A team of MIT scientists has developed software capable of analyzing existing satellite imagery to identify separate structures, significantly speeding up the process of deciding where to apportion aid in rural areas in the developing world. The system will be tested against existing methods on a project in India.
Identifying suitable sites for the deployment of aid programs, such as supplying solar-powered electricity systems, is a difficult and time-consuming task. Due to poor mapping of the size and location of structures, it's usually necessary to put people on the ground to perform lengthy field studies, picking out appropriate locations for the projects.
A team of graduate students from MIT has developed a new system to tackle the problem, creating software that allows for large parts of the selection process to be automated. To test the system, the team focused on two projects, the first of which centered on selecting the poorest villages in sub-Saharan Africa, providing residents with unrestricted cash grants for buying livestock, equipment, or whatever else is required. To select the villages most in need of support, the grant-giving agency adopted a system of counting the percentage of houses with thatched roofs in comparison to those with more costly metal roofs.
A second project focused on selecting rural villages in India suitable for the installation of solar-powered microgrids. Once villages were shortlisted, it was necessary to determine the best placement of panels, battery storage systems and power distribution networks.
Both of these projects would usually require exploratory fieldwork before appropriate sites could be identified. However, with the new software, potential sites can be selected automatically, significantly cutting down on planning time.
In order for the software to identify structures, the first step was to have people study satellite images, manually picking out buildings. These served as examples that the software could use to create general criteria for autonomously identifying structures.
Adding a greater number of examples to the software allows it to more accurately identify individual buildings, overcoming difficulties such as houses with a color similar to the ground around them, or structures that are close together, appearing to be a single building instead of two.
For the African project, the team then used the same method to detect thatch and metal roofs, with the later being more reflective. With the Indian project, once the positions of the houses had been determined, the software ran thousands of different configurations for the placement of the microgrids, allowing the team to pick the setups that provided the best combination of coverage area and required equipment.
Moving forward, four villages will be selected in India, with two having microgrids installed using the software recommendation and two using existing methods. The cost and performance of the two approaches will then be monitored to determine how beneficial the new system may be.
The team believes that the software has a range of potential applications beyond its initial uses. For example, it could be used on a wider scale to map population movement in India by tracking the position of houses over time.
"We're hoping that public agencies eventually see the wisdom of mapping 100 million rural households in developing countries," said Stewart Craine, chair of the UN Foundation's mapping group. "Preliminary mapping can reduce wasting expensive field-time mapping households."
Source: MIT
