PRACTICAL WISION – Automatic identification and mapping of weeds

Image analysis has revolutionized research fields from medicine to nature conservation. Weed research is an area where image analysis has both create potential but also many challanges. Identifying small green weeds among many similar small crop plants is not easy — especially when doing so from a vehicle moving over uneven terrain and large field areas. Recently, however, significant advances have been made in methods for identifying different weed species using artificial intelligence (AI)-based image analysis.

Automatic weed identification could enable species-specific weed control, reducing both the amount of pesticides, such as glyphosate/Round-Up, and the need for tillage. Furthermore, it could help weed researchers study how the weed flora is affected by different practices, such as leys and cover crops, reduced plowing, the creation of flower strips, and other methods associated with environmental farming practices. In this way, it would be possible to design cropping systems that require fewer pesticides and less tillage to manage weeds — creating better conditions for preserving soil health and biodiversity. It could even allow for the design of cropping systems that retain some weeds to support biodiversity. After all, weeds are not just weeds; they are plants that can provide food for pollinators, serve as hosts for beneficial natural enemies of pests, increase soil carbon storage, and reduce nutrient runoff. However, to keep weeds at a level where they don’t cause significant yield losses — without the control measures themselves harming surrounding ecosystems — we need to know where the weeds are in the field and which species are present. This knowledge is essential for implementing targeted and resource-efficient management strategies.

AI-based image analysis for automatic weed identification works well in theory, but for it to be useful in weed research (and eventually in agriculture), it needs to be tested and applied in practice. PRACTICAL WISION is a collaboration between RISE Research Institutes of Sweden, the Swedish University of Agricultural Sciences (SLU), Hushållningssällskapet Skåne, and NBR Nordic Beet Research. Together, we will: (1) evaluate how well AI-based image analysis can identify and map different weed species in Swedish field trials, (2) build an iterative process for collecting high-quality, annotated weed images from Swedish field trials (to be published as open-access) — leading to continuously improving weed recognition, and (3) further develop AI models to make them more suitable for field trial applications. This is the first step toward bringing AI-based image analysis into practical weed research — a crucial step for developing more sustainable cropping systems that secure future food supply while promoting ecosystem and biodiversity conservation.

PRACTICAL WISION is funded by FORMAS.