Efficient Poverty Mapping from High Resolution Remote Sensing Images

However, the accuracy afforded by high-resolution imagery comes at a cost, as such imagery is extremely expensive to purchase at scale. This creates a substantial hurdle to the efficient scaling and widespread adoption of high-resolution-based approaches. To reduce acquisition costs while maintaining accuracy, a reinforcement learning approach is proposed. This approach uses free low-resolution imagery to dynamically identify where to acquire costly high-resolution images, prior to performing a deep learning task on the high-resolution images. The method leverages publicly available Sentinel-2 images (10-30m) to sample smaller amounts of high-resolution images (<1m). This concept is inspired by recent studies in computer vision literature that perform conditional inference to reduce computational complexity of convolutional networks in test time. The proposed approach is applied to the task of poverty prediction in Uganda, building on an earlier approach that used object detection to count objects and use these counts to predict poverty. The method exceeds previous performance benchmarks on this task while using 80% fewer high-resolution images. This could be useful in many domains that require high-resolution imagery. Poverty is typically measured using consumption expenditure, the value of all the goods and services consumed by a household in a given period. A household or individual is said to be poverty-stricken if their measured consumption expenditure falls below a defined threshold (currently $1.90 per capita per day). The focus is on this consumption expenditure as the outcome of interest, using "poverty" as shorthand for "consumption expenditure" throughout the paper. The dataset consists of data on consumption expenditure (poverty) from the Living Standards Measurement Study (LSMS) survey conducted in Uganda by the Uganda Bureau of Statistics between 2011 and 2012. The survey consists of data from 2,716 households in Uganda, grouped into unique locations called clusters. The latitude and longitude of a cluster are given, with noise of up to 5 km added in each direction by the surveyors to protect respondent privacy. Individual household locations in each cluster are also withheld to preserve anonymity. High-resolution satellite imagery is acquired for Uganda, corresponding to each cluster. The high-resolution satellite imagery is represented by 34x34 images of 1000x1000 pixels each with 3 channels, arranged in a 34x34 square grid. This corresponds to a 10kmx10km spatial neighborhood centered at the cluster location. The high-resolution images come from DigitalGlobe satellites with 3 bands (RGB) and 30cm resolution. Low-resolution satellite imagery is also acquired, which is free of charge and has a resolution of 10-30m.