Projektsuche

Federal State

Thuringia

Project Title

Development of an artificial intelligence to distinguish weeds from main crops (Sustainable PhenoAI).

Project description

An open-interface digital ecosystem is to be established that allows the use of plant recognition algorithms to be made more generalizable and scalable. This is to be tested using the example of weed control with the agricultural robot E-TERRY in order to increase the acceptance of the solution and drive forward automation in agriculture by entering the market. The innovation consists of embedding an AI that can recognize crops and weeds with pixel precision in an environment with a radically shortened training cycle.

Final report

Download report (PDF file)

Results focusing on practiceorientation

The Sustainable PhenoAI project developed a sustainable AI system that uses semantic segmentation to distinguish between weeds and main crops.

To develop and train the AI system, a database of real and synthetically generated image files was created and subsequently annotated. In the synthetic generation of image material, GAN shows no measurable improvements over real data, meaning that more realistic methods must be chosen in the future. The synthetic datasets were further expanded through photogrammetry, 3D modeling, and automated image synthesis, making the AI more robust against variations. The AI model for semantic segmentation was built, expanded, and adapted using a Mask2Former architecture. A customized MLOps system was designed to implement the following pipeline: data acquisition → annotation → model training → deployment → visualization. The complete automation of interfaces significantly increased efficiency. The subsequent integration of the p

Main funding source

Rural development 2014-2020 for Operational Groups (in the sense of Art 56 of Reg.1305/2013)

Project status

completed

Total budget applied for in EUR

236,735

Link 1

Link 2

NUTS3 CODE

DEG0D Sömmerda