“…Conversely, convolutional neural networks (CNNs) can automatically determine complex and highly discriminative features directly from images. Recently developed CNN‐based weed and crop classification systems (Lottes, Behley, Milioto, & Stachniss, ; McCool, Perez, & Upcroft, ; Milioto, Lottes, & Stachniss, ; Mortensen, Dyrmann, Karstoft, Jørgensen, & Gislum, ; Sa et al, ) also integrate background removal (e.g., separating plants from soil) into the classification step and have been reported to achieve excellent classification performance as well as real‐time processing speeds, a strong requirement for the commercial deployment of such systems. One of the remaining obstacles towards the commercial feasibility of weed and crop classification systems is the high annotation effort required to generate the amount of labeled data needed for the training, as well as for adaptation to different environmental and crop conditions (Lottes et al, ; Slaughter, Giles, & Downey, ).…”