Detection of informal settlements in LATAM: Workshop + Dymaxion Community
The analysis of geospatial imagery through machine learning techniques has great potential to contribute to the solution of many major problems that societies face. During July, we launched a workshop where we worked on the detection of informal settlements in Latin America through aplatam-net, a new neural network that we incorporated as one of our open-source algorithms.
Although there are local singularities, the problem of precarious housing in Latin America is present in all the countries in this region. The importance of having a complete and updated map of the location, features and evolution of informal settlements is essential to design and implement policies to revert this situation and improve the quality of life of the population.
Even though informal settlements vary depending on the region, it is possible to determine their location based on certain observable elements from satellite imagery, such as agglomeration and density patterns.
We divided the workshop into three meetings about specific topics in order to focus on every relevant aspect for the detection and to show how a project unfolds, from the beginning to the creation of the final map.
Three meetings to detect informal settlements from scratch
The first meeting was focused on the dataset creation and it covered all the challenges that may arise: obtaining the images and annotations, loading them to the cloud and processing them to work on them later on without difficulty, among others. Since we know that in the data science field more than 80% of the time is devoted to this stage, we incorporated Satproc, one of our open-source tools which streamlines this process.
On the second meeting, with the data ready, we focused on training the model and executing, analyzing and improving the first predictions. To accomplish this, we used imagery segmentation techniques and convolutional neural networks.
Lastly, on the third meeting, we focused on the application of post-processing techniques to the results obtained during the previous meeting. We also reviewed the potential and advantages of executing the model prediction on the Dymaxion Labs cloud, through the Dymaxion Labs Platform.
Dymaxion Labs Community
We had the pleasure of sharing this meeting with professionals from different fields, who want to learn more about these technologies in order to be able to apply them to their professional areas.
Since we believe that collaborating and exchanging ideas is essential to keep developing tools with a positive impact on society, we created the Dymaxion Labs Community on the Slack platform. There, we share all the material and full videos of the meetings. Besides, we want it to be a channel to keep this conversation going.
If you did not participate on the meeting, but you are interested in being part of the Community and accessing the workshops material, contact us at comunidad@dymaxionlabs.com.
