摘要： 在高分辨率光学遥感图像变化检测中,多数面向对象的方法只能利用简单的特征组合得到对象特征,难以进行高层特征的设计和提取。针对该问题,提出一种基于深度置信网络和对象融合的图像变化检测方法。将变化检测转化为二分类问题,并把图像像素作为分类单元,在特征学习和分类阶段设计多尺度的图像特征学习和分类方法,以充分利用图像目标的上下文信息。在此基础上设计基于对象的分类融合方法,对利用深度置信网络分类得到的结果进行融合,从而减小局部噪声的影响。在QucikBird影像数据集上的实验结果表明,该方法可有效提高图像变化检测的准确率。

关键词: 图像变化检测, 遥感图像, 深度置信网络, 对象融合, 多尺度特征

Abstract: In high-resolution optical remote sensing image change detection,most of the object-oriented method can only use simple features combination to get the object features,which cannot implement design and characteristic extraction for high-level features.Aiming at this problem,a method based on Deep Belief Network(DBN) and object fusion is proposed in this paper.The change detection is transformed into dichotomous problem,and image pixel is used as the classification unit.Multi-scale image feature learning and classification methods are designed in feature learning and classification stage to make full use of contextual information of image targets.On this basis,a fusion method based on object classification is proposed to fuse the classification results classified by DBN,so as to reduce the effects of local noise.Experiments are conducted on the dataset acquired by QuickBird,and the results show that the proposed method can improve the accuracy of image change detection.

Key words: image change detection, remote sensing image, Deep Belief Network(DBN), object fusion, multiscale feature

中图分类号: 

• TP751