基于CGAN的居住区景观功能平面生成方法

GENERATIVE DESIGN METHOD OF LANDSCAPE FUNCTIONAL LAYOUT IN RESIDENTIAL AREAS BASED ON CONDITIONAL GENERATIVE ADVERSARIAL NETS

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。

Residential landscape design are subjected to the requirements of design specifications,which makes designers often modify the design repeatedly,resulting in low design efficiency.Taking deep learning as the starting point,the design method of residential landscape functional layout generation is established.The method is based on conditional generative adversarial nets(CGAN),including data set production,CGAN model construction,and verification of generated images.The results show that the green space rate of the layout generated by the method is more than 37%,and the green space exposed to sunshine for more than 2h in great cold days for more than 54%,which meets the requirements of design specifications.Within 3s,residential landscape designs are generated after training the modules.This method may provide the reference for designers to compare multiple landscape functional layouts,and reduce the modification of the design due to the inconsistency with the specification,improving the design efficiency and quality.