Extracting Agricultural Fields from Remote Sensing Imagery Using Graph-Based Growing Contours

Abstract

Knowledge of the location and extent of agricultural fields is required for many applications, including agricultural statistics, environmental monitoring, and administrative policies. Furthermore, many mapping applications, such as object-based classification, crop type distinction, or large-scale yield prediction benefit significantly from the accurate delineation of fields. Still, most existing field maps and observation systems rely on historic administrative maps or labor-intensive field campaigns. These are often expensive to maintain and quickly become outdated, especially in regions of frequently changing agricultural patterns. However, exploiting openly available remote sensing imagery (e.g., from the European Union’s Copernicus programme) may allow for frequent and efficient field mapping with minimal human interaction. We present a new approach to extracting agricultural fields at the sub-pixel level. It consists of boundary detection and a field polygon extraction step based on a newly developed, modified version of the growing snakes active contours model we refer to as graph-based growing contours. This technique is capable of extracting complex networks of boundaries present in agricultural landscapes, and is largely automatic with little supervision required. The whole detection and extraction process is designed to work independently of sensor type, resolution, or wavelength. As a test case, we applied the method to two regions of interest in a study area in the northern Germany using multi-temporal Sentinel-2 imagery. Extracted fields were compared visually and quantitatively to ground reference data. The technique proved reliable in producing polygons closely matching reference data, both in terms of boundary location and statistical proxies such as median field size and total acreage.

Deep Learning and Adaptive Graph-Based Growing Contours for Agricultural Field Extraction

“…We compare the default version of the GGC method as presented in our previous publication and the new adaptive version [25]. Settings were selected through manual testing of different settings and visual interpretation of results.…”

“…For field polygon extraction, we employ an active contours algorithm based on growing contours (or snakes) that we presented recently [25,44]. Active contours in general were introduced for object delineation and are based on energy minimization [45].…”

“…Following the concept of angular dispersion introduced by Gregson and the fractional anisotropy described in Aganj et al, we calculated a modified angular dispersion metric similar to the local anisotropy we used for seed point selection in our previous study [25,36,37]. The concept is based on the observation that gradient direction tends to be homogeneous along edges or boundaries in an image while they are heterogeneous ("dispersing") if no clear boundary is present or boundary directions are changing, for example, at crossings of multiple edges or along sharp curves.…”

“…Following our previous methodology, we therefore transform the contour output of the GGC method into a binary edge map [25]. We use a flood fill algorithm initiated on local maxima of a Euclidian distance transform from the binary boundaries to create a segment representing the rough extent of a field as described by the extracted contours.…”

“…In a previous publication, we presented a full workflow combining agricultural field boundary detection and polygon extraction steps [25]. Starting from image pre-processing and enhancement, we combined edge detection, contour detection, and, finally, field polygon extraction.…”