基于深度学习的自然灾害遥感影像语义分割①王关茗,胡乃平(青岛科技大学信息科学技术学院,青岛266061)通信作者:王关茗,E-mail:348102401@qq.com摘要:自然灾害种类繁多,通过遥感影像语义分割相对比较困难.为了能够更好实现遥感影像分割,本文提出一种基于生成对抗网络的3层遥感影像语义分割模型,针对不同场景的解析,基于全卷积神经网络FCN,设计一种多层次的遥感语义分割框架.有效对遥感图像语义分割进行处理,从而提高了模型的分割精度.实验表明利用这种模型是有效的,特别是受损建筑的分割结果,mIoU为82.28%,通过该模型与其他网络模型进行对比,其性能评价指标明显优于其他网络模型.最后,通过对自然灾害各种场景影像进行分析,为应急管理部门提供一份可靠的数据报告.关键词:自然灾害;遥感影像;深度学习;语义分割引用格式:王关茗,胡乃平.基于深度学习的自然灾害遥感影像语义分割.计算机系统应用,2023,32(2):322–328.http://www.c-s-a.org.cn/1003-3254/8994.htmlSemanticSegmentationofNaturalDisasterRemoteSensingImageBasedonDeepLearningWANGGuan-Ming,HUNai-Ping(CollegeofInformationScienceandTechnology,QingdaoUniversityofScienceandTechnology,Qingdao266061,China)Abstract:Therearemanykindsofnaturaldisasters,anditisrelativelydifficulttosemanticallysegmentremotesensingimages.Inordertobetterrealizeremotesensingimagesegmentation,thisstudyproposesathree-layersemanticsegmentationmodelforremotesensingimagesbasedonagenerativeadversarialnetwork.Fortheanalysisofdifferentscenes,amulti-levelremote-sensingsemanticsegmentationframeworkisdesignedbasedonafullyconvolutionalnetwork(FCN).Thesemanticsegmentationofremotesensingimagesiseffectivelyperformed,andthusthesegmentationaccuracyofthemodelisenhanced.Experimentsshowthatthismodeliseffective,whichcanbedirectlyobservedfromthesegmentationresultsofdamagedbuildings,withmIoUbeing82.28%.Inaddition,thismodeliscomparedwithothernetworkmodels,anditsperformanceevaluationindexissignificantlybetterthanthatofothernetworkmodels.Finally,areliabledatareportisprovidedtoemergencymanagementdepartmentsbyanalyzingvarioussceneimagesofnaturaldisasters.Keywords:naturaldisaster;remotesensingimage;deeplearning;semanticsegmentation近年来,地...
