基于U-Net卷积神经网络的遥感影像变化检测方法研究

Research on remote sensing image change detection method based on U-Net convolutional neural network

针对目前遥感影像传统变化检测方法中存在的影像预处理技术要求苛刻、部分环节需采取人工干预、难以做到信息提取自动化、难以处理海量多源数据等问题,基于深度学习卷积神经网络方法,进行U-Net网络模型在遥感影像变化检测中的应用研究。以水体为例,利用该方法对两时相的遥感影像进行变化检测,通过对比基于支持向量机(SVM)的分类后比较法后发现,在给予大量充分训练数据的情形下,利用该方法对试验数据进行变化检测,得到的卡帕系数Kappa为0.88,总体精度为97.06%,相比传统方法精度有所提升,说明本文方法进行变化检测有一定的可用性。研究可为自然资源调查管理提供极强的现势性数据,对开展自然资源管理工作的动态监测提供一个可行的方案。

Aiming at the problems existing in the current traditional change detection methods of remote sensing images,such as demanding image pre-processing technology,manual intervention in some aspects,difficulty in automating information extraction,and difficulty in processing massive multi-source data,the application of U-Net network model in remote sensing image change detection is studied based on the deep learning convolutional neural network method.Taking the water body as an example,this method is used to detect the changes of remote sensing images in two phases.After comparing the post classification comparison methods based on support vector machine(SVM),it is found that,given a large number of sufficient training data,the method is used to detect the changes of test data,and the Kappa coefficient obtained is 0.88,with an overall accuracy of 97.06%,which is improved compared with the traditional methods.It shows that the method in this paper has certain usability for change detection.The research can provide strong current data for natural resource survey and management,and provide a feasible scheme for dynamic monitoring of natural resource management.