Better-Weeds – Next generation weed management

Abstract

Digitalisation in agriculture is currently expanding very rapidly. Digital field maps, geo-referenced machine movements, and a vast set of sensors on different platforms are already utilised on arable farms and even robotic field management is tested under practical field conditions. Digitalisation and artificial intelligence enable more effective and precise weed management through automated weed recognition. Sensor-based recognition systems automatically differentiate between weed and crop plants allowing for selective and site-specific weed control measures. By reducing herbicide inputs, site-specific herbicide application can mitigate negative impacts on the environment and reduce production costs. With support from the Federal Ministry of Food and Agriculture (BMEL), the project “Better-Weeds” uses the technical possibilities of GIS-based imagery from unmanned aerial vehicles (UAV) and artificial intelligence (AI)-driven plant recognition on arable fields. The novel and innovative part of the project is a knowledge-based merging process. It combines the location data on the spatial distribution of different weed species with biological and ecological weed traits, as well as soil and climatic data to create extensive field maps. In addition, farm-specific agronomic conditions such as the crop rotation on the respective field and available weed control technology are considered. Together, this combined information is used to create a site-specific weed management plan for a given arable field taking into account weed control thresholds and the competitive ability and ecological benefits (e.g. habitat for beneficial insects) of the present weed species. The application of the generated management plan forms a step towards reducing reduce herbicide applications and increasing in-field weed diversity.

Keywords

• automated flying
• hotspot analysis
• image recognition
• knowledge-based maps
• machine learning
• mission planning
• site specific weed management