DOI:10.13905/j.cnki.dwjz.2022.12.002基于CGAN的居住区景观功能平面生成方法GENERATIVEDESIGNMETHODOFLANDSCAPEFUNCTIONALLAYOUTINRESIDENTIALAREASBASEDONCONDITIONALGENERATIVEADVERSARIALNETS曲广滨,薛博洋(哈尔滨工业大学建筑学院,寒地城乡人居环境科学与技术工业和信息化部重点实验室,哈尔滨150001)QUGuangbin,XUEBoyang(SchoolofArchitecture,HarbinInstituteofTechnology,KeyLaboratoryofColdRegionUrbanandRuralHumanSettlementEnvironmentScienceandTechnology,MinistryofIndustryandInformationTechnology,Harbin150001,China)【摘要】在居住区景观方案设计过程中方案需要满足设计规范的要求,这使设计师往往需要对方案进行反复的修改,导致设计效率较低。文中以深度学习为切入点,基于CGAN构建居住区景观功能平面生成设计方法,其中包括数据集制作、CGAN模型构建和生成图像的验证。结果表明该方法生成的方案绿地率在37%以上,大寒日2h以上日照的绿地占比在54%以上,可以满足设计规范要求。模型训练完成后,可以在3s内生成多个居住区景观功能平面方案。该方法可以为设计师提供多方案比对的参考,减少方案因不满足规范而做出的修改,从而提高居住区景观设计效率和质量。【关键词】风景园林;居住区景观设计;验证评价【中图分类号】TU984.12【文献标志码】A【文章编号】1001-6864(2022)12-0005-05Abstract:Residentiallandscapedesignaresubjectedtotherequirementsofdesignspecifications,whichmakesde⁃signersoftenmodifythedesignrepeatedly,resultinginlowdesignefficiency.Takingdeeplearningasthestartingpoint,thedesignmethodofresidentiallandscapefunctionallayoutgenerationisestablished.Themethodisbasedonconditionalgenerativeadversarialnets(CGAN),includingdatasetproduction,CGANmodelconstruction,andverifi⁃cationofgeneratedimages.Theresultsshowthatthegreenspacerateofthelayoutgeneratedbythemethodismorethan37%,andthegreenspaceexposedtosunshineformorethan2hingreatcolddaysformorethan54%,whichmeetstherequirementsofdesignspecifications.Within3s,residentiallandscapedesignsaregeneratedaftertrainingthemodules.Thismethodmayprovidethereferencefordesignerstocomparemultiplelandscapefunctionallayouts,andreducethemodificationofthedesignduetotheinconsistencywiththespecific...