基于孪生卷积网络的SAR图像变化检测

SAR Image Change Detection Based on Siamese Convolutional Network

SAR图像检测容易受到相干斑噪声的干扰,传统的基于差异图的变化检测算法难以获取变化区域鲁棒的差异信息,进而影响了变化检测的精度。为了解决该问题,文中提出一种基于孪生卷积网络的多时相SAR图像变化检测算法。该算法由两支路卷积网络构成,每一支路卷积网络用于提取某一时相SAR图像的特征。然后将两时相的SAR图像特征输入由测度函数构成的差异信息提取层,从而获得两时相SAR图像的差异信息。我们用两组黄河地区的两时相SAR图像来验证所提出的算法,实验结果表明,相对于传统差异图的变化检测算法,文中所提出的算法能获得更好的效果。

Due to the speckle in SAR images,the change detection methods based on traditional difference image fail to obtain robust the difference information of changed region and will reduce its performance of change detection.To solve this issue,in this paper,we will propose a multi-temporal SAR image change detection based on a Siamese convolutional network.The proposed Siamese network consists of two branches of convolutional network,each of which is employed to extract the feature for a temporal SAR image.Then the features extracted from the bitemporal SAR images will be compared and we will get the difference information via a layer constructed from a metric function.Finally,we verify our proposed method on two sets of bitemporal SAR images on Yellow River areas and have it compared with the method based on traditional difference image.The experimental results show that our proposed method outperforms those methods based on the traditional difference image.