基于多尺度学习型字典表示的极地浅层探冰雷达图像去噪算法

为了更好地解决极地浅层探冰雷达回波信号中的杂波和噪声问题,提出了一种基于多尺度学习型字典表示的极地浅层探冰雷达图像去噪算法。该算法首先通过曲波变换构建曲波系数矩阵,在曲波域使用自适应字典学习得到原子尺寸不同的多尺度字典,最后利用去噪和修正后的曲波系数重建浅层探冰雷达剖面图像,完成最终的去噪。结果表明:相较于曲波变换去噪算法、K-SVD(K-奇异值分解)去噪算法,改进的算法不但能够有效地去除噪声,提高图像的峰值信噪比,而且探冰雷达图像的边缘轮廓信息得到了较好的保留,有着良好的视觉效果。

Image Denoising Algorithm of Shallow Ice-penetrating Radar Based on Multi-scale Learning-type Dictionary

In order to solve the problem of clutter and noise in the echo signal of polar shallow ice-penetrating radar better, an image denoising algorithm based on multi-scale learning-type dictionary for polar shallow ice-penetrating radar is proposed. Firstly, the curvelet coefficient matrix is constructed by the curvelet transform, and then the multi-scale dictionary with atoms of different sizes is acquired by using the adaptive dictionary in the curvelet domain. Finally, the denoised and modified curvelet coefficients are used to reconstruct the shallow ice-penetrating radar profile image, and the final denoising is completed. The results show that compared with the curvelet transform denoising algorithm and the K-SVD (K-singular value decomposition) denoising algorithm, the proposed algorithm can not only remove noise and improve the peak signal-to-noise-ratio of the image effectively, but also preserve the edge contour information of the ice-penetrating radar image and create a good visual effect.