基于卷积自编码网络的SAR图像相干斑抑制算法  

A SAR Image Despeckling Method Based on Convolutional Auto-encoder Network

作　　者：赵婧媛[1] 张志伟[1] ZHAO Jing-yuan;ZHANG Zhi-wei(Beijing Aeronautical Technology Research Center,Beijing 100000,China)

机构地区：[1]北京航空工程技术研究中心,北京100000

出　　处：《航空计算技术》2021年第4期76-80,共5页Aeronautical Computing Technique

摘　　要：针对SAR图像中存在的相干斑噪声影响目标识别性能的问题,设计了基于卷积自编码网络的相干斑噪声抑制模型,利用多层卷积自动编码器挖掘SAR图像潜在的空间特征,并使用混合正则约束优化网络训练过程提高网络泛化性能。通过仿真数据和真实数据两部分实验,用峰值信噪比、结构相似性以及等效视数作为算法性能评价指标,并与SAR-BM3D、PPB、ID-CNN进行对比,结果表明,网络能够有效地抑制SAR图像中的相干斑噪声,且对不同水平的噪声有较强的鲁棒性。A SAR image despeckling method based on convolutional auto-encoder network:aiming at the problem that speckle noise in SAR image affects the performance of target recognition,a despeckling model based on convolutional auto-encoder network is designed.The potential spatial characteristics of SAR image are mined by multi-layer convolutional auto-encoder,and the training process of the network is optimized by using mixed regularization constraints to improve network generalization performance.Through simulation data and real data two part experiments,peak signal to noise ratio,structural similarity,and equivalent number of looks are used as performance evaluation indexes,and compared with SAR-BM3D,PPB,and ID-CNN.The results show that the network can effectively suppress speckle noise in SAR image,and has strong robustness to different levels of noise.

关 键 词：SAR图像 相干斑噪声 噪声抑制 卷积自编码 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]