Agricultural Field Delineation Using Satellite Imagery

Abstract

Defining the borders of agricultural fields is fundamental for precision agriculture and one of the key parts of the new European Agricultural Policy. The agricultural fields’ boundaries are basic building blocks for monitoring agricultural land in the context of climate change, food production and security. The aim of the field delineation process is to automatically determine the borders of agricultural fields from satellite images. It is based on the similarity of spatial, spectral, and temporal properties of pixels belonging to the same field. The basic method was developed within the NIVA project on data from the Sentinel-2 satellite constellation of the European Space Agency. The u-net based deep neural network predicts three image variables from the satellite image: the segmentation of the field, its boundary, and the distance of the segmented image points to the boundary. From these an image of the boundaries of the fields is constructed, either from a single image or from a time series of images. In the post-processing phase, the image prediction is transformed into vector format, which represents the result of the field delineation process.