摘要
针对当前基于卷积神经网络的建筑提取方法在数据输入层对建筑语义信息利用不足的问题,提出一种融合类别语义特征的卷积神经网络建筑物提取方法。首先,从影像上提取形态学建筑指数,该指数能够直接表征建筑覆盖信息,是一种高层次的语义特征;然后,将该建筑语义特征作为原始影像的补充通道一起输入到卷积神经网络模型中训练,从数据层进一步增强建筑与背景的可分性。采用国际上公开的具有多种地物形态的标准建筑数据集验证本文方法的有效性。实验表明,所提出的方法取得了满意的精度(准确率为85.6%,召回率为93.1%,F指数为88.4%),相对于原始的RGB影像输入,建筑物语义特征的加入整体上提升了建筑物提取的精度。
The traditional methods for building extraction based on convolutional neural networks are insufficient in using building semantic information.To solve this problem,this study proposes a method for building extraction based on convolutional neural network fused with categorical semantic feature.First,the morphological building index is extracted from the images.This index,a kind of high-level semantic feature,can directly describe the building coverage information.Then,the building semantic feature is served as a supplementary channel of the original image and put into the convolutional neural network for model training,in order to strengthen the seperability between buildings and backgrounds at the data level.The standard open-source building dataset with various land covers is used to evaluate the effectiveness of the proposed method.The experiments show that the proposed method achieves satisfactory results(precision is 85.6%,recall is 93.1%,and F-score is 88.4%).Compared to the input of the original RGB image,the addition of building semantic feature improves the overall accuracy of the building extraction results.
作者
张涛
丁乐乐
史芙蓉
ZHANG Tao;DING Lele;SHI Furong(Tianjin Survey Design Institute Group Co. Ltd.,Tianjin 300191,China;State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan 430079,China)
出处
《遥感信息》
CSCD
北大核心
2021年第5期49-55,共7页
Remote Sensing Information
基金
天津市重点研发计划科技支撑重点项目(18YFZCSF00620)
天津市重点研发计划院市合作项目(18YFYSZC00120)。
关键词
建筑提取
语义分割
语义特征
高分辨率影像
深度学习
building extraction
semantic segmentation
semantic feature
high-resolution image
deep learning