CN103091668A - Sea surface small target detection method based on chaos theory - Google Patents
Sea surface small target detection method based on chaos theory Download PDFInfo
- Publication number
- CN103091668A CN103091668A CN2013100109457A CN201310010945A CN103091668A CN 103091668 A CN103091668 A CN 103091668A CN 2013100109457 A CN2013100109457 A CN 2013100109457A CN 201310010945 A CN201310010945 A CN 201310010945A CN 103091668 A CN103091668 A CN 103091668A
- Authority
- CN
- China
- Prior art keywords
- chaos
- neural network
- extra large
- detection method
- sea
- 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
Images
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention aims at providing a sea surface small target detection method based on a chaos theory. The sea surface small target detection method includes a first step of carrying out amplitude and phase correction, filtering and normalized processing to original sea clutter data, a second step of carrying out extraction of correlation dimension and the biggest Lyapunov index chaos characteristic quantity to carry out the chaos characteristic verification to sea clutter, a third step of using a traditional statistics detecting method if the sea clutter is not provided with the chaos characteristic, and using a neural network to carry out modeling to chaos time sequence if the sea clutter is provided with the chaos characteristic, a fourth step of training the neural network through a chaos background signal generation mode, carrying out a single-step detection to the received signals, obtaining a predicted value and calculating a predicted error, if the predicted error is larger than a threshold value, a target exists, and on the contrary, the target does not exist. The sea surface small target detection method does not need priori knowledge, is not sensitive to an initial parameter, and can not sink into a local minimum point. A recursive least square method can be adopted to confirm the weight between a hidden layer and an output layer, and fast convergence speed can be guaranteed.
Description
Technical field
What the present invention relates to is a kind of object detection method of radar signal processing field.
Background technology
Traditional sea small target detecting method is that extra large clutter is carried out analysis modeling as stochastic process, along with the more deep research of people to extra large clutter, finds that extra large clutter is not only stochastic process, but has chaotic characteristic.With respect to traditional detection method based on statistical model, more can reflect the dynamics of extra large clutter based on the sea small target detecting method of sea-clutter chaos property.Neural network by genetic algorithm optimization makes the detection performance that further raising arranged.
The report relevant to the application has: 1, " based on the extra large clutter processing scheme design of chaology " (" science and technology and engineering " the 10th the 6th phase of volume in 2010), proposed the design proposal that under a kind of extra large clutter background, the radar return data are processed.It is carried out pre-service not changing on raw data dynamic perfromance basis, then the chaos of check data, utilize at last radial basis function neural network as fallout predictor, target is had or not judge.2, the neural network prediction of the chaos time sequence " research " (" naval aviation engineering college journal " the 23rd the 1st phase of volume in 2008), studied the forecasting problem of chaos system.By reconfiguration system state phase space analysis chaos time sequence, then with multilayer feedforward neural network, it is predicted.3, " based on the neural network performance optimization of genetic algorithm " (" computer technology and development " the 17th the 12nd phase of volume in 2007), analyze the characteristics of genetic algorithm and the characteristics of neural network in literary composition, thereby drawn, two kinds of algorithms have been combined the thought of using.Use the method for theoretical contrast, illustrated the reason of carrying out the neural network performance optimization with genetic algorithm, and reached a conclusion, carry out the neural network performance optimization with genetic algorithm and impelled neural network further to use.
Although existing extra large clutter time series is predicted with neural network much about the researchs of sea small target detecting method, and is used detection method that genetic algorithm is optimized neural network also seldom, can further improve the detection performance.
Summary of the invention
A kind of sea small target detecting method based on chaology that the object of the present invention is to provide under extra large clutter background, neural network is optimized as the basis, take neural network as instrument and by genetic algorithm take chaology.
The object of the present invention is achieved like this:
A kind of sea small target detecting method based on chaology of the present invention is characterized in that:
(1) original extra large clutter data are carried out amplitude and phase correction, filtering and carried out normalized;
(2) carry out the extraction of correlation dimension and maximum Lyapunov exponent chaos characteristic amount, extra large clutter is carried out the chaotic characteristic checking;
(3) if extra large clutter does not have chaotic characteristic, use traditional statistical detection method;
If extra large clutter has chaotic characteristic, with neural network to the chaos time sequence modeling, with neural network as a fallout predictor, namely produce pattern to neural metwork training with the Chaotic Background signal, make its predicated error reach 0.001, thereby realize that time series is had short-term forecasting performance preferably; Otherwise change topology of networks and re-start training, until meet the demands, after neural metwork training is good, carry out to the received signal Single-step Prediction, obtain predicted value and calculate predicated error, the analyses and prediction error: if predicated error greater than threshold value, target exists, on the contrary the explanation driftlessness.
Advantage of the present invention is: the present invention compares with classic method, need not priori, and insensitive to initial parameter, so can not be absorbed in local minimum point.Least square method of recursion can be adopted for the definite of the weights between hidden layer and output layer, speed of convergence faster can be guaranteed.
Description of drawings
Fig. 1 is process flow diagram of the present invention;
Fig. 2 is amplitude and phase-correcting circuit figure;
Fig. 3 is the detection method schematic diagram based on neural network;
Fig. 4 is RBF neural network structure figure;
Fig. 5 is the genetic algorithm process flow diagram.
Embodiment
For example the present invention is described in more detail below in conjunction with accompanying drawing:
In conjunction with Fig. 1～5, the present invention proposes solution mainly for modeling and the Testing of Feeble Signals problem thereof of extra large clutter, take chaology as the basis, take neural network as instrument, from Phase-space Reconstruction, utilize the RBF neural network to come the internal motivation of the extra large clutter of reconstruct, the detection method of introducing based on predicated error detects analysis to small-signal.This new detection technique has the application potential that is better than the traditional detection technology, has this prior imformation of chaotic behavior because it takes full advantage of extra large clutter, is predictable and chaos system is a kind of deterministic system, has at least at short notice predictability.The specific implementation process can be divided into following four steps:
1, the pre-service of extra large clutter data
The pre-service of measured data is a basic step in all analysis of experimental data work.The extra large clutter data of collecting with different radars in the different location exist the impact of error, and we can be by carrying out on data the impact that pre-service reduces error before extracting chaotic parameter.Preprocessing process comprises the following steps:
(1) amplitude and phase correction (I-Q calibration)
(2) filtering
(3) normalized of extra large clutter data
2, the checking of sea-clutter chaos property
(1) to extra large clutter seasonal effect in time series phase space reconfiguration
In processing based on the Nonlinear Time Series of chaology, be that the prediction that the foundation of calculating, the nonlinear model of chaos invariant also is based on chaotic model is all carried out in phase space, so phase space reconfiguration is the very important first step in the Nonlinear Time series processing.The Takens embedding theorems are bases of phase space reconfiguration, when utilizing time series to carry out phase space reconfiguration, pay special attention to embed choosing of dimension m and time-delay τ, because choosing of m and τ will directly affect " coupling " of reconfiguration system and original system degree, affect the reconstruction quality of Chaotic Sea Clutter, and then the effect of impact prediction and detection.The general mutual information method computation delay τ that adopts, G-P method and Cao method are calculated and are embedded dimension.
(2) extra large clutter time series is carried out chaotic Property Analysis
The key property index of describing chaos system comprises maximum Lyapunov exponent, correlation dimension etc.By analyzing the chaotic characteristic of the extra large noise signal of Lyapunov exponential sum correlation dimension checking.
3, use the predictive ability of neural network to carry out little target detection
Chaos time sequence has long-term unpredictable and predictable characteristics of short-term.Like this, can utilize neural network first to the chaos time sequence modeling during echo signal under detecting Chaotic Background, then predict to the received signal.
Object detection method based on neural network under Chaotic Background can simply be summed up as following several step:
(1) with neural network to the chaos time sequence modeling, neural network as a fallout predictor, is namely produced pattern to neural metwork training with the Chaotic Background signal, make its predicated error less, and time series had short-term forecasting performance preferably.Otherwise, change topology of networks and re-start training, until meet the demands.
(2) neural metwork training good after, (original data in training process) carry out Single-step Prediction (n is the total length of the data point that obtains in the observation stage) to the received signal, obtain predicted value and calculate predicated error.
(3) according to the CFAR processor, predicated error is analyzed, if predicated error greater than the threshold value that presets, illustrates the existence of target, on the contrary the explanation driftlessness.
4, use genetic algorithm that neural network is optimized, improve and detect performance
Radial basis function neural network (RBFNN) has simple network structure, learning method, Generalization Ability preferably fast, is widely used in the approximation of function field.Yet, how effectively to determine network structure and the parameter of RBF neural network, do not have so far systematic rule to follow.The parameter that needs to determine in the RBF neural network comprises that the central value of hidden layer node number, hidden layer basis function is connected the connection weights of hidden layer to output layer with width.Here utilize genetic algorithm optimization to choose the central value of hidden layer node, improve the neural network performance.Compare with traditional algorithm, the genetic algorithm neural network training need not priori, and insensitive to initial parameter, so can not be absorbed in local minimum point.Least square method of recursion can be adopted for the definite of the weights between hidden layer and output layer, speed of convergence faster can be guaranteed.
The principal feature of method of the present invention is as follows:
1, the checking of sea-clutter chaos property.Theoretical according to Chaos, can judge from following 5 aspects whether a process is chaos system.(1) process is limited; (2) process is nonlinear; (3) along with the increase correlation dimension that embeds dimension is limited; (4) to the susceptibility of starting condition, namely has a positive Lyapunov index at least; (5) though Lyapunov index sum is arranged for negative.If above-mentioned any one is false, illustrate that this process is not chaos system.Due to the back scattering clawback that extra large clutter is radar illumination behind the sea, be therefore bounded, it is a dissipative system simultaneously, is physically realizable, therefore set up (1) (5); If (4) set up, show two tracks of phase space with exponent separation, set up (2).To sum up, as long as calculate correlation dimension and Lyapunov index, just can judge that whether extra large clutter system is chaos.
2, based on the little target detection in the sea of RBF neural network.At first, extra large clutter time series is carried out phase space reconfiguration.In processing based on the Nonlinear Time Series of chaology, be that the prediction that the foundation of calculating, the nonlinear model of chaos invariant also is based on chaotic model is all carried out in phase space, so phase space reconfiguration is the very important first step in the Nonlinear Time series processing.The Takens embedding theorems are bases of phase space reconfiguration, when utilizing time series to carry out phase space reconfiguration, pay special attention to embed choosing of dimension m and time-delay τ, because choosing of m and τ will directly affect " coupling " of reconfiguration system and original system degree, affect the reconstruction quality of Chaotic Sea Clutter, and then the effect of impact prediction and detection.The general mutual information method computation delay τ that adopts, G-P method and Cao method are calculated and are embedded dimension.Chaos time sequence has long-term unpredictable and predictable characteristics of short-term.Like this, can utilize neural network first to the chaos time sequence modeling during echo signal under detecting Chaotic Background, then predict to the received signal.
Object detection method based on neural network under Chaotic Background can simply be summed up as following several step:
(1) with neural network to the chaos time sequence modeling, with neural network as a fallout predictor, namely produce pattern to neural metwork training with the Chaotic Background signal, make its predicated error reach 0.001, thereby realize that time series is had short-term forecasting performance preferably.Otherwise, change topology of networks and re-start training, until meet the demands.
(2) neural metwork training good after, (original data in training process) carry out Single-step Prediction (n is the total length of the data point that obtains in the observation stage) to the received signal, obtain predicted value and calculate predicated error.
(3) according to the CFAR processor, predicated error is analyzed, if predicated error greater than threshold value, illustrates the existence of target, on the contrary the explanation driftlessness.
3, utilize genetic algorithm optimization RBF neural network.
Radial basis function neural network (RBFNN) has simple network structure, learning method, Generalization Ability preferably fast, is widely used in the approximation of function field.Yet, how effectively to determine network structure and the parameter of RBF neural network, do not have so far systematic rule to follow.The parameter that needs to determine in the RBF neural network comprises that the central value of hidden layer node number, hidden layer basis function is connected the connection weights of hidden layer to output layer with width.Here utilize genetic algorithm optimization to choose central value and the width of hidden layer node, improve the neural network performance.Compare with traditional algorithm, the genetic algorithm neural network training need not priori, and insensitive to initial parameter, so can not be absorbed in local minimum point.Least square method of recursion can be adopted for the definite of the weights between hidden layer and output layer, speed of convergence faster can be guaranteed.
For example the present invention is done more detailed description below in conjunction with accompanying drawing:
In conjunction with shown in Figure 1, the pre-service of original extra large clutter data is basic steps in the little target detection in whole sea.Before data that radar is received are processed, must first carry out suitable pre-service, the impact of error is reduced.Preconditioning technique is used for radar data in the following order: (1) amplitude and phase correction; (2) filtering; (3) normalized of data.After raw data is carried out pre-service, carry out the extraction of the chaos characteristic amounts such as correlation dimension and maximum Lyapunov exponent, thereby extra large clutter is carried out the chaotic characteristic checking.If extra large clutter has chaotic characteristic, can use the object detection method based on chaology.Otherwise, use traditional statistical detection method.After having verified the chaotic characteristic of extra large clutter, just can utilize the RBF neural network that extra large clutter time series is predicted, the existence that judges the little target in sea according to the predicated error between predicted value and actual value whether.At last, to detect performance in order further improving, to utilize genetic algorithm to carry out parameter optimization to the RBF neural network, thus the realize target judgement.
In conjunction with shown in Figure 2, amplitude and phase correction (I-Q calibration) circuit.Due to the impact of experimental error, amplitude and phase mismatch (variance does not wait and the simple crosscorrelation non-zero) appear in the original homophase (I) of coherent radar and quadrature (Q) component.This is by direct current (DC) biasing and gain and the phase difference of receiving cable cause separately.The I-Q calibration process is used for measuring the imbalance of I and Q two interchannel gains and phase place.Low-pass filter in figure (LPF) filtering appears at the high fdrequency component of multiplier output terminal.By take the I passage as reference, the receiving cable of I and Q two passages gains by normalization.This makes the I passage have unity gain, and the gain of Q passage is G
eθ
eOn expression Q passage with respect to the relative phase-angle error of I passage.Cos (ω
IFT+ α) be input intermediate-freuqncy signal, DC
IAnd DC
QThe direct current biasing that represents respectively I and Q passage.
In conjunction with shown in Figure 3, based on the detection method schematic diagram of neural network.At first, extra large clutter time series x (n) is carried out phase space reconfiguration, use RBF neural metwork training one batch data, obtain a neural network prediction device, then with the other one group of not used data of its prediction, obtain its predicted value
Calculate its predicated error ε (n).Through many experiments, determine its threshold value.Data to be tested are input in the RBF that trains, the error of its output and the threshold value of having set are compared, if predicated error greater than threshold value, thinks that target is arranged, otherwise explanation does not have target.
In conjunction with shown in Figure 4, RBF neural network structure figure.Radial basis function neural network is three layers of feedforward neural network that only have a hidden layer, is to comprise an input layer, a hidden layer, the simplest pattern of an output layer.Input layer has m node, and output layer has 1 node.It compares maximum different being from feedforward network, the transfer function of hidden layer is the Gaussian function of local acknowledgement.Due to local acknowledgement, radial basis function network can approach the arbitrary continuation function with arbitrary accuracy.
In conjunction with shown in Figure 5, the genetic algorithm process flow diagram.At first, the parameter of RBF neural network is encoded, generate initial population.Then according to objective function, calculate fitness, fitness value is estimated detection.By selecting, intersect, the genetic operator of variation three basic is selected and heredity, generates colony of future generation.Carry out at last the end condition judgement, if satisfied end condition, export as optimum solution with resulting individuality with maximum adaptation degree in evolutionary process, stop computing.
Claims (1)
1. sea small target detecting method based on chaology is characterized in that:
(1) original extra large clutter data are carried out amplitude and phase correction, filtering and carried out normalized;
(2) carry out the extraction of correlation dimension and maximum Lyapunov exponent chaos characteristic amount, extra large clutter is carried out the chaotic characteristic checking;
(3) if extra large clutter does not have chaotic characteristic, use traditional statistical detection method;
If extra large clutter has chaotic characteristic, with neural network to the chaos time sequence modeling, with neural network as a fallout predictor, namely produce pattern to neural metwork training with the Chaotic Background signal, make its predicated error reach 0.001, thereby realize that time series is had short-term forecasting performance preferably; Otherwise change topology of networks and re-start training, until meet the demands, after neural metwork training is good, carry out to the received signal Single-step Prediction, obtain predicted value and calculate predicated error, the analyses and prediction error: if predicated error greater than threshold value, target exists, on the contrary the explanation driftlessness.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013100109457A CN103091668A (en) | 2013-01-11 | 2013-01-11 | Sea surface small target detection method based on chaos theory |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013100109457A CN103091668A (en) | 2013-01-11 | 2013-01-11 | Sea surface small target detection method based on chaos theory |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103091668A true CN103091668A (en) | 2013-05-08 |
Family
ID=48204471
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013100109457A Pending CN103091668A (en) | 2013-01-11 | 2013-01-11 | Sea surface small target detection method based on chaos theory |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103091668A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104007423A (en) * | 2014-05-27 | 2014-08-27 | 电子科技大学 | Sky wave radar sea clutter suppression method based on chaos sequence prediction |
| CN105738888A (en) * | 2016-03-31 | 2016-07-06 | 西安电子科技大学 | Double-characteristic sea surface floating small-target detection method based on sea clutter suppression |
| CN108387880A (en) * | 2018-01-17 | 2018-08-10 | 西安大衡天成信息科技有限公司 | Detection method of small target under a kind of sea clutter background based on multiple dimensioned oriented Lyapunov indexes |
| CN112068085A (en) * | 2020-10-16 | 2020-12-11 | 中国电波传播研究所（中国电子科技集团公司第二十二研究所） | Radar sea clutter original data rapid preprocessing method based on deep learning |
| CN112668507A (en) * | 2020-12-31 | 2021-04-16 | 南京信息工程大学 | Sea clutter prediction method and system based on hybrid neural network and attention mechanism |
-
2013
- 2013-01-11 CN CN2013100109457A patent/CN103091668A/en active Pending
Non-Patent Citations (4)
| Title |
|---|
| 史忠植: "《神经网络》", 31 May 2009, 高等教育出版社 * |
| 徐伟 等: "基于海杂波混沌特性的微弱目标信号检测", 《雷达与对抗》 * |
| 钱玉莹 等: "基于混沌理论的海杂波处理方案设计", 《科学技术与工程》 * |
| 韩长喜: "海杂波背景下基于混沌理论的目标检测", 《微型机与应用》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104007423A (en) * | 2014-05-27 | 2014-08-27 | 电子科技大学 | Sky wave radar sea clutter suppression method based on chaos sequence prediction |
| CN104007423B (en) * | 2014-05-27 | 2016-09-14 | 电子科技大学 | Folded Clutter in Skywave Radars ocean clutter cancellation method based on chaos sequence prediction |
| CN105738888A (en) * | 2016-03-31 | 2016-07-06 | 西安电子科技大学 | Double-characteristic sea surface floating small-target detection method based on sea clutter suppression |
| CN108387880A (en) * | 2018-01-17 | 2018-08-10 | 西安大衡天成信息科技有限公司 | Detection method of small target under a kind of sea clutter background based on multiple dimensioned oriented Lyapunov indexes |
| CN108387880B (en) * | 2018-01-17 | 2020-06-09 | 西安大衡天成信息科技有限公司 | Multi-scale directed Lyapunov index-based method for detecting weak and small targets in sea clutter background |
| CN112068085A (en) * | 2020-10-16 | 2020-12-11 | 中国电波传播研究所（中国电子科技集团公司第二十二研究所） | Radar sea clutter original data rapid preprocessing method based on deep learning |
| CN112068085B (en) * | 2020-10-16 | 2022-05-06 | 中国电波传播研究所（中国电子科技集团公司第二十二研究所） | Radar sea clutter original data rapid preprocessing method based on deep learning |
| CN112668507A (en) * | 2020-12-31 | 2021-04-16 | 南京信息工程大学 | Sea clutter prediction method and system based on hybrid neural network and attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Detection of low observable moving target in sea clutter via fractal characteristics in fractional Fourier transform domain | |
| CN103091668A (en) | Sea surface small target detection method based on chaos theory | |
| Shiri et al. | Estimation of daily suspended sediment load by using wavelet conjunction models | |
| Can et al. | Daily streamflow modelling using autoregressive moving average and artificial neural networks models: case study of Ç oruh basin, T urkey | |
| CN102879766A (en) | Method and apparatus for detecting and tracking faint target of high frequency ground wave radar | |
| Darbandi et al. | Real‐time stability assessment in smart cyber‐physical grids: a deep learning approach | |
| Meng et al. | A threshold artificial neural network model for improving runoff prediction in a karst watershed | |
| Liu et al. | Rolling bearing fault diagnosis based on EEMD sample entropy and PNN | |
| Ghanbari et al. | Anomaly detection in a smart grid using wavelet transform, variance fractal dimension and an artificial neural network | |
| Rezaeian Koochi et al. | New phasor‐based approach for online and fast prediction of generators grouping using decision tree | |
| Chai et al. | Non-linear system identification and fault detection method using RBF neural networks with set membership estimation | |
| Xin et al. | Intelligent fault diagnosis method for rotating machinery based on vibration signal analysis and hybrid multi‐object deep CNN | |
| Dong et al. | Multi‐task learning method for classification of multiple power quality disturbances | |
| Afrasiabi et al. | Two-stage deep learning-based wind turbine condition monitoring using SCADA data | |
| Huang et al. | False phasor data detection under time synchronization attacks: A neural network approach | |
| Arvani et al. | Detection and protection against intrusions on smart grid systems | |
| Saleki et al. | High‐speed directional protection based on cross correlation of Fourier transform components of voltage and current | |
| Pan et al. | Multi-sensor signal fusion method for rolling bearing based on the standard relative mean–variance value and random weighting algorithm | |
| Peng et al. | Short‐term wind power prediction based on stacked denoised auto‐encoder deep learning and multi‐level transfer learning | |
| Zhang et al. | The method for determining optimal analysis length of vibration data based on improved multiscale permutation entropy | |
| Fan et al. | Composite Fault Diagnosis of Rolling Bearing Based on Optimized Wavelet Packet AR Spectrum Energy Entropy Combined with Adaptive No Velocity Term PSO‐SOM‐BPNN | |
| CN109525453B (en) | Networked CPS (control performance Standard) anomaly detection method and system based on node dependency relationship | |
| Mou et al. | Motion classification for radar moving target via STFT and convolution neural network | |
| CN109506936B (en) | Bearing fault degree identification method based on flow chart and non-naive Bayes inference | |
| Bao et al. | Nautical radar clutter suppression and small target detection based on image spectrum filtering and Hough transform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C05 | Deemed withdrawal (patent law before 1993) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130508 |