摘要
遥感能够全面、立体、快速、有效地探明地上和地下自然资源的分布情况,这使其逐步成为从多维和宏观角度去认识世界的重要方法和手段。目前,遥感数据采集技术逐渐成熟,但精准的遥感专题信息提取主要靠全人工目视解译实现,迫切需要自动化的高精度遥感影像信息提取技术实现快速的变化检测和信息提取,为自然资源调查监测快速提供高精度产品成果。文章提出一种基于深度学习的影像变化检测方法,通过构建训练样本库,采用残差全卷积神经网络作为模型的骨干框架,进行模型训练并解译不同时相的影像,然后对解译结果求差,并采用形态学处理,从而有效辨别变化的区域。
Remote sensing can fully,stereoscopically,rapidly and effectively detect the distribution of natural resources on the ground and underground,which makes it an important means to know the world from a multi-dimensional and macro-perspective.At present,remote sensing data acquisition technology is becoming mature,but accurate remote sensing thematic information extraction is mainly realized by full manual visual interpretation.Therefore,it is urgent to realize rapid change detection and information extraction by automatic high-precision remote sensing image information extraction technology,so as to provide rapid high-precision product results for natural resource survey and monitoring.In this paper,a method of image change detection based on deep learning is presented.By constructing training sample library and adopting residual full convolution neural network as the backbone frame of model,the model training is carried out,the images at different time phases are interpreted,the differences of interpretation results are calculated,and morphological treatment is conducted,so as to effectively identify the change areas.
作者
张戬
高雅
Zhang Jian;Gao Ya(Jiangsu Provincial Research Institute of Surveying and Mapping,Nanjing 210013,China)
出处
《江苏科技信息》
2020年第32期40-44,57,共6页
Jiangsu Science and Technology Information
基金
省属科研院所江苏省测绘研究所自主科研项目,项目名称:深度学习影像解译在自然资源监测中的应用,项目编号:JSCHZZKY202002。
关键词
深度学习影像解译
全卷积神经网络
变化检测
deep learning image interpretation
full convolution neural network
change detection