改进U-Net网络的高分辨率遥感影像变化检测算法

High Resolution Remote Sensing Image Change Detection Based on Improved U-Net

针对高分辨率遥感影像变化检测中出现的伪变化较多、检测边界模糊、小目标漏检等问题,提出一种孪生结构的Siam-FAUnet变化检测模型。该模型可以实现端到端的变化检测任务。首先,利用改进的VGG16作为编码器提取双时相的影像特征;其次,通过空洞空间金字塔池化模块获取图像多尺度上下文信息;然后,使用流对齐模块将编码器中的低层特征融合至解码器,以此来获取影像的变化区域。实验采用公开的CDD和STAKI数据集进行训练和测试。结果表明,相对于其他主流的深度学习变化检测方法,Siam-FAUnet变化检测模型在准确率、精确率、召回率和F1分值上均有提升,表明该模型具有良好的检测性能。

Aiming at the problems of pseudo-change,blurred detection boundary and small target miss detection in high resolution remote sensing image change detection,a change detection algorithm based on siamese neural networks is proposed.Siam-FAUnet can implement end-to-end change detection tasks.Firstly,the improved VGG 16 is utilized as an encoder to extract the image features.Secondly,the multi-scale contextual information of the image is obtained through the atrous spatial pyramid pooling module.Finally,the flow alignment module is used to fuse the low-level features from the encoder to the decoder to capture the changing regions of the images.The experiments are trained and tested using publicly available CDD and STAKI datasets.The results show that the Siam-FAUnet change detection model has improved in accuracy,precision,recall and F1 score,compared with other mainstream deep learning change detection methods,indicating that the model has good detection performance.