基于多重对抗信息生成网络的建筑景观格局特征研究  

Research on the Characteristics of Architectural Landscape Pattern Based on Multi Adversarial Information Generation Network

作　　者：谢静[1] XIE Jing(Department of Art and Design,Ma'anshan Normal College,Ma'anshan Anhui 243000,China)

机构地区：[1]马鞍山师范高等专科学校艺术设计系,安徽马鞍山243000

出　　处：《佳木斯大学学报（自然科学版）》2024年第5期98-101,135,共5页Journal of Jiamusi University：Natural Science Edition

基　　金：2021年度马鞍山师范专科学校校级《环境艺术设计专业岗课赛证综合育人改革》(2021xjgksz02)。

摘　　要：当前对于建筑景观格局特征研究的方法缺乏对其隐藏性特征进行提取,因此研究引入共享的思想构建了多重对抗信息共享生成网络模型(Generation Network for Multiple Counter Information Sharing,GNMCIS),并对其有效性和实用性进行了验证。实验结果表明,GNMCIS模型在甄别器训练出现过好状态时会对其权值参数进行及时调整,训练过差时会对其进行拉升操作,从而有效规避了梯度下降和消失的出现。同时其在50×100的像素下的训练时间为124min,在50×200的像素下的训练时间为154min,均优于对比模型。并且其监测出三个区域内最优监测点的平均温度维持在26℃左右,有效分析出相关热环境信息。综合来看,GNMCIS模型在形成具备获取特征的建筑景观格局上具备较高的有效性,对实际优化建筑景观格局上具备借鉴意义。The current methods for studying the characteristics of architectural landscape patterns lack the extraction of hidden features.Therefore,the study introduced the concept of sharing to construct a Generation Network for Multiple Counter Information Sharing(GNMCIS)model,and verified its effectiveness and practicality.The experimental results show that the GNMCIS model adjusts its weight parameters in a timely manner when the discriminator training is in a good state,and performs a pull-up operation when the training is poor,effectively avoiding the occurrence of gradient descent and disappearance.At the same time,the training time at 50×100 pixels is 124 minutes,the training time at 50×200 pixels is 154 minutes,which is better than the comparison model.And it monitors that the average temperature of the optimal monitoring points within the three regions is maintained at around 26℃,effectively analyzing relevant thermal environment information.Overall,the GNMCIS model has high effectiveness in forming architectural landscape patterns with acquired features,and has reference significance for optimizing architectural landscape patterns in practice.

关 键 词：GNMCIS 建筑景观格局 甄别器 DNN 

分 类 号：TU984.114[建筑科学—城市规划与设计]