Book chapter

Extraction of agricultural field boundaries from remotly sensed images
Authors: Rydberg, A., Borgefors, G.
Book title: Geospatial Pattern Recognition
Editors: Binaghi, E., Brivio. P.A., Serpico, S.B.
Publisher: Research Signpost, Trivandrum, Kerala, India, pp. 57-86, 2002
Abstract: We describe a system for automated detection of agricultural fields, which integrates new and existing methods for segmentation and where knowledge about agricultural fields is incorporated. The difficulties associated with different image resolutions and spectral characteristics are also investigated. We thus propose a novel multispectral edge- and line detection method, which combines the results from oriented filters. The detected boundaries are combined with information generated by a region oriented segmentation method, which produces an over-segmentation. Known shape properties of fields are used for merging. Evaluation, including comparison with interpreted ground truth, show that almost 80% of the boundaries are detected correctly, but that the results may vary significantly with image properties.