Sensor technology for precision weeding in cereals. Evaluation of a novel convolutional neural network to estimate weed cover, crop cover and soil cover in near-ground red-green-blue images

Abstract

Precision weeding or site-specific weed management (SSWM) take into account the spatial distribution of weeds within fields to avoid unnecessary herbicide use or intensive soil disturbance (and hence energy consumption). The objective of this study was to evaluate a novel machine vision algorithm, called the ‘AI algorithm’ (referring to Artificial Intelligence), intended for post-emergence SSWM in cereals. Our conclusion is that the AI algorithm should be suitable for patch spraying with selective herbicides in small-grain cereals at early growth stages (about two leaves to early tillering). If the intended use is precision weed harrowing, in which also post-harrow images can be used to control the weed harrow intensity, the AI algorithm should be improved by including such images in the training data. Another future goal is to make the algorithm able to distinguish weed species of special interest, for example cleavers (Galium aparine L.).

Sensor technology for precision weeding in cereals. Evaluation of a novel convolutional neural network to estimate weed cover, crop cover and soil cover in near-ground red-green-blue images