Residential scene classification for gridded population sampling in developing countries using deep convolutional neural networks on satellite imagery

Abstract

BackgroundConducting surveys in low- and middle-income countries is often challenging because many areas lack a complete sampling frame, have outdated census information, or have limited data available for designing and selecting a representative sample. Geosampling is a probability-based, gridded population sampling method that addresses some of these issues by using geographic information system (GIS) tools to create logistically manageable area units for sampling. GIS grid cells are overlaid to partition a country’s existing administrative boundaries into area units that vary in size from 50 m × 50 m to 150 m × 150 m. To avoid sending interviewers to unoccupied areas, researchers manually classify grid cells as “residential” or “nonresidential” through visual inspection of aerial images. “Nonresidential” units are then excluded from sampling and data collection. This process of manually classifying sampling units has drawbacks since it is labor intensive, prone to human error, and creates the need for simplifying assumptions during calculation of design-based sampling weights. In this paper, we discuss the development of a deep learning classification model to predict whether aerial images are residential or nonresidential, thus reducing manual labor and eliminating the need for simplifying assumptions.ResultsOn our test sets, the model performs comparable to a human-level baseline in both Nigeria (94.5% accuracy) and Guatemala (96.4% accuracy), and outperforms baseline machine learning models trained on crowdsourced or remote-sensed geospatial features. Additionally, our findings suggest that this approach can work well in new areas with relatively modest amounts of training data.ConclusionsGridded population sampling methods like geosampling are becoming increasingly popular in countries with outdated or inaccurate census data because of their timeliness, flexibility, and cost. Using deep learning models directly on satellite images, we provide a novel method for sample frame construction that identifies residential gridded aerial units. In cases where manual classification of satellite images is used to (1) correct for errors in gridded population data sets or (2) classify grids where population estimates are unavailable, this methodology can help reduce annotation burden with comparable quality to human analysts.

Remote Sensing in Environmental Justice Research—A Review

“…Furthermore, temporal monitoring of changing environmental conditions can be achieved by accessing historical and recent images. Especially in combination with modern achievements in machine learning [214], the highly complex relationships between the environment, peoples' social status, and their health are about to be modeled on a new level of detail.…”

Artificial Neural Network and Machine Learning Based Methods for Population Estimation of Rohingya Refugees: Comparing Data-Driven and Satellite Image-Driven Approaches

“…satellite image, has been performed using convolutional neural network-based deep learning methodology. 28,29 However, this study uses arti¯cial neural networks due to the relatively smaller size of the study area and consequently, smaller number of features and samples as well. Statistical regression models have also been used along with satellite imagery for the task of population estimation 37,38 but di®er from the current study due to use of readily available gridded population datasets for those regions, which is not available for the current study area.…”

“…Machine learning methods such as neural network-based methods have widely been used for population estimation. [20][21][22][23][27][28][29] The task has also been performed using SVM 39 and Random Forest models. 40 Due to lack of existing studies on population estimation of Rohingya refugees, both data-driven and imagedriven, we could not provide a direct comparative study with others."…”

“…27 Deep learning and grid-based approach of satellite image classi¯cation for population estimation is also present. 28 Tang et al 20 and Folorunso et al 21 focused on data-driven approach using traditional features such as birth, death, school enrollment etc. which di®ers from the satellite image-driven approach used in this study.…”

“…which di®ers from the satellite image-driven approach used in this study. Chew et al 28 also employed the satellite image-driven approach for population estimation, albeit using convolutional neural networks rather than arti¯cial neural networks (unlike this study) for the task of population estimation. Robinson et al 29 also used satellite image and deep learning for population estimation of the U.S.…”