基于频域信息融合和稀疏贝叶斯学习的高分辨ISAR成像

为了在ISAR成像中更好地压制噪声，消除条纹干扰，提高成像分辨率，本文提出一种基于双向插值处理和频域信息融合的稀疏贝叶斯学习算法，称之为BI?FF SBL算法。该方法首先对回波信号分别进行径向和横向插值预处理，将预处理得到的两份数据通过LA?VB算法进行ISAR成像；然后将得到的两幅ISAR图像通过二维傅里叶变换进入频域，并将两个二维频谱进行信息融合处理，以消除噪声和条纹干扰的相关信息并保留目标结构信息；最后对融合处理后的频谱进行二维傅里叶逆变换，得到最终的ISAR图像。为了验证BI?FF SBL算法的ISAR成像效果，本文进行了基于仿真数据和实测数据的成像实验，并将实验结果与R?D算法、L1?BP算法、LA?VB算法进行对比，发现BI?FF SBL算法在压制噪声和去除条纹干扰方面具有明显的优势，且能提供分辨率更高的ISAR图像。当实验数据信噪比降到0 dB时，BI?FF SBL算法依然能够提供清晰的ISAR图像，明显优于其他三种算法。测试超分辨重构误差的实验结果表明，相比于L1?BP算法和LA?VB算法，BI?FF SBL算法的重构误差更低，在实验数据信噪比为0 dB时，重构信噪比可以达到13.55 dB。

High Resolution ISAR Imaging Based on Frequency Domain Information Fusion and Sparse Bayesian Learning

Aiming at suppressing noise, eliminating stripe interference and improving the resolution of ISAR image, a sparse Bayesian learning algorithm based on bidirectional interpolation and information fusion in frequency domain is proposed in this paper, named BI?FF SBL algorithm here. Firstly, radial and transverse interpolations are respectively performed on the echo signal. Secondly, the preprocessed data are put in parameter reconstruction using the LA?VB algorithm to obtain two ISAR images. Thirdly, the two ISAR images are transferred into the frequency domain by two?dimensional Fourier transform. Subsequently, through two?dimensional spectra fusion, the information related to noise and streak interference is eliminated and the structure information of the target is retained. Finally, the fused spectrum is inversely transformed by two?dimensional Fourier transform to obtain the final ISAR image. To verify the effectiveness of the BI?FF SBL algorithm, imaging experiments based on both simulated and measured data are performed in this paper. Compared with the R?D, L1?BP, and LA?VB algorithms, the experimental results indicate that the BI?FF SBL algorithm has obvious advantages in suppressing noise, removing stripe interference, and providing higher resolution in ISAR image. When the SNR of the experimental data drops to 0 dB, the BI?FF SBL algorithm can still provide clarified ISAR images, which is significantly better than the other three algorithms. The experimental results based on the super?resolution reconstruction error show that the reconstruction error of the BI?FF SBL algorithm is lower than that of the L1?BP and LA?VB algorithms, and the reconstruction SNR can reach 13.55 dB when the SNR of the experimental data is 0 dB.