高分影像变化检测的孪生差分特征融合网络

Siam-differential feature fusion network for change detection of high-resolution images

针对目前高空间分辨率光学遥感影像地表变化检测面临的挑战,该文提出一种孪生差分特征融合网络方法,一方面增强了对深层变化特征的提取能力,通过差分特征能更好地引导网络学习;另一方面在网络末端引入深监督策略,有效融合多尺度信息,充分利用不同语义层次特征,从而生成高精度的变化检测结果。此外,还设计了顾及样本不均衡问题的损失函数,降低正负样本极度不平衡对模型训练的负面影响。为了评估该文提出方法的有效性和优势,在两个公开变化检测数据集上将其与5种具有代表性的变化检测方法进行对比实验,结果证明该方法能有效提升变化检测的精度,并且对尺度差异明显的地物有较强的检测能力,在轮廓细化上具有显著优势。

To address the challenges of land surface change detection in current high spatial resolution optical remote sensing images,this paper proposes a siam-differential feature fusion network method,which on the one hand improves the model's ability to extract deep change features and the differential features can better guide network learning;on the other hand,it introduces a deep supervision strategy at the end of the network to effectively fuse multi-scale information and make full use of different semantic level features,so as to generate high accuracy change detection results.In addition,this paper also designs a loss function that takes into account the sample imbalance problem to reduce the negative impact of the extreme imbalance between positive and negative samples on the model training.In order to evaluate the effectiveness and advantages of the proposed method,the experiments are conducted on two public change detection datasets and compared with five representative change detection methods,and the results demonstrate that the proposed method can effectively improve the accuracy of change detection,and can better detect features with significant scale differences,and has significant advantages in contour refinement.