一种改进U 2-Net的高分辨率遥感影像道路提取方法  

An Improved U 2-Net Method for Road Extraction in High-resolution Remote Sensing Images

作　　者：许锐 庄振兴 黄风华 XU Rui;ZHUANG Zhenxing;HUANG Fenghua(School of Computer Science and Mathematics,Fujian University of Technology,Fuzhou 350118,China;Fujian Key Laboratory of Spatial Information Perception and Intelligent Processing(Yango University),Fuzhou 350015,China)

机构地区：[1]福建理工大学计算机科学与数学学院,福州350118 [2]福建省空间信息感知与智能处理重点实验室(阳光学院),福州350015

出　　处：《遥感信息》2024年第4期33-43,共11页Remote Sensing Information

基　　金：福建省空间信息感知与智能处理重点实验室开放基金(阳光学院,FKLSIPIP1020)。

摘　　要：目前基于深度学习的道路提取方法普遍存在忽略影像细节特征的问题。针对此问题,设计了一种改进U 2-Net的高分辨率遥感影像道路信息提取新方法。该方法以U 2-Net为主体框架,引入了卷积注意力机制模块和自注意力机制模块,既增加影像的全局语义信息又保留了空间细节特征,实现不同类型特征的有效融合。在DeepGlobe和CHN6-CUG两个道路数据集上的实验结果表明,该方法具有更强的特征提取和抗干扰能力,整体性能优于其他同类研究成果,能够更有效地从高分辨率遥感影像中提取道路。Currently,there is a common problem in deep learning-based road extraction methods,which is the tendency to ignore the detailed features of images.To address this issue,this paper proposes a novel approach for extracting road information from high-resolution remote sensing images by enhancing the U 2-N et al gorithm.This method incorporates the modules of convolutional attention mechanism and self-attention mechanism into the original U 2-Net model,which not only increases the global semantic information of images but also preserves sufficient spatial features,achieving effective fusion of different types of features.Experimental results on two road datasets,which are DeepGlobe and CHN6-CUG,demonstrate that the proposed method has stronger feature extraction and anti-interference capabilities,and overall performance is better than other similar research achievements,enabling more effective extraction of roads from high-resolution remote sensing images.

关 键 词：深度学习 道路提取 语义分割 U 2-Net 双注意力机制 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]