多通道U型网络遥感影像变化检测

Change Detection of Remote Sensing Images with Multi-channel U-shaped Network

遥感影像变化检测是遥感领域的一个重要研究方向,在农业、灾害评估和城市建设等诸多方向发挥着重要作用。目前的变化检测任务大多使用深度学习方法完成,但现存的诸多深度学习网络存在影像特征提取能力不强、变化区域不能精细化区分等问题。提出了一种多通道、多尺度特征融合的深度U型网络——MCFFNet。将Unet扩展为三通道结构,并在下采样过程中获得相应尺度特征影像的预分类特征信息和融合特征。在上采样过程中将对应尺度的特征信息融合。通过卷积激活等操作将特征图映射为单次最优变化检测结果图。在遥感影像变化检测领域常用数据集CDD和WHU数据集上实验得到了较对照方法更高的变化检测精度。

Change detection of remote sensing images is an important research direction in the field of remote sensing,which plays an important role in many fields such as agriculture,disaster assessment,and urban construction.At present,most change detection tasks are completed using deep learning methods,but many existing deep learning networks have problems such as weak image feature extraction ability and inability to finely distinguish between change regions.A deep U-shaped network MCFFNet with multi-channel and multi-scale feature fusion is proposed.Firstly,the Unet network is extended to a three-channel structure,and the pre-classification feature information and fusion features of the corresponding scale feature images are obtained during the down-sampling process.Then,during the up-sampling process,the feature information of the corresponding scale is fused.Finally,the feature map is mapped into a single optimal change detection result map through convolutional activation and other operations.Experiments on the commonly used datasets CDD and WHU in the field of remote sensing image change detection have achieved higher change detection accuracy than the methods for comparison.