面向遥感图像变化检测的特征提取网络结构改进

Improvement of Feature Extraction Network Structure for Remote Sensing Image Change Detection

遥感图像变化检测技术作为一项重要的应用型技术,在地表变化监测领域有着广泛的应用。针对目前多数特征提取网络无法兼顾空间信息与语义信息的问题,以残差网络为基础,提出一种用于遥感图像变化检测的特征提取结构——CDResNet。使用该结构对不同时间的遥感图像进行特征提取,可以平衡所提取特征图中的空间信息和语义信息,还可以更准确地识别和定位图像中的变化区域。两个公共数据集上的实验验证了所提方法的鲁棒性和精确性。

Remote sensing image change detection technology,as an important applied technology,has a wide range of applications in the field of surface change monitoring.Aiming at the problem that most current feature extraction networks cannot balance spatial and semantic information,a feature extraction structure called CDResNet is proposed for remote sensing image change detection based on residual networks.Using this structure for feature extraction of remote sensing images at different times can balance the spatial information and semantic information in the extracted feature maps,and more accurately identify and locate the changing regions in the image.The robustness and accuracy of the proposed method are validated through experiments on two public datasets.