To the synthetic aperture radar (SAR) imaging system which using Compressed Sensing (CS) technique,radio frequency interference (RFI) would undermine the priori sparse condition and cause deteriorationof image quality. In this paper, under the condition of the sparse target scene, it builds the target echo signal and RFI redundant dictionary, according to RFI sparse features on their redundant dictionary, greedy algorithm was used in conjunction with the minimum description length (MDL) criterion to estimate the sparsenessof RFI, and SAR imaging RFI suppression algorithms on the compressed domain was designed based on this basis. Simulation results show that in the larger JSR conditions the algorithmcan effectively suppress SAR radio interference and achieve better imaging results. I.INTRODUCTION Synthetic Aperture Radar (SAR) are not affected by the weather and time limitations, in military and civil fields it has been widely used, but large amount of data caused by large bandwidth brings great burden to store and transport system[1]. Compressive Sensing (CS) technology can achieve good imaging results by using a small amount of data based on the signal sparse sampling and reconstruction precision; but due to the intrinsic characteristics of CS theory, when the SAR echo signal exists strongRFI, the SAR image quality will be serious decline[2]. Therefore, it is of great significance for the SAR imaging to study how to detect and inhibition the RFI. Existing RFI suppression methods can be divided into parametric method and nonparametric method[3]. The main idea of parametric method is modeled on the RFI, considering the RFI as a model with many constant amplitude single frequency signal superimposed or random process in compliance with the AR model, using minimum mean square error (LMS) or maximum likelihood (Maximun likelihood, ML) criterion to estimate the signal model parameters, and then use the estimate subtraction structure to filter the obtained RFI estimated from the raw data[4]; however, in the case of highly intensive RFI modeling are more complex, it will produce the model parameter errors and computationally intensive problems. Non-parametric methods are usually detecting the interference and signal based on signal characteristics RFI with the use of spectrum estimation method, then suppressing the RFI on the time-frequency domain using the filtering method. It mainly includes the frequency domain filtering method, subspace filtering method and image subtraction method[5]. Frequency domain notch method suppresses the interference by way of making the position of RFI the frequency domain zero;however, the signal will be filtered out at the same position in the frequency domain while suppressinginterference,resulting in the loss of signal energy[6]. Subspace filtering method projects the interference and signal, then filtering the corresponding subspace of interference out, this method can suppress the steady RFI better, butfor distance-changingRFI it will have a greater signal loss when suppressing interference[7]. Subtraction imaging through a sit channel to get RFI without signal filtered it out in the image data directly; however, this method requires a separate sit channel[7]. In summary, although the existing method can suppress RFI, but there are some problems of energy loss, intensive computation and high complexity modeling. Within the spectrum of SAR, RFI components in the frequency domain has obvious sparse features,the number of principal 4th International Conference on Machinery, Materials and Computing Technology (ICMMCT 2016) © 2016. The authors Published by Atlantis Press 734 components (the sparseness of RFI) of RFI components can be accessed by OMP algorithm combined with minimal description length criterion (MDL). Because RFI has high emission power, one-way communication and other characteristics, so the energy of RFI components in the frequency domain is higher than SAR echo signal.Based on this, each echo received by the SAR, firstly, uses the OMP algorithm to reconstruct the interference coefficient and set it to zero, then uses the SAR echo signal dictionary to carry on the compressed sensing image. II.CS BASIC THEORY For a N dimensional signal 1 N x x × ∈R ( ) whose projection coefficient on the N N × basis matrix { } 1 2 , , , , , i N ψ ψ ψ ψ = Ψ ( i ψ is a N dimensional vector) were only contain ( K K N . The signal x can be expressed as