Semantic segmentation of crops from time series of Sentinel-2 imagery (CZE only)
Satellites of the Sentinel-2 mission are currently very valuable data sources for the automation of land coverage mapping. In addition to sufficient spatial resolution, the multispectral images offered by these satellites also have a high temporal resolution in the order of units of days, which is suitable for monitoring phenomena that exhibit temporal dynamics. In the case of semantic segmentation of crop types, when the spectral response of the surface depends on the phenological phase of the crop, it is essential that the temporal information contained in the time series of images is also used to maximize the resulting accuracy. As part of the presentation, we will present U-TAE, the model we used to solve the semantic segmentation of crops, we will further describe the dataset of crop types consisting of time series of Sentinel-2 images over the Czech Republic and summarize the preliminary results.