基于小波变换的改进DnCNN网络的遥感图像去噪

Remote sensing image denoising based on wavelet transform and DnCNN

摘要:

遥感图像去噪一直是遥感领域的重要难题，现有的去噪算法会使图像边缘信息模糊，导致图像中有用信息丢失，为了提高遥感图像的质量，提出了一种改进DnCNN（Denoising Convolutional Neural Network）的遥感图像去噪方法，通过小波变换将原始图像分解成不同子带，采用基于遗传算法的网络结构自动搜索方法对于不同子带搜索出不同结构和参数的DnCNN网络实现去噪，使对噪声成分的提取更加有针对性。实验采用峰值信噪比（PSNR）和结构相似性（SSIM）两项评价指标对实验结果进行量化评判，标准差为20时，较原始的DnCNN方法相比PSNR值平均提高了3.5%，图像细节清晰，能有效地保护遥感图像边缘特征和轮廓结构的完整性。

Abstract:

Remote sensing image denoising is an important problem in the field of remote sensing research. Considering the problem that the existing denoising algorithms lead to the loss of useful edge information of the image, a remote sensing image denoising method based on improved DnCNN (Denoising Convolutional Neural Network) is proposed. The original image is transformed into different subbands based on wavelet transform, and the network structure automatic search method based on genetic algorithm is used to Denoise the DnCNN network with different subbands with different structures and parameters, which makes the extraction of noise components more targeted. Peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) are used to quantitatively evaluate the experimental results. When the standard deviation is 20, compared with the original DnCNN method, the PSNR value is increased by 3.5%, and the image details are clear. The?experimental?results?show?that?the proposed?method can effectively protect the integrity of edge features and contour structure of remote sensing images.