合成孔径雷达能够全天时,全天侯产生高分辨率SAR图像.SAR图像中由于工作环境及成像机制会受到噪声影响,大多数去噪算法去除SAR图像噪声时会出现噪声去除不完全,图像信息损失的问题.针对这一问题,提出了一种基于U-Net网络结构改进的SAR图像去噪算法.该算法采用VGG16 网络结构作为特征提取模块,对SAR图像进行去噪的下采样操作,提取SAR图像中的关键特征,保留去噪后SAR图像的细节信息;采用修改的U-Net上采样网络结构,让包含特征的低分辨率图片在保留特征的同时变为高分辨率,并通过特征融合使得去噪后SAR图像恢复更多细节,实现SAR图像的智能去噪.选择峰值信噪比(Peak Signal to Noise Ratio,PSNR)和结构相似性指数(Structural Similarity Index Measure,SSIM)作为实验的评价指标.仿真实验结果表明,该方法对添加噪声的SAR图像进行去噪,其主观视觉效果及客观评价指标PSNR和SSIM相比于实验对照去噪方法较高.所提方法兼顾了SAR图像噪点的去除和细节的保留,去噪获取的SAR图像具备更清晰的细节特征,具有较强的SAR图像去噪现实意义.
Research on SAR image denoising technology based on U-Net network structure
Synthetic Aperture Radar(SAR)is capable of producing SAR images around the clock.SAR images are affected by noise due to the working environment and imaging mechanism,and most of the denoising algorithms for removing the noise from SAR images will have the problem of incomplete noise removal and loss of image information.To address this issue,an enhanced SAR image denoising algorithm based on the improvement of U-Net network structure is proposed.The algorithm adopts the VGG16 network structure as the feature extraction module,and performs the downsampling operation for denoising SAR images.The key features in the SAR image are extracted,and the detail information is retained;the modified U-Net upsampling network structure is used to make the low-resolution image containing features become high-resolution while retaining the features.And through feature fusion,more details are recovered in the denoised SAR image to realize denoising of SAR image.Peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)are chosen as evaluation metrics.The results of experiments show that the method denoises SAR images with added noise.Its subjective visual effect and objective evaluation indexes are higher compared to the experimental control denoising method.The method proposed in this paper takes into account the removal of SAR image noise and the preservation of details,and the details of the denoised SAR image are more clear and real,with better visual effect,which holds significant practical significance for SAR image denoising.
SAR imagesconvolutional neural networkU-NetVGG16image denoising
NETL
NSTL
万方数据
