Research On The Application Of Transfer Entropy And Fractals In Sea Clutter Target Detection

Radar detects and locates targets by transmitting electromagnetic signals,using the phenomenon of reflection or scattering of electromagnetic waves by the target.The process of radar reception is often accompanied by clutter,and sea clutter is one of the most complex forms of radar clutter,which strongly interferes with radar for target detection.Therefore,the study of the characteristics of sea clutter is of great importance for the design of radar systems and for effective sea surface target detection.This thesis focuses on how to extract the features of the sea clutter and the target,and to identify the target effectively based on the difference between the two features.This thesis investigates and analyses target detection methods in the context of sea clutter based on IPIX data sets of measured radar signals on the sea.The main work and innovations of this thesis are as follows:1.On the basis of the IPIX data set of real sea radar measurements,the amplitude statistical characteristics of sea clutter and the chaotic characteristics are analyzed;based on the phase space reconstruction theory,and using a reasonable method to select the phase space reconstruction parameters,the one-dimensional signal is extended to high dimensions,and the intrinsic dynamics of the sea surface echo signal is restored,laying the foundation for the subsequent algorithm.2.Based on the phase space reconstruction theory,the Weighted Permutation Entropy(WPE)algorithm is introduced to analyze the chaotic nature of the sea surface echo signal.The complexity and randomness of the target and sea clutter are also characterized by the WPE algorithm,which verifies the feasibility of using the WPE values of the sea clutter and the target as features for target detection,and that the HH has better differentiation than the WPE features under the VV polarization approach.Wavelet analysis is introduced on the basis of WPE,and a target detection method based on wavelet decomposition and WPE is proposed.The results show that the WPE values of the approximation and detail components of the wavelet decomposition are used to calculate the feature vector,and the ROC curve is used to verify the effectiveness of the wavelet decomposition and WPE based target detection.The results show that the area under the ROC curve for the eigenvectors consisting of the WPE values of the approximate component reaches 100% in low sea state conditions,99.9%in mid sea state conditions and 99.2% in high sea state conditions;the area under the ROC curve for the eigenvectors consisting of the WPE values of the detail component is 97.9% in low sea state conditions,76.5% in mid sea state conditions and 75.4% in high sea state conditions.The area under the ROC curve was 97.9% in low sea state,76.5% in medium sea state and 75.4% in high sea state.The results show that the eigenvectors are better differentiated between sea clutter and targets.3.Based on mutual information,the concept of Transfer Entropy(TE)is introduced to analyze the transfer relationships between the time series of sea surface echoes.The effectiveness of the TE algorithm is verified by means of a dynamical linear system AR model,and the feasibility of using the transmission relationship between sequences(TE)as a feature of sea clutter and signals for target detection is demonstrated based on the echo signals of different sea conditions in the IPIX sea radar real-world dataset.The Symbolic Phase Transfer Entropy(SPTE)algorithm is proposed on the basis of time series symbolization theory and more signal variation characteristics in phase information,and the phase information of the sea surface echo signal is symbolized and its TE value is calculated with the help of phase space reconstruction theory,and the effect on sea clutter and target is analyzed in different embedding dimensions.The effect of the embedding dimension on the distinction between sea clutter and target is analyzed.Finally,three classifiers,ROC,Support Vector Machine(SVM)and Extreme Learning Machine(ELM),were used to compare the detection performance of TE and SPTE algorithms,and it was found that SPTE was more effective in discriminating between sea clutter and targets under different sea conditions and different polarization methods The TE algorithm was used to extract the echo signal features: the average recognition performance was 62.1% for mid-sea conditions and 55.6% for high-sea conditions in VV polarization;85.5% for mid-sea conditions and 87.0% for high-sea conditions in HH polarization.The SPTE algorithm was used to extract the echo signal features: the average recognition performance was 79.9% for medium sea state and 84.5% for high sea state in VV polarization method;the average recognition performance was 93.7%for medium sea state and 97.1% for high sea state in HH polarization method.The results show that SPTE is more effective in distinguishing between sea clutter and targets under different sea conditions and different polarization methods.4.The fractal theory is introduced and the applicability of the two single fractal dimensions in sea surface target detection is verified based on the verification of the single fractal characteristics of sea clutter waves by two methods: box dimension and Hurst index.Then the Multifractal Detrended Fluctuation Analysis(MFDFA)is introduced on the basis of fractal theory,and the MFDFA method is used to verify the multifractal properties of sea clutter and analyze the parameters of sea clutter multifractals.Finally,the performance of target detection using the singularity index of the multiple fractal spectrum of the target and sea clutter as a feature is analyzed by three detection methods,namely ROC,SVM and ELM.The results show an average detection performance of 52.3% in medium sea conditions and54.3% in high sea conditions for the VV polarization method and 85.7% in medium sea conditions and 82.4% in high sea conditions for the HH polarization method.The results demonstrate that the algorithm is effective in detecting the surface echo signals with HH polarization under different sea state conditions.