摘要
WorldView-3是目前空间分辨率最高的商业卫星数据,具有立体采集、1∶5000测图能力,同时又具有更多的光谱波段,有利于地物识别。本文尝试基于深度学习分类工艺和方法,探讨利用0.31 m WorldView-3多光谱、高分辨率遥感影像的地物识别能力,通过选取试验区对房屋建筑区、植被、阴影、水体、棚房等城市目标进行分类试验,分析针对分米级高分辨率遥感影像城市目标分类试验工艺和关键参数设置、样本特征选取。试验结果表明模型具有一定样本容错能力,并在一定程度上可以有效挖掘WorldView-3的不同材质地物识别能力。
In this paper,using deep learning based classification work flow and algorithms,the object recognition capability with 0.31 m WorldView-3 multispectral and high resolution remote sensing image was discussed. An experiment to classify construction area,vegetation,shadow,water,shed housing etc.was carried out to explore the work flow,the key parameter setting and the sample feature selection for decimeter image classification. Finally,the model was validated with the fault tolerance and effectively mining the recognition of object with different material with WorldView-3.
出处
《测绘通报》
CSCD
北大核心
2017年第S2期40-43,共4页
Bulletin of Surveying and Mapping
基金
2017年度国家测绘地理信息局青年学术和技术带头人科研计划课题
