Rapidai4Eo: Mono-and Multi-Temporal Deep Learning Models for Updating the Corine land Cover Product

Abstract

In the remote sensing community, Land Use Land Cover (LULC) classification with satellite imagery is a main focus of current research activities. Accurate and appropriate LULC classification, however, continues to be a challenging task. In this paper, we evaluate the performance of multitemporal (monthly time series) compared to mono-temporal (single time step) satellite images for multi-label classification using supervised learning on the RapidAI4EO dataset. As a first step, we trained our CNN model on images at a single time step for multi-label classification, i.e. monotemporal. We incorporated time-series images using a LSTM model to assess whether or not multi-temporal signals from satellites improves CLC classification. The results demonstrate an improvement of approximately 0.89% in classifying satellite imagery on 15 classes using a multi-temporal approach on monthly time series images compared to the monotemporal approach. Using features from multi-temporal or mono-temporal images, this work is a step towards an efficient change detection and land monitoring approach.

Mapping everything, everywhere, all the time : Modeling European land cover using data fusion and machine learning

“…Because such classes can be ambiguous and hard to map by computer programs, they are not often reproduced at scale is among the most popular and widely used European land cover datasets and should not be discarded. As using human cartographers at a finer spatial and temporal scale would be prohibitively difficult, slow, and costly, attempts have been made to automate the production of CLC [29], but usually at a lower thematic resolution, for example for 12 [199] or 14 [15] classes. There have also been other projects that accurately map European land cover at high spatial resolution using different legends that are more optimized for a remote sensing context, such as S2GLC [161], and attempts to specifically map crop types with specialized approaches [48,158] Validation data All maps are wrong [175], and maps produced with machine learning and earth observation data are no exception.…”