摘要
量化的城市街区品质评价是街区设计规划的重要依据,图像数据是街区品质评价模型的重要维度。目前的研究中存在街区品质标注成本较高的问题。因此本文改进基于子空间的小样本学习方法,对街区卫星图像特征进行奇异分解生成类别子空间,并将训练集子空间参数继承到街区品质评估模型中。实验结果表明,在小样本街区品质评估问题上,本文方法相比传统小样本学习方法的正确率提高约30%,一致性提高约15%。
Quantitative urban block quality evaluation is an important foundation for block design and planning,and image data is an important dimension of the block quality evaluation model.Currently,there are some problems in this field of research,such as the high cost of block quality labeling.This paper improves the small sample learning method based on subspace,performs singular decomposition on the satellite image features of the block to generate class subspace,and inherits the subspace parameters of the training set into the block quality evaluation model.The experimental results show that this method is about 30%more accurate and 15%more consistent than the traditional small sample learning method.
作者
郭茂祖
王偲佳
王鹏跃
赵玲玲
GUO Maozu;WANG Sijia;WANG Pengyue;ZHAO Lingling(School of Electrical and Information Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Beijing Key Laboratory of Intelligent Processing for Building Big Data,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;School of Architecture and Urban Planning,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001,China)
出处
《智能系统学报》
CSCD
北大核心
2022年第6期1254-1262,共9页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金面上项目(61871020)
北京市属高校高水平创新团队建设计划项目(IDHT20190506).
关键词
街区品质评估
卫星图
小样本学习
自适应子空间
深度神经网络
奇异值分解
不平衡数据集
欠采样
block quality assessment
satellite map
few-shot learning
adaptive subspace
depth neural network
singular value decomposition
unbalanced dataset
under-sampling
