing the value of deltaδ(Su) A- (-lg delta) curve, wherein the slope of the curve is an AR box-counting dimension, and the slope and the intercept of a logarithm-logarithm curve form an AR spectrum joint fractal characteristic parameter;
step two: time-frequency transformation;
four polarization data of HH, HV, VH and VV are sequentially extracted from H, V polarization dual-channel echo signals, and the polarization echo data is expressed as xa(n) performing short-time fourier transform on the echo data to obtain:
the subscript 'a' represents a polarization mode, N belongs to [1, N ] is discrete time sampling, K belongs to [1, K ] is discrete frequency sampling number, h is a window function, m is a window length, and N is a Fourier transform point number;
step three: time-frequency feature extraction
(1) Time-frequency kurtosis Ta,kurtosis：
Kurtosis is used to measure how steep the probability distribution of random variables is;
wherein m isTAnd σTAre respectively a feature vector [ TEa(1),TEa(2),……,TEa(N)]E (-) represents the averaging operation;
(2) time frequency skewness Ta,skeness：
The skewness is used for measuring the asymmetry of random variable probability distribution, the positive or negative of a skewness value represents that statistical data is distributed to the right or left, the greater the absolute value of the skewness value is, the higher the skewness is, and the skewness equal to zero represents that the statistical data are relatively uniformly distributed on the two sides of the average value;
(3) time-frequency concentration Ta,concentration：
The time-frequency aggregation is used for measuring the definition degree of the time resolution and the frequency resolution of the measured area:
step three: preprocessing the characteristic parameters;
dividing a data set into a training set and a test set, respectively calculating two fractal features and three time-frequency features under HH, HV, VH and VV polarization, combining 14 features into a fourteen-dimensional feature vector, and performing t-SNE (space-time interaction) dimension reduction processing on the fourteen-dimensional feature vector;
step four: intelligent target detection
Sending the training set into an SVM for model training, wherein in the SVM training process, the optimal parameters of the classification model are iteratively calculated through an SVM internal parameter optimization algorithm, and the model and the corresponding parameters are output; and finally, sending the test set into an SVM classification model output by training for target detection, and realizing robust recognition of the target and the clutter.
According to the flow of fig. 2, the specific implementation process of the present invention is as follows:
the sea surface target multi-feature intelligent identification technology is characterized by comprising the following steps of:
the method comprises the following steps: fractal feature extraction
Two fractal features under four polarizations (HH/VV/HV/VH) are extracted, as shown in FIG. 1, the difference of the fractal features under different polarizations is small, and for convenience of explanation, the method calculates the two fractal features under the HH polarization;
(1) AR box-counting dimension DAR：
The box-counting dimension of the power spectrum is used for describing the basic characteristics of a fractal set, and the fractal dimension is measured by calculating the number of lattices covering the fractal when the side length is delta;
wherein S isuNormalized power spectrum, N, representing a data setδ(Su) Representing the number of bins covering the fractal;
(2) intercept of AR spectrum box dimension curve
lgN is fitted by changing the value of delta by least square methodδ(Su) A- (-lg delta) curve, wherein the slope of the curve is an AR box-counting dimension, and the slope and the intercept of a logarithm-logarithm curve form an AR spectrum joint fractal characteristic parameter;
step two: time-frequency transformation;
four kinds of polarization data of HH, HV, VH and VV are sequentially extracted from a dual-channel (H, V polarization) echo signal, and the polarization echo data is expressed as xa(n) performing a short-time fourier transform on the echo data to obtain:
wherein, subscript 'a' represents polarization mode, N belongs to [1, N ] and K belongs to [1, K ] are discrete time sampling and discrete frequency sampling number; 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 three: time-frequency feature extraction
(1) Time-frequency kurtosis Ta,kurtosis：
According to mathematical knowledge of mathematical statistics, kurtosis is used for measuring the steepness degree of probability distribution of random variables; according to the geometrical significance of kurtosis, the fact that the kurtosis value is higher than 3 indicates that the measured curve is sharper than a normal distribution curve, and conversely, the measured curve is short and fat;
wherein m isTAnd σTAre respectively a feature vector [ TEa(1),TEa(2),……,TEa(N)]E (-) represents the averaging operation;
(2) time frequency skewness Ta,skeness：
Skewness is used to measure the asymmetry of the probability distribution of random variables. The positive or negative of the skewness value represents that the statistical data is distributed to the right or left, the greater the absolute value of the skewness value is, the higher the skewness is, and the skewness equal to zero represents that the statistical data is relatively uniformly distributed on the two sides of the average value;
(3) time-frequency concentration Ta,concentration：
The time-frequency aggregation degree is used for measuring the definition degree of the time resolution and the frequency resolution of the measured area. According to the concept of kurtosis of statistics, the short-time Fourier transform X of the detected regionaKurtosis T of (n, k)a,concentrationThe larger the time frequency aggregation is, the higher the corresponding time frequency aggregation is, and the higher the time and frequency resolution of the detected area in the time frequency plane is.
Step three: preprocessing the characteristic parameters;
dividing a data set into a training set and a test set, respectively calculating two fractal features and three time-frequency features under HH, HV, VH and VV polarization, combining the 14 features into a fourteen-dimensional feature vector, and performing t-SNE dimension reduction processing on the fourteen-dimensional feature vector;
step four: intelligent target detection
Sending the training set into an SVM for model training, wherein in the SVM training process, the optimal parameters of the classification model are iteratively calculated through an SVM internal parameter optimization algorithm, and the model and the corresponding parameters are output; and finally, sending the test set into an SVM classification model output by training for target detection, and realizing robust recognition of a target and a clutter, wherein the flow of a processing algorithm is shown in figure 2.
The embodiment adopts IPIX to publicly measure sea clutter data, the data provides echo data of 14 distance units, and each distance unit acquires 131072 pulse echo signals and two polarization channels H and V. Aiming at the data, the technology for intelligently detecting the sea target by multi-polarization of fractal-time frequency feature fusion comprises the following specific steps:
the method comprises the following steps: performing fractal feature combined extraction;
calculation of AR box-counting dimension D under HH polarization according to equation (1)ARFitting a logarithm-logarithm curve, and calculating two fractal characteristics of the intercept of the AR spectrum box dimension curve;
step two: time-frequency transformation and time-frequency feature extraction;
and respectively carrying out short-time Fourier transform on the echo data of the 14 range units, wherein the length of an echo signal is 2048 sampling points, and the window length is 192.
Respectively calculating three characteristics of time-frequency kurtosis, time-frequency skewness and time-frequency aggregation degree of HH, HV, VH and VV data with different polarizations according to formulas (2) to (4), namely THH,kurtosis、THH,skeness、THH,concentration、THV,kurtosis、THV,skeness、 THV,concentration、TVH,kurtosis、TVH,skeness、TVH,concentration、TVV,kurtosis、TVV,skeness、TVV,concentrationTwelve polarization-time frequency characteristics, as shown in fig. 3.
Step three: feature parameter preprocessing
Combining IPIX data prior information, concentrating the main energy of a target in a 7 th distance unit, considering the expansion of the target energy, marking the fractal-time-frequency joint characteristics of the 6 th and 8 th distance units as the target, marking the fractal-time-frequency characteristics calculated by other distance units as clutter, randomly selecting 100 sample points from the clutter units 1-4 and the clutter units 10-14 respectively by a training set and a testing set, randomly selecting 100 sample points from the target units 6-8, totally 1200 sample points, carrying out [ -1, 1] normalization and parameter optimization preprocessing, and then carrying out dimension reduction processing by using a t-SNE method, wherein the processing result is shown in FIG. 4.
Step four: SVM model training and target detection
Inputting the preprocessed training set into a Support Vector Machine (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 the detection and identification of the target. The results of the measurements for different target unit compositions are shown in table 1.
TABLE 1 test results for different target unit compositions
The method provided by the invention is compared and analyzed with the method I and the method II under the same conditions. The first method is fractal dual features (AR box-counting dimension-intercept) under four polarizations (HH/VV/HV/VH), and the second method is time-frequency triple features (time-frequency kurtosis-time-frequency skewness-time-frequency aggregation) under single polarization. The first method, the second method and the identification accuracy of the invention are shown in table 1, 1200 sample points are tested, and the identification accuracy is highest when the target unit is 7, and the accuracy is 98.67%.
In conclusion, due to the combination of polarization and time-frequency domain characteristics and the realization of dimension reduction visualization by using the t-SNE technology, the invention can effectively improve the detection performance of the slow small target under the strong clutter background.
Claims (3)
1. A sea surface target multi-feature intelligent identification method is characterized by comprising the following steps:
the method comprises the following steps: extracting fractal characteristics;
two fractal characteristics are calculated in HH polarization, and the specific steps are as follows:
(1) AR box-counting dimension DAR：
The box-counting dimension of the power spectrum is used for describing the basic characteristics of a fractal set, and the fractal dimension is measured by calculating the number of grids covering the fractal when the side length is delta;
wherein S isuNormalized power spectrum, N, representing a data setδ(Su) Representing the number of bins covering the fractal;
(2) AR spectrum box-dimension curve intercept:
lgN is fitted by least square method by changing the value of deltaδ(Su) A- (-lg delta) curve, wherein the slope of the curve is an AR box-counting dimension, and the slope and the intercept of a logarithm-logarithm curve form an AR spectrum joint fractal characteristic parameter;
step two: time-frequency transformation;
four polarization data of HH, HV, VH and VV are sequentially extracted from H, V polarization dual-channel echo signals, and the polarization echo data is expressed as xa(n) performing short-time fourier transform on the echo data to obtain:
the subscript 'a' represents a polarization mode, N belongs to [1, N ] is discrete time sampling, K belongs to [1, K ] is discrete frequency sampling number, h is a window function, m is a window length, and N is a Fourier transform point number;
step three: time-frequency feature extraction
(1) Time-frequency kurtosis Ta,kurtosis：
Kurtosis is used to measure how steep the probability distribution of random variables is;
wherein m isTAnd σTAre respectively a feature vector [ TEa(1),TEa(2),……,TEa(N)]E (-) represents the averaging operation;
(2) time frequency skewness Ta,skeness：
The skewness is used for measuring the asymmetry of random variable probability distribution, the positive or negative of a skewness value represents that statistical data is distributed to the right or left, the greater the absolute value of the skewness value is, the higher the skewness is, and the skewness equal to zero represents that the statistical data are relatively uniformly distributed on the two sides of the average value;
(3) time-frequency concentration Ta,concentration：
The time-frequency aggregation is used for measuring the definition degree of the time resolution and the frequency resolution of the measured area:
step three: preprocessing the characteristic parameters;
dividing a data set into a training set and a test set, respectively calculating two fractal features and three time-frequency features under HH, HV, VH and VV polarization, combining 14 features into a fourteen-dimensional feature vector, and performing t-SNE (space-time interaction) dimension reduction processing on the fourteen-dimensional feature vector;
step four: intelligent target detection
Sending the training set into an SVM for model training, wherein in the SVM training process, the optimal parameters of the classification model are iteratively calculated through an SVM internal parameter optimization algorithm, and the model and the corresponding parameters are output; and finally, sending the test set into an SVM classification model output by training for target detection, and realizing robust recognition of the target and the clutter.
2. The sea surface target multi-feature intelligent identification method according to claim 1, characterized in that:
the two fractal characteristics are calculated by adopting two fractal characteristics under VV/HV/VH polarization.
3. The sea surface target multi-feature intelligent identification method according to claim 1, characterized in that:
and the window function h takes a Hamming window.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111606750.XA CN114429585A (en) | 2021-12-26 | 2021-12-26 | Sea surface target multi-feature intelligent identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111606750.XA CN114429585A (en) | 2021-12-26 | 2021-12-26 | Sea surface target multi-feature intelligent identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114429585A true CN114429585A (en) | 2022-05-03 |
Family
ID=81311536
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111606750.XA Pending CN114429585A (en) | 2021-12-26 | 2021-12-26 | Sea surface target multi-feature intelligent identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114429585A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111707999A (en) * | 2020-06-15 | 2020-09-25 | 西安电子科技大学 | Sea surface floating small target detection method based on combination of multiple features and ensemble learning |
| CN112327293A (en) * | 2020-10-20 | 2021-02-05 | 西北工业大学 | Sea surface target detection method based on fractal feature intelligent learning |
| WO2021036286A1 (en) * | 2019-08-30 | 2021-03-04 | 华为技术有限公司 | Target behavior recognition method, apparatus and radar system |
| CN112462355A (en) * | 2020-11-11 | 2021-03-09 | 西北工业大学 | Sea target intelligent detection method based on time-frequency three-feature extraction |
-
2021
- 2021-12-26 CN CN202111606750.XA patent/CN114429585A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021036286A1 (en) * | 2019-08-30 | 2021-03-04 | 华为技术有限公司 | Target behavior recognition method, apparatus and radar system |
| CN111707999A (en) * | 2020-06-15 | 2020-09-25 | 西安电子科技大学 | Sea surface floating small target detection method based on combination of multiple features and ensemble learning |
| CN112327293A (en) * | 2020-10-20 | 2021-02-05 | 西北工业大学 | Sea surface target detection method based on fractal feature intelligent learning |
| CN112462355A (en) * | 2020-11-11 | 2021-03-09 | 西北工业大学 | Sea target intelligent detection method based on time-frequency three-feature extraction |
Non-Patent Citations (1)
| Title |
|---|
| 顾智敏;张兴敢;王琼;: "FRFT域内的海杂波多重分形特性与目标检测", 南京大学学报(自然科学), no. 04, 31 August 2017 (2017-08-31), pages 731 - 737 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111580064B (en) | Sea surface small target detection method based on multi-domain and multi-dimensional feature fusion | |
| CN104036239B (en) | Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering | |
| CN102508219B (en) | Turbulent current target detection method of wind profiler radar | |
| CN112462355B (en) | Intelligent sea target detection method based on time-frequency three-feature extraction | |
| CN106599808A (en) | Full-waveform laser radar data based concealed target extraction method | |
| CN103353988B (en) | Allos SAR scene Feature Correspondence Algorithm performance estimating method | |
| CN106569193B (en) | The small targets detection in sea clutter method filtered based on anterior-posterior to income reference particle | |
| CN113516052B (en) | Imaging millimeter wave radar point cloud target classification method based on machine learning | |
| CN105093215A (en) | Doppler information based method for tracking low-altitude low-speed small target through radar | |
| Ningbo et al. | High-dimensional feature extraction of sea clutter and target signal for intelligent maritime monitoring network | |
| CN105372717A (en) | FOD fusion detection method and device based on radar and image signal | |
| CN111505649B (en) | Towed passive array sonar low signal-to-noise ratio ship moving target detection method | |
| CN104569923A (en) | Velocity restraint-based Hough transformation fast track starting method | |
| Lin et al. | Unknown radar waveform recognition based on transferred deep learning | |
| CN104199002B (en) | The radar information processing method of suppression wind-power electricity generation vane false-alarm targets | |
| Li et al. | Pedestrian liveness detection based on mmwave radar and camera fusion | |
| CN106291550A (en) | The polarization SAR Ship Detection of core is returned based on local scattering mechanism difference | |
| CN109389053B (en) | Method and system for detecting position information of vehicle to be detected around target vehicle | |
| CN107861108A (en) | A kind of method for detecting targets at sea based on power spectrum box counting dimension | |
| CN114429585A (en) | Sea surface target multi-feature intelligent identification method | |
| CN111707999A (en) | Sea surface floating small target detection method based on combination of multiple features and ensemble learning | |
| Wang et al. | SAR water image segmentation based on GLCM and wavelet textures | |
| Tang et al. | Floating Small Target Detection Algorithm based on Features Fusion of Full Polarization Information in Sea Clutter | |
| Yu et al. | CNN-based weather signal detection algorithm for airborne weather radar | |
| Yuankui et al. | Automatic target recognition of ISAR images based on Hausdorff distance |
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 |