Published in October 2021
A study carried out by researchers from the Department of Computer Science (DCC) at UFMG, PUC-RJ and USP, all Brazilian institutions, which addresses the challenges of multitemporal recognition of cultures and proposes a new approach called OpenPCS++ capable of not only learning known classes , but also identifies new crops in the forecasting phase, was published in IEEE Xplore, in early October 2021.
According to scientists, multi-temporal remote sensing images play a fundamental role as a source of information for automated crop mapping and monitoring. “The evolution of the spatial/spectral pattern over time provides information about crop dynamics and is very useful for estimating productivity. Although the multitemporal mapping of crops has progressed considerably, with the advent of deep learning in recent years, the classification models obtained still have limitations when exposed to unknown classes in the prediction phase, reducing their usefulness.
Therefore, these models are trained to identify a closed set of crops (e.g., soybeans and sugarcane) and are therefore unable to recognize other types of crops (e.g., corn). “In this work titled “Open Set Semantic Segmentation for Multitemporal Crop Recognition”, we deal with the challenges of multitemporal crop recognition by proposing a new approach called OpenPCS++, which is not only capable of learning known classes, but is also capable of identifying new crops in the prediction phase. The proposed approach was evaluated on two challenging public datasets located in tropical climates in Brazil. The results showed that OpenPCS++ achieved increases of up to 0.19 in terms of area under the receiver operating characteristic curve (ROC) compared to baselines,” they explained.
The article is authored by researchers Jorge A. Chamorro Martinez and Raul Queiroz Feitosa, from the Department of Electrical Engineering at the Pontifical Catholic University of Rio de Janeiro, Brazil, Hugo Oliveira, from the Institute of Mathematics and Statistics (IME), at the University of São Paulo. Paulo (USP), Brazil; Jefersson A. dos Santos, from DCC/UFMG, Belo Horizonte, Brazil.
To learn more, the code is available at https://github.com/DiMorten/osss-mcr.