The work carried out at our chair is concerned with software and systems engineering in the field of digital farming. For this purpose, we consider systems along the entire value chain. Our location in Rhineland-Palatinate offers an excellent basis due to the versatile coverage of different areas, such as wine, vegetable and grain cultivation, but also the immediate proximity to research and industry.

Interested students looking for theses and projects are always welcome to contact us to discuss interests and possible topics.

Our research focus:

• Software and systems engineering in the field of digital farming
• Requirements engineering of the different actors in the agricultural ecosystem
• Improvement of interoperability and networking between actors and systems
• Improvement of the acceptance of digital farming solutions (e.g. FMIS, decision support systems, agricultural machinery)

Forschungsschwerpunkte:

• Untersuchen Sie die aktuellen Methoden zur Erkennung von Pflanzenkrankheiten und ihre Grenzen, insbesondere im Hinblick auf Datenknappheit.

• Schlagen Sie neue Strategien für die effektive Nutzung begrenzter Daten beim Training von Deep-Learning-Modellen zur Erkennung von Pflanzenkrankheiten vor.

• Entwerfen und implementieren Sie ein Deep-Learning-Framework, das auf die Erkennung von Pflanzenkrankheiten zugeschnitten ist und dabei den Schwerpunkt auf Techniken wie Transferlernen, Datenerweiterung und halbüberwachtes Lernen legt.

• Bewerten Sie den vorgeschlagenen Rahmen anhand verschiedener Datensätze zu Pflanzenkrankheiten mit unterschiedlichem Grad an Datenknappheit und vergleichen Sie seine Leistung mit den Basismethoden.

• Analysieren Sie die Wirksamkeit verschiedener im vorgeschlagenen Rahmenwerk eingesetzter Techniken und geben Sie Einblicke in deren Beiträge zur Modellverallgemeinerung und -robustheit.

• Erkunden Sie potenzielle Anwendungen und Auswirkungen des entwickelten Frameworks in realen landwirtschaftlichen Umgebungen und berücksichtigen Sie dabei Faktoren wie Skalierbarkeit, Recheneffizienz und praktische Verwendbarkeit.

Erwartete Beiträge:

• Entwicklung eines neuartigen Deep-Learning-Frameworks zur Erkennung von Pflanzenkrankheiten mit begrenzten Daten unter Einbeziehung innovativer Techniken zur Bewältigung der Herausforderungen der Datenknappheit.

• Empirische Bewertung des vorgeschlagenen Rahmens für verschiedene Datensätze zu Pflanzenkrankheiten, um seine Wirksamkeit und Robustheit im Vergleich zu bestehenden Methoden zu demonstrieren.

• Einblicke in die Wirksamkeit verschiedener Strategien beim Umgang mit begrenzten Daten zur Erkennung von Pflanzenkrankheiten und Bereitstellung von Leitlinien für zukünftige Forschung und praktische Anwendungen in der Landwirtschaft.

Kontakt:

Vishal Sharbidar Mukunda

Overview:

Digital Farming solutions, such as precision agriculture technologies, crop management software, soil irrigation systems, have the potential to increase agricultural productivity, reduce costs, and improve sustainability. However, the adoption of these technologies by farmers has been slow. One reason for this could be missing explainability how these systems come to decisions. The opaque nature of these models limits the farmer's understanding of the technology and their ability to trust the system.

The goal of this study is:

• To identify the key factors that influence farmers' acceptance of Digital Farming solutions.
• To explore the role explainability plays in user acceptance.
• Investigate how explainability is integrated in other domains (e.g. medical, finance).
• Provide recommendations for the development of explainable Digital Farming solutions in agriculture.

Methodologies:

• Online literature review
• Surveys and interviews to measure the relationship between explainability, user understanding, trust, and acceptance

Contact :

Mengisti Berihu

Overview:

• Images in arable farming already enable many use cases, such as monitoring plant growth or early detection of pest infestations in the field. The analysis of imagery can be done in many ways and often relies on AI (e.g. for the detection of yellow rust on wheat).
• In order to plan appropriate actions after analyzing the imagery (e.g. application of pesticides), the detected image content must be mapped to real-world coordinates in order to locate the position of the findings in the image in the field. A suitable file format for storing the images is GeoTIFF, which allows the transformation of image points (pixels) into real coordinates (longitude, latitude).

Problem definition:

• Numerous georeferenced stereo images in PNG file format are available, which were recorded by a tractor on farmland. For each image, the position of the tractor during the recording is known (longitude, latitude).
• A transformation of the image points to real coordinates is to be made possible. It is to be investigated independently, how this can be realized with the available data. Finally, the images in PNG format are to be transformed into GeoTIFF format and enriched with this information.

Research Question:

• How can georeferenced images be subsequently transformed pixel-wise into real coordinates? This question is to be worked out exemplarily on the basis of the available data.

Contact:

Felix Möhrle