Artificial intelligence recognizes the type of land use
A master's student in environmental technology at the Swiss Federal Institute of Technology in Lausanne has used artificial intelligence to advance the partially automated process of classifying land use. This reduces a lot of manual work.
Thanks to a student at the Swiss Federal Institute of Technology in Lausanne ( EPFL ), the regularly necessary classification of land use is now much less time-consuming than before. According to a press release from the university, she has developed and trained her own machine learning algorithm that can not only distinguish forests from other types of land. Instead, Valérie Zermatten's algorithm also recognizes rivers, lakes, camping and sports fields, cemeteries, water treatment stations, public parks, airports and dams. This makes it clearly superior to the algorithm developed by the Federal Statistical Office (FSO) called Areal Statistics Deep Learning, or ADELE for short.
The results produced by their program as part of a master’s thesis are similar to the official data published by the FSO. According to the announcement, this suggests that it could be used for land use classification in the future. The big advantage lies in the processing time for aerial photos, because their classification into around 40 different categories is still largely done by hand.
All of Switzerland is photographed from the air every three years. Because manual categorization takes so long, the results are only published every six years. With this mapping, land consumption can be better tracked, soil permeability can be monitored and urban sprawl can be combated.
“Our goal is not to replace humans with artificial intelligence,” says Devis Tuia, one of Zermatt's doctoral supervisors at EPFL. “Although Valérie's algorithm will reduce the amount of tedious work that has to be done manually.” But even then there is still enough for people to do – for example, to recognize whether it is a house or a school, a football field or a football field Wiese act.