从遥感影像中提取建筑物是计算机视觉领域的一项基本任务.近年来,基于深度学习的方法已成为遥感影像中自动提取建筑物的主流方法.由于建筑物结构复杂、尺度多样等特点,从遥感影像中准确高效地提取建筑物仍然是一个挑战.针对建筑物尺度多样导致在提取过程中无法同时兼顾小型和大型建筑物的问题,提出一种基于多尺度指导的遥感影像建筑物提取网络.通过4条路径分别提取小尺度、大尺度以及其他尺度特征,通过基于交互的尺度指导模块和可选择核(Selective Kernel,SK)卷积模块分别对特征进行指导和优化特征,融合不同路径提取的特征预测建筑物信息.分别在WHU数据集和inria数据集上评估提出网络的有效性,对比实验结果表明,所提出的网络在WHU数据集上的交并比(Intersection over Union,IoU)较网络 SegNet、ENet、MMB-Net、Refine-UNet、MAP-Net 分别提高 2.37％、1.48％、1.05％、0.83％、0.59％,在inria数据集上IoU较其他网络分别提高3.65％、4.93％、2.42％、1.82％、1.21％.结果显示,所提出的网络是一种有效、提取结果完整性更高、鲁棒性更强的目标提取网络.
Building Extraction Network Based on Multi-scale Guidance in Remote Sensing Images
Extracting buildings from remote sensing images is a fundamental task in the field of computer vision.In recent years,deep learning based methods have become the mainstream method for automatically extracting buildings from remote sensing images.Due to the complex structure and diverse scale of buildings,accurately and efficiently extracting buildings from remote sensing images remains a challenge.A remote sensing image building extraction network based on multi-scale guidance is proposed to address the issue of the inability to simultaneously consider small and large buildings during the extraction process due to the diversity of building scales.This network extracts small-scale,large-scale and other scale features through four paths,and the features are then guided and optimzed by an interaction-based scale guidance module and a Selective Kernel(SK)convolution module,respectively.Finally,the features extracted from different paths are fused to predict building information.The effectiveness of the proposed method is evaluated on the WHU dataset and the inria dataset respectively.Comparative experimental results show that the Intersection over Union(IoU)of the proposed method on the WHU dataset is 2.37％,1.48％,1.05％,0.83％and 0.59％higher than SegNet,ENet,MMB-Net,Refine-Net and MAP-Net respectively.In the inria dataset,the IoU is 3.65％,4.93％,2.42％,1.82％and 1.21％higher than other networks respectively.The results show that the proposed method is an effective,more complete,and robust object extraction method.
deep learningremote sensing imagesbuilding extractionmulti-scale guidance
国家科技期刊平台
NSTL
万方数据
