The relative spatial scale relationship of observers and ecosystem and their aesthetic dynamic interaction process are fundamental to evaluation and optimization of aesthetic ecosystem service(AES).A comprehensive and...The relative spatial scale relationship of observers and ecosystem and their aesthetic dynamic interaction process are fundamental to evaluation and optimization of aesthetic ecosystem service(AES).A comprehensive and efficient framework for the assessment of AES is lack in the integration of scale relationship and dynamic process.This study took 9 villages in 4 different developmental stages(traditional,folk,rapidly changed,newly built)in Honghe Hani Rice Terraces,a world heritage site,as the research object.From two scales,viewing from inside and outside,the bi-scale assessing framework was established,which includes the three components of interaction process,connection area(as precondition of interaction),quality(as result of interaction),and influencing factors of quality(elements’characteristics of villages).Among them,the connection areas were evaluated with visual and traffic accessibility along the route.The quality and influencing factors were evaluated through participatory preferences methods by expert group.The influencing factors include 9 characteristics,such as,space size,architecture layout,vegetation species richness,color diversity.The results suggested that villages with high AES quality and low accessibility need to be optimized,and the key influencing factors are space size,architecture layout,color harmony and surrounding sanitation.Therefore,the bi-scale assessing framework can provide important references for decision making and visual protection regulations on the villages.展开更多
脑电信号(electroencephalogram,EEG)在情感识别领域受到广泛关注。然而,现有方法多侧重于空间和时间维度,却忽视频段维度,存在时空频特征提取不足问题。提出一种基于多尺度卷积与时序建模的特征融合网络(multi-scale convolution and t...脑电信号(electroencephalogram,EEG)在情感识别领域受到广泛关注。然而,现有方法多侧重于空间和时间维度,却忽视频段维度,存在时空频特征提取不足问题。提出一种基于多尺度卷积与时序建模的特征融合网络(multi-scale convolution and temporal modeling for feature fusion network,MSCTF-Net)进行脑电情绪识别。将脑电信号重构为多维形式输入,设计多尺度空间卷积和频段卷积模块,从三维空频矩阵的空间和频段维度提取空间-频率信息;引入双向长短时记忆网络(Bi-LSTM)对二维时频矩阵进行时序建模,提取时间-频率特征;提出门控特征融合模块进行特征融合。模型在SEED数据集上的平均准确率为96.63%,在SEED-Ⅳ数据集上的平均准确率为91.58%,优于现有的多种深度学习方法。展开更多
基金funded by the National Natural Science Foundation of China(grant numbers 41761115,41271203)。
文摘The relative spatial scale relationship of observers and ecosystem and their aesthetic dynamic interaction process are fundamental to evaluation and optimization of aesthetic ecosystem service(AES).A comprehensive and efficient framework for the assessment of AES is lack in the integration of scale relationship and dynamic process.This study took 9 villages in 4 different developmental stages(traditional,folk,rapidly changed,newly built)in Honghe Hani Rice Terraces,a world heritage site,as the research object.From two scales,viewing from inside and outside,the bi-scale assessing framework was established,which includes the three components of interaction process,connection area(as precondition of interaction),quality(as result of interaction),and influencing factors of quality(elements’characteristics of villages).Among them,the connection areas were evaluated with visual and traffic accessibility along the route.The quality and influencing factors were evaluated through participatory preferences methods by expert group.The influencing factors include 9 characteristics,such as,space size,architecture layout,vegetation species richness,color diversity.The results suggested that villages with high AES quality and low accessibility need to be optimized,and the key influencing factors are space size,architecture layout,color harmony and surrounding sanitation.Therefore,the bi-scale assessing framework can provide important references for decision making and visual protection regulations on the villages.
文摘脑电信号(electroencephalogram,EEG)在情感识别领域受到广泛关注。然而,现有方法多侧重于空间和时间维度,却忽视频段维度,存在时空频特征提取不足问题。提出一种基于多尺度卷积与时序建模的特征融合网络(multi-scale convolution and temporal modeling for feature fusion network,MSCTF-Net)进行脑电情绪识别。将脑电信号重构为多维形式输入,设计多尺度空间卷积和频段卷积模块,从三维空频矩阵的空间和频段维度提取空间-频率信息;引入双向长短时记忆网络(Bi-LSTM)对二维时频矩阵进行时序建模,提取时间-频率特征;提出门控特征融合模块进行特征融合。模型在SEED数据集上的平均准确率为96.63%,在SEED-Ⅳ数据集上的平均准确率为91.58%,优于现有的多种深度学习方法。
