差异特征融合的无监督SAR图像变化检测

针对Siamese网络忽略不同层级差异特征之间的关联导致检测精度有限的问题，提出了基于差异特征融合的无监督SAR(synthetic aperture radar)图像变化检测算法。首先，利用对数比值算子和均值比值算子构建两幅信息互补的差异图，通过引入能量矩阵对差异图进行像素级融合以提高其信噪比；其次，设计了一种基于差异特征融合的Siamese网络(difference feature fusion for Siamese，DFF-Siamese)，该网络能够通过差异特征提取模块在决策层综合衡量不同层级特征之间的差异程度，从而有效增强网络的特征表达能力；最后，利用模糊聚类算法对融合结果进行分类构建“伪标签”，用于训练DFF-Siamese网络以实现高精度SAR图像变化检测。在3组真实遥感数据集上的实验结果表明，本文提出的算法与其他对比算法相比具有更高的检测精度和更低的错误率。

Unsupervised SAR image change detection based on difference feature fusion

The Siamese network ignores the correlation between different levels of different features, leading to limited detection accuracy. To overcome this limitation, an unsupervised synthetic aperture radar (SAR) image change detection algorithm based on the fusion of different features is proposed. The method uses log-ratio and mean-ratio operators to construct two complementary difference images and introduces an energy matrix to perform pixel-level fusion on these difference maps to improve its signal-noise ratio. Next, a Siamese network based on different feature fusion (DFF-Siamese) is designed. By introducing a difference feature extraction module, the DFF-Siamese comprehensively measure the differences and similarities between the multitemporal images using high-level semantic features, effectively improving the feature representation capability of the network. Finally, a fuzzy clustering algorithm is used to classify the fusion images to construct fake labels used to train the developed DFF-Siamese network, achieving accurate change detection for the SAR images. Experimental results from three real remote sensing datasets showed that the proposed method exhibits higher change detection accuracy with lower error rates compared with other popular algorithms.