“…A number of studies have developed weed detection algorithms based on specific plant features such as colour (Burks et al, 2000;Esau et al, 2018;Woebbecke et al, 1995a), shape (Lee et al, 1999;Woebbecke et al, 1995b), texture (Chang et al, 2012;Tian and Reid, 1999) or a combination of these features (Burgos-Artizzu et al, 2011;Golzarian and Frick, 2011;Kavdir, 2004). Importantly, non-machine learning algorithms typically have lower computational requirements and perform faster on less powerful processors, such as the Raspberry Pi, improving the likelihood of real-time operation in large-scale cropping systems (Chechliński et al, 2019;Tufail et al, 2021). Critically, the computational requirements of the algorithm determine the framerate and hence the real-time capability of the device.…”