Synthetic Aperture Radar(SAR)can obtain high-quality ground profile images,which has unique advantages compared to other observation methods.It has broad prospects for military and civil applications such as battlefield reconnaissance,precision guidance,terrain mapping and environmental monitoring etc.However,the reuse of electromagnetic spectrum makes the environment of SAR more and more complicated.The received SAR echoes are inevitably mixed into electromagnetic waves emitted by other electromagnetic equipment sharing the same frequency band.These interference signals are aliased with useful signals and lead to lines or noise which cover the ground target on the images and seriously degrade the SAR imaging quality.At the same time,in the actual imaging process,the SAR platform is unstable due to the influence of airflow,so Doppler parameter estimation is essential to perform motion compensation before imaging,whereas the interference hinders the accuracy of parameter estimation,resulting in blurred SAR images.Therefore,it is a critical topic with significant practical significance to effectively detect and mitigate the interference in the field of SAR research,and it is also a difficult problem.Based on this practical requirement,this thesis analyzes the characteristics of narrowband interference and wideband interference,and studies several mitigation schemes for radio frequency interference(RFI),all of which achieve good results.The specific research contents are as follows:1.Firstly,a mathematical model of SAR echo signals in complex electromagnetic environments is constructed,then Range-Doppler imaging algorithm and Chirp Scaling imaging algorithm are introduced.In addition,in order to filter the interference more effectively,the characteristic analysis of the original SAR echo contaminated by narrowband interference and wideband interference is provided in different representation domains.Finally,several qualitative and quantitative interference mitigation evaluation criteria are introduced so as to obtain filtering methods with superior performance,which lays a theoretical foundation for the subsequent chapters.2.The interference detection method based on eigenvalue analysis and the nonparametric narrowband RFI mitigation scheme based on matrix decomposition are studied.Since the interference is usually incoherent to the target signal,and generally has a large power,the eigenvalue decomposition of the correlation matrix of the original echo can be performed,thereby detecting whether the echo contains interference according to the distribution of the eigenvalues.The interference mitigation method based on matrix decomposition firstly performs the short-time Fourier transform(STFT)on the original SAR echo to obtain the STFT matrix,and secondly makes full use of the low-rank characteristics of the narrow-band interference STFT matrix and the sparse characteristics of the useful signal STFT matrix.The interference mitigation problem is transformed into a convex optimization problem of matrix decomposition.Then,the existing matrix decomposition method is used to separate the STFT matrix of target signal and the STFT matrix of interference from that of the original echo.The inverse short-time Fourier transform is applied to signal recovery.Finally,the effectiveness of the interference mitigation method is verified by results of by simulated and real-measured datasets.3.A kurtosis-based interference detection method is studied and a parameterized narrowband RFI mitigation approach based on Bayesian estimation is presented.The echoes of different scattering points in the scene differ in direction and distance,so that the overall useful signals obey the complex Gaussian distribution,while the narrowband interference has narrow bandwidth and high power.Therefore,the spectrum of SAR echo mixed with narrowband interference usually contains "spikes" and is no longer obey the complex Gaussian distribution.Kurtosis measures the degree of non-Gaussianness of a signal,which can be utilized to identify the existence of interference.The narrow-band interference mitigation method based on Bayesian estimation makes full use of the complex Gaussian property of the useful signal and the frequency sparseness of the narrowband RFI.Firstly,the frequency dictionary matrix of the interference signal is established.Then,the statistical probability density distribution function of the original echo and the interference signal is derived.The maximum a posteriori criterion and the gradient iteration algorithm are then used for interference parameters estimation.According to the frequency dictionary matrix,the interference can be reconstructed and extracted from SAR echo.Finally,the interference mitigation method is verified by simulation experiments and measured data.4.An interference detection scheme based on deep convolutional neural network and a novel narrowband and wideband interference mitigation method based on deep residual network are proposed.The deep convolutional neural network can automatically extract and select the texture features and spatial information in the image,and the time-frequency spectrum containing the interference is greatly different from that without interference,so it can be used to detect whether the original echo contains interference.The interference mitigation method based on deep residual network first utilizes STFT to transform the radar echo signal from time domain into time-frequency domain.Then,the time-frequency feature of the target signal is extracted by deep residual network and skip-connection structure,and the timefrequency image of the target signal is reconstructed.Finally,the inverse short-time Fourier transform is used to transform the signal from time-frequency domain into time domain,thereby obtaining signal without interference.Since the interference in the training samples includes narrowband interference,chirp modulated wideband interference and sinusoidal modulated wideband interference,the method can be applied to both narrowband and wideband interference mitigation.The effectiveness of the interference mitigation scheme is verified by simulated and real-measured data,and the superiority of the approach is verified by the performance evaluation function. |