基于多源信息与深度学习特征提取的人口空间抽样方法

Spatial sampling of population based on multi-source information and deep learning feature extraction

人口抽样调查是通过人口样本估算区域人口总体的一种手段。由于人口分布通常具有空间差异性,传统的抽样调查理论难以满足日益增长的空间抽样需求,合理高效的人口空间抽样调查方法对于人口统计、研究人类活动、解决城市问题等有重要意义。本文提出一种基于多源信息与深度学习特征提取的人口空间抽样方法。在不透水面信息的辅助下,利用四叉树分割进行分层抽样,初步选择出可能存在人口分布的调查样本,并通过深度学习的常用模型——卷积神经网络估算样本建筑物密度,以辅助最终调查样本的选择与调查方案的制定。研究结果证明,该方法能够有效地筛选与人口分布密切相关的抽样区域,排除大量的无用样本,提高了人口调查的效率,节约了大量调查成本。

Population sampling survey is a means to estimate the population of a region through population samples.Due to the spatial difference of population distribution,the traditional sampling survey theory is difficult to meet the growing demand of spatial sampling.The research on reasonable and efficient population spatial sampling survey method is of great significance to population statistics,human activities and urban problems.This paper proposes a population spatial sampling method based on multi-source information and deep learning feature extraction.Firstly,we use quadtree segmentation for stratified sampling with the help of impervious surface information,initially select the survey samples that may have population distribution.Secondly,we estimate building density of sample by convolution neural network that is a common model of deep learning to assist in the final sample selection and survey scheme formulation.The results show that this method can effectively screen the sampling areas closely related to population distribution,eliminate a large number of useless samples,improve the efficiency of population survey and save a lot of survey costs.