摘要
遥感影像建筑物信息提取对于自然资源监测、土地利用现状调查、生态修复等具有重要的现实意义。但在实际应用中,建筑物提取面临"小目标""有遮挡"的问题,导致识别效果不理想。本文基于高分辨率遥感影像,提出运用多示例卷积神经网络的方法对建筑物场景进行识别。试验表明,多示例卷积神经网络相较于经典的卷积神经网络对建筑物场景有更好的识别效果,尤其是"小目标""有遮挡"的建筑物场景,识别效果有显著的提升。
Extracting building information from remote sensing imagery plays a significant role in monitoring natural resources,investigating the status of land use and ecological restoration,etc.But in practical applications,building extraction faces the problem of"small target","covered",leading to an unsatisfactory recognition result.A multi-instance convolutional neural network method was developed for building scene recognition with high-resolution remote sensing imagery in this study.The results illustrate that the multi-instance convolutional neural network has a better recognition effect than that of the classical convolutional neural network on the building scene,especially for the"small target","covered"building scene.
作者
刘强
解加粉
陈建忠
孙如瑶
赵中飞
LIU Qiang;XIE Jiafen;CHEN Jianzhong;SUN Ruyao;ZHAO Zhongfei(Shandong Provincial Institute of Land Surveying and Mapping,Jinan 250000,China)
出处
《测绘通报》
CSCD
北大核心
2021年第S01期124-128,182,共6页
Bulletin of Surveying and Mapping
关键词
多示例卷积神经网络
建筑物场景识别
高分辨率遥感影像
multi-instance convolutional neural network
building scene recognition
high-resolution remote sensing image
