On-farm evaluation of UAV-based aerial imagery for season-long weed monitoring under contrasting management and pedoclimatic conditions in wheat

作者：Jonas Anderegg, Flavian Tschurr, Norbert Kirchgeßner, Simon Treier, Manuel Schmucki, Bernhard Streit, Achim Walter · 发表于：Computers and Electronics in Agriculture · 年份：2022 · DOI：10.1016/j.compag.2022.107558 · 被引用次数：59 · 研究领域：Remote Sensing in Agriculture、Smart Agriculture and AI、Biological Control of Invasive Species

Timely availability of weed infestation maps is a key prerequisite for the implementation of site-specific weed management practices. Low-altitude aerial imagery obtained from unmanned aerial vehicles (UAVs) has shown significant potential for weed detection in crops. However, most studies focused on wide-spaced row crops such as maize and sunflower and evaluated proposed methods on single or few well-characterized experimental sites at specific points in time representing a limited range of application scenarios. This study evaluated the feasibility of weed detection in on-farm wheat fields characterized by a narrow row spacing, throughout the early and late developmental stages using UAV imagery and ground-based high-resolution imagery. Image data was obtained for nine sites, representing a wide range of management and pedoclimatic conditions. These sites can be seen as a representative sample of scenarios that would be encountered in practice. A high within- and across-site as well as temporal variation was observed for weed infestation levels and weed population species composition, highlighting the need for spatially and temporally resolved weed mapping. Image-based classification of vegetation objects as crop or weed plants was achieved with an accuracy of 0.88 and 0.72 in ground-based high-resolution images and UAV-based aerial images captured from an altitude of 10 m, respectively. The accuracy of pixel-wise, vegetation-index-based weed infestation estimation during t...