MIKIL - Multiskalige Integration moderner KI-Technologie in dezentralen Landwirtschaftssystemen (NaLamKI-Subproject)
Project duration:
01.01.2021 to 31.12.2023
Project description:
In the NaLamKI joint project, sensor and machine data from remote sensing, soil sensors, robotics, manual data collection and inventory data are aggregated in order to optimise agricultural processes such as irrigation, fertilisation and pest control more sustainably, efficiently and transparently using modern AI methods and (partially) autonomous vehicle and agent networks. NaLamKI creates the possibility of a high-dimensional evaluation and fusion of all aggregated data. GAIA-X compliant services are made available in a decentralised and interoperable manner while preserving the data sovereignty of the actors.
For the University of Hohenheim, the MIKIL sub-project focuses on the integration of modern AI methods at all levels of the NaLamKI system. In addition to the collaborative development of an overarching complex system architecture for a GAIA-X-compliant distributed agricultural system based on determined domain-specific requirements, this includes in particular the AI-based processing, fusion and analysis of the data collected and available in the project at different levels, as well as the federated knowledge exchange between cooperating machines and subsystems.
Scientists involved on site:
- Jonathan Heil, M.Sc.
- Mortesa Hussaini, M.Sc.
- Jun.-Prof. Dr. Anthony Stein
Departments involved on site:
- Fg. 440g Artificial Intelligence in Agricultural Engineering
Consortium of the research network:
- John Deere GmbH & Co. KG (Consortium leader)
- Fraunhofer Heinrich-Hertz-Institut
- Deutsche Forschungszentrum für Künstliche Intelligenz (DFKI) Kaiserslautern
- Technische Universität Kaiserslautern
- Julius-Kühn-Institut
- OptoPrecision GmbH
- Robot Makers GmbH
- Planet Labs Germany GmbH
- NT Neue Technologien AG
- Deutsche Landwirtschafts-Gesellschaft DLG e.V.
- Förderverein Digital Farming (FDF)
- Landwirtschaftliche Lehr- und Versuchsanstalt Hofgut Neumühle
Funding provider:
- Bundesministerium für Wirtschaft und Energie (BMWi)
- Förderkennzeichen: 01MK21003J
More Information:
Projectwebsite: https://nalamki.de/
KINERA - Künstliche Intelligenz für eine effiziente und resiliente Agrartechnik
Project duration:
19.04.21 to 18.04.2024
Project description:
The targeted use of artificial intelligence methods is intended to increase the efficiency and resilience of procedural processes in agricultural technology. In particular, the process of crop establishment is being examined as a real-life experiment using a conventional tractor-implement combination and autonomously with a robot. The aim is to achieve easier operability through setting support or self-adaptation and thus to specifically reduce the performance gap of the machine. The machine should also access information through communication with other machines and the cloud. In the context of machine communication, the potential of swarm robotics is simulated. The reliability of the system is ensured by a three-layer, fault-tolerant information and system architecture with integration of a farm server.
The utilisation of a sowing combination is increased while at the same time the machine operator is relieved by easier operability. This is measurable through the development of a prototype by retrofitting a conventional drill combination and embedding the machine in a developed system architecture. The three-layer system architecture becomes resilient to external disturbances such as network failures through the integration of a farm server at the operating level, and the integration of the cloud level addresses possibilities for process optimisation on the tactical planning horizon as well as the integration of additional information sources. The simulative evaluation of swarm robotics will show possible potentials of scalable robots in the agricultural context.
The Department of Artificial Intelligence in Agricultural Engineering is developing the required AI methods in the project. For this purpose, methods for observing the environment from sensor data as well as methods for the automated adaptation of machine configurations are being developed. In addition, simulative AI methods are being developed specifically for the use of robots and the potential of swarm robotics is being explored.
Scientists involved on site:
- Jonas Boysen, M.Sc.
- Georg Feyrer, M.Sc.
- Jun.-Prof. Dr. Anthony Stein
Departments involved on site:
- Fg. 440g Artificial Intelligence in Agricultural Engineering
- Fg. 440d Process Engineering in Plant Production
- Chair of Business Informatics
Consortium of the research network:
- Exatrek, EXA Computing GmbH
- Smart Site Solutions GmbH
- CLAAS Vertriebsgesellschaft mbH
- Horsch Maschinen GmbH
- VDMA Landtechnik
Funding provider:
- Bundesministerium für Ernährung und Landwirtschaft (BMEL)
- Bundesanstalt für Landwirtschaft und Ernährung (BLE)
NOcsPS - Landwirtschaft 4.0 ohne chemisch-synthetischen Pflanzenschutz / VP25 KI-basierte Hyperspektraldatenanalyse für ein effizientes Pflanzenpathogenmonitoring
Project duration:
01.04.2022 to 30.11.2023
Project description:
A sufficient and high-quality food and biomass supply that is produced in an even more environmentally and nature-friendly manner is a strong socio-political concern. The use of synthetic chemical pesticides (csPSM) is increasingly criticised due to residues in food and nature as well as threats to biodiversity.
This means that an Agriculture 4.0 can be established that follows biological principles using the latest networked technologies, while dispensing with csPSM. At the same time, the use of mineral fertilisers is made possible to ensure soil fertility to produce the required amount of biomass yields.
This approach represents a complete reorientation in arable farming and requires careful accompanying research from all angles and at all scales.
The aim of the research network of the University of Hohenheim (UHOH) and Georg-August University Göttingen (UGOE) and the Julius Kühn Institute (JKI) is to develop, analyse and describe NOcsPS cultivation systems in comparison with other cultivation systems. This comparison is carried out in system, exact and on-farm trials at plot, field, farm and landscape level as well as from an ecological, economic and social perspective.
Our part in this project:
AI-based hyperspectral data analysis for automated and efficient plant pathogen detection
A fundamental change in the application of chemical-synthetic crop protection must be accompanied by an increase in the level of automation and reliability in pest monitoring. The application of alternatives to chemical-synthetic crop protection requires a more detailed knowledge of the prevailing pathogen-induced stress in the field as well as the most precise localisation of the occurrence. Monitoring of this information with high temporal and spatial resolution can be made possible by means of field flights by UAVs (unmanned aerial vehicles, often also "drones") in combination with AI-based data analysis.
Based on novel systematic development and operationalisation principles for data-centric AI or machine learning applications (MLOps), an analysis pipeline is developed for the purpose of automated pathogen detection based on hyperspectral data. First, the existing database and the underlying pathogen monitoring process are analysed manually (WP 1: Process & Data Understanding). Subsequently, necessary data pre-processing steps and procedures for data augmentation are elicited according to the determined characteristics of the existing hyperspectral data and those to be further collected, and the technical (i.e. programmatic) development of the AI pipeline is initiated (WP 2: Data Engineering). This is followed by the informed and systematic selection and configuration of suitable Deep Learning models from the state of the art in machine vision, which are carried out in an iterative model development and validation process (WP 3: Model Engineering). The final part of the outlined work programme is the iterative technical implementation of the AI pipeline, which ensures both continuous systematic data management and, as a consequence, continuous model adaptation to newly available data (WP 4: MLOps).
At the end of the project, a robust and automated pathogen detection pipeline should exist, which can be adapted to varying conditions of use in a data-efficient manner and can reduce future costs for the essential task of pathogen monitoring in integrated crop protection through the use of AI.
Scientists involved on site:
- Georg Feyrer, M.Sc.
- Jun.-Prof. Dr. Anthony Stein
Departments involved on site:
- Fg. 440g Artificial Intelligence in Agricultural Engineering
Consortium of the research network:
Funding provider:
- Bundesministerium für Bildung und Forschung (BMBF)