基于DeepLabv3+模型的高分辨率遥感影像建筑物提取

Building Extraction on High Spatial Resolution Remote Sensing Image Based on DeepLabv3+Segmentation Model

从遥感影像中自动化地检测和提取建筑物,在智慧城市建设、土地利用调查、灾害应急评估、军事目标侦察等领域具有极为重要的意义。提出一种基于DeepLabv3+语义分割模型的高分辨率遥感影像建筑物提取方法,建立一套多源、多尺度的高分辨率建筑物样本数据集,依托深度学习方法和数据集,实现了模型训练、建筑物提取以及精度评估。实验结果表明,所提算法提取建筑物准确率可达94.4%,精确率为90.46%,召回率为92.94%,MIoU为88.27%,可实现建筑物有效提取。该研究可以为高分辨率遥感影像建筑物提取提供参考。

Automatic extraction of buildings from remote sensing images is significant to smart city construction,land use survey,emergency disaster assessment and military target reconnaissance.A building extraction method is proposed on high spatial resolution remote sensing image based on deeplabv3+segmentation model,a multi-source,multi-scale,high-resolution building dataset is created,relying on deep learning and building dataset model training,building extraction and accuracy evaluation is implemented.Experimental result shows the building extraction accuracy rate can reach 94.4%,the precision rate is 90.46%,the recall rate is 92.94%,the MIOU is 88.27%,which can extract buildings effectively.