【作者】 李嘉恒； 武越； 公茂果； 张明阳； 王善峰；

【Author】 Jiaheng Li;Yue Wu;Maoguo Gong;Mingyang Zhang;Shanfeng Wang;Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education,International Research Center for Intelligent Perception and Computation,Xidian University;School of Computer Science and Technology,Xidian University;School of Cyber Engineering,Xidian University;

【机构】 西安电子科技大学电子工程学院； 西安电子科技大学计算机科学与技术学院； 西安电子科技大学网络与信息安全学院；

【摘要】 光学图像和SAR图像之间的互补特性及巨大的特征差异使得基于它们的异源遥感图像变化检测在实际应用中变得相当有意义而且富有挑战性。在本文中,我们提出了一种新颖的基于卷积耦合自编码器的无监督异源图像变化检测方法。卷积耦合自编码器中的卷积层首先将两副输入图像中每个像素点及其邻域编码为特征向量,该过程可以代替降噪算法并在两个图像之间建立更一致的特征表示。接着,一个经典的自编码器结构将两个特征空间连接,以探索每对特征向量间的潜在关系,并学习特征对之间的映射函数。最后,我们通过特征相似性分析建立变化图,并采用阈值方法生成最终的变化检测结果。我们对两个真实数据集的实验结果表明,与现有的几种方法相比,该方法具有更好的性能。更多还原

【Abstract】 The complementary properties between optical images and SAR images with their huge feature differences make heterogeneous images change detection based on them become significant and challenging.In this paper,we propose a novel convolutional coupling autoencoder for heterogeneous images change detection.The convolutional layers of the convolutional coupling autoencoder firstly transform the local neighborhood of each pixel in the two input images to a feature vector,that can replace the denoising process and provide a more consistent feature representation between the two images.Then,a classic autoencoder which connects the two feature space is applied to explore the inner relationships between them.And the autoencoder can be used to learn a mapping function of the feature-pairs.Finally,we can build a change map through feature similarity analysis and apply a thresholding algorithm to generate the ultimate change detection result.Our experimental results on two real datasets demonstrate the promising performance of the proposed framework compared with several existing approaches.更多还原