基于多尺度交互结构卷积神经网络的SAR图像相干斑抑制方法  

SAR image speckle suppression method based on muti-scale interactive structure convolutional neural network

作　　者：申仕煜 叶晓东[1] 王昊[1] 陶诗飞 SHEN Shiyu;YE Xiaodong;WANG Hao;TAO Shifei(School of Electronic and Optical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学电子工程与光电技术学院,江苏南京210094

出　　处：《系统工程与电子技术》2021年第12期3526-3532,共7页Systems Engineering and Electronics

基　　金：国家自然科学基金(61701240);中央高校基本科研业务费专项资金(30918011317)资助课题。

摘　　要：结合深度学习思想,提出了一种基于多尺度交互结构卷积神经网络(convolutional neural network,CNN)的合成孔径雷达(synthetic aperture radar,SAR)图像相干斑抑制方法。首先,通过不同尺寸的卷积核及跳跃连接构成多尺度交互特征提取模块以获得不同感受野的特征并加快网络收敛速度。然后,在多尺度交互特征提取模块之间利用简化的密集连接方式使网络能够充分利用浅层纹理特征。最后,采用残差学习策略得到抑制后的图像。实验结果表明,与已有方法相比,所提方法不仅使用较少的计算参数量,还能保证性能的提升。Combined with the idea of deep learning,a speckle suppression method for synthetic aperture radar(SAR)images based on multi-scale interactive convolutional neural network(CNN)is proposed.Firstly,a multi-scale extraction module is constructed by convolution kernels and skip connection with different sizes to obtain the features of different receptive fields and speed up the convergence of the network.Then,the network can make the best of shallow texture features by using simplified dense connection between multi-scale interactive feature extraction modules.Finally,the suppressed image is obtained by residual learning strategy.The experimental results show that compared with the existing methods,the proposed method not only uses less calculation parameters,but also ensures the improvement of performance.

关 键 词：深度学习 合成孔径雷达 相干斑抑制 卷积神经网络 残差学习策略 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]