基于双注意力融合的遥感影像道路提取  

Road Extraction from Remote Sensing Images Based on Dual Attention Fusion

作　　者：许鑫 邵飞 XU Xin;SHAOFei(Field Engineering College,Nanjing Jiangsu210000,China)

机构地区：[1]陆军工程大学野战工程学院,江苏南京210000

出　　处：《机电产品开发与创新》2023年第4期129-131,156,共4页Development & Innovation of Machinery & Electrical Products

摘　　要：获取准确的道路信息对城市规划、交通导航、数字地图更新和构建实时道路交通信息系统等方面具有重要促进作用。然而,遥感图像中的道路信息容易受到周边复杂环境以及道路尺寸等因素的干扰,从而影响道路信息提取的精度。针对这一问题,本文提出了一种基于双注意力融合的遥感影像道路提取方法。研究结果显示,本研究方法提取的道路信息的使用精度(Precision)、F1分数(F1-score)和召回率(Recall)分别为0.83、0.71、0.55,高于DLinkNet、DeepLabv3+、SegNet和UNet等模型所获得的结果。这充分证明了本文提出方法的有效性和优越性。Obtaining accurate road information plays an important role in promoting urban planning,traffic navigation,digital map updating and building real-time road traffic information system.However,road information in remote sensing image is susceptible to the interference of complex surrounding environment and road size,so as to affect the accuracy of road information extraction.To solve this problem,a road extraction method based on dual attention fusion is proposed in this paper.Research results showed that Precision,F1-score and Recall of road information extracted by this research method were 0.83,0.71 and 0.55,respectively,which are higher than those obtained by DLinkNet,DeepLabv3+,SegNet and UNet.This fully proves the effectiveness and superiority of the proposed method.

关 键 词：遥感影像 深度学习 双注意力 道路信息 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]