4 Michael Xie and Neal Jean and Marshall Burke and David Lobell and Stefano Ermon. 2016. Transfer Learning from Deep Features for Remote Sensing and Poverty Mapping. Association for the Advancement of Artificial
Intelligence (www.aaai.org)
Department of Computer Science, Stanford University
fxie, nealjean, [log in to unmask]
Department of Earth System Science, Stanford University
fmburke,[log in to unmask]
Abstract
The lack of reliable data in developing countries is a major obstacle to sustainable development, food security, and disaster relief. Poverty data, for example, is typically scarce, sparse in coverage, and labor-intensive to obtain. Remote sensing data such as high-resolution satellite imagery, on the other hand, is becoming increasingly available and inexpensive. Unfortunately, such data is highly unstructured and currently no techniques exist to automatically extract useful insights to inform policy decisions and help direct humanitarian efforts. We propose a novel machine learning approach to extract large-scale socioeconomic indicators from highresolution satellite imagery. The main challenge is that training data is very scarce, making it difficult to apply modern techniques such as Convolutional Neural Networks (CNN). We therefore propose a transfer learning approach where nighttime light intensities are used as a data-rich proxy. We train a fully convolutional CNN model to predict nighttime lights from daytime imagery, simultaneously learning features that are useful for poverty prediction. The model learns filters identifying different terrains and man-made structures, including roads, buildings, and farmlands, without any supervision beyond nighttime lights. We demonstrate that these learned features are highly informative for poverty mapping, even approaching the predictive performance of survey data collected in the field.
And
5. David Lobell. 2013. The use of satellite data for crop yield gap analysis. Field Crops Research 143: 56-64.
Field experiments and simulation models are useful tools for understanding crop yield gaps, but scaling up these approaches to understand entire regions over time has remained a considerable challenge. Satellite data have repeatedly been shown to provide information that, by themselves or in combination with other data and models, can accurately measure crop yields in farmers’ fields. The resulting yield maps provide a unique opportunity to overcome both spatial and temporal scaling challenges and thus improve understanding of crop yield gaps. This review discusses the use of remote sensing to measure the magnitude and causes of yield gaps. Examples from previous work demonstrate the utility of remote sensing, but many areas of possible application remain unexplored. Two simple yet useful approaches are presented that measure the persistence of yield differences between fields, which in combination with maps of average yields can be used to direct further study of specific factors. Whereas the use of remote sensing may have historically been restricted by the cost and availability of fine resolution data, this impediment is rapidly receding.
Cheers,
Robert A. Washington-Allen