Labels for land-cover classification in livestock farms
This dataset contains a set of georeferenced labels used in the LIVE4ENV project for training and evaluating an AI model for land-cover classification in livestock farms. It includes 3 classes:
- 0: non-productive areas
- 1: woodland
- 2: pasture
The aim of the LIVE4ENV project is to reduce the environmental impact of livestock farming by leveraging IoT devices, Artificial Intelligence and Earth Observation. One of the first steps towards the completion of this goal, has been the implementation of an AI algorithm for land-cover classification in livestock farms.
This dataset contains a set of georeferenced labels used for training this algorithm. The input data for this algorithm have been multitemporal Sentinel-2 images, specifically the annual time series of cloudless Sentinel-2 images in year 2022. The input variables for the algorithm have been derived from these time series, extracting statistical features from it (mean, median, standard deviation, maximum, minimum, etc.).
These labels have been manually annotated using high-resolution satellite imagery. They have been collected from farms with diverse agroclimatic conditions around Europe.