Pre-harvest weed mapping of Cirsium arvense in wheat and barley with off-the-shelf UAVs

• Published in: Precision agriculture Vol. 20; no. 5; pp. 983 - 999
• Main Authors: Rasmussen, J., Nielsen, J., Streibig, J. C., Jensen, J. E., Pedersen, K. S., Olsen, S. I.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01-10-2019; Springer Nature B.V
• Subjects: Agriculture; Altitude; Atmospheric Sciences; Barley; Biomedical and Life Sciences; Cameras; Cereals; Chemistry and Earth Sciences; Cirsium arvense; Computer Science; Environmental conditions; Flight; Flight altitude; Glyphosate; Harvesting; Image detection; Life Sciences; Mapping; Patches (structures); Physics; Plant species; Remote Sensing/Photogrammetry; Soil Science & Conservation; Spraying; Statistics for Engineering; Stubble; Vegetation; Visual flight; Visual thresholds; Weeds; Wheat; Weed detection; Image analysis; Perennial weeds; Cereals; Site-specific weed management
• ISSN: 1385-2256; 1573-1618
• DOI: 10.1007/s11119-018-09625-7

The paper describes a procedure to detect green weeds in pre-harvest cereals using images from off-the-shelf UAVs with RGB cameras. All images used to develop and test the detection procedure were from fields infested with
Cirsium arvense
and in consequence, the procedure is called Thistle Tool. Thistle tool, however, may also detect other green weeds and may also be useful in post-harvest stubble.
C. arvense
maps may be used for pre-harvest glyphosate spraying in countries allowing this practice or used post-harvest or in the following year because
C. arvense
patches are relatively stable from one year to the next. The detection procedure exclusively used colour analysis and discriminated green and senescent vegetation without the ability to discriminate between green plant species. Thistle Tool divides images into patches of 1 m
2
irrespective of the flight altitude, and calculates a classifier called TopMaxExG used for visual threshold editing. Patches are classified into two categories: with or without green vegetation. When
C. arvense
was the main contributor to green vegetation in pre-harvest cereals, 92–97% patches were classified correctly under varying environmental conditions with different consumer-grade RGB cameras. With small consumer UAVs, such as Phantom 3 or 4, it is possible to map 10 ha in 20 min at 40 m flight altitude, which corresponds to the duration of one battery. This study has demonstrated that UAV imagery is practically manageable for
C. arvense
mapping. Barriers to reach the end user and pros-and-cons of using more advance weed detection algorithms are discussed.