P358 - 45. GIL-Jahrestagung 2025 - Fokus: Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft
Autor*innen mit den meisten Dokumenten
Auflistung nach:
Neueste Veröffentlichungen
- Konferenzbeitrag45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft - Komplettband(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025)
- KonferenzbeitragDrinking event detection of dairy cows using deep learning(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Mukunda, Vishal Sharbidar; Jagadale, Anish Bipin; Girmay, Mengisti Berihu; Möhrle, Felix; Burkhardt, Franziska Katharina; Hayer, Jason Jeremia; Doerr, Joerg; Steinhoff-Wagner, JuliaWater is crucial for dairy cows, making up 55-70% of their body weight and 85% of milk [Be12]. Restricted access to water affects health, welfare, milk quality and quantity, so adequate water intake is essential. Dairy cows’ drinking behavior is influenced by trough design and cleanliness, making monitoring important but tedious [Bu22]. This study introduces a deep-learning approach to detect the drinking event and monitor the total duration of drinking. The approach is divided into cow detection, identification, drinking event detection and tracking the total duration of the cow’s drinking. Various You Only Look Once (YOLO) models were used for cow detection, ResNet-18 and ResNet-50 for identification, and Deep SORT with OCR for detecting and tracking the drinking event. Various YOLO versions and ResNet models were compared for performance. The approach achieved 98% precision in cow detection, 98% accuracy in identification, and 95% accuracy in detecting the duration of drinking, with a 97% F1 Score, ensuring reliable monitoring of dairy cows’ health through their drinking behavior.
- KonferenzbeitragControlled vocabularies in metadata(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Martini, Daniel; Turbati, Andrea; Subirats, ImmaInnovation in research and development relies on findability of information on prior work. For that purpose, FAO maintains AGRIS, an aggregator system making publication and dataset metadata distributed all across the globe searchable. Keyword metadata are a crucial factor for navigation and information retrieval in such systems. Thesauri like AGROVOC provide a means for standardized keyword assignment following best practices of subject indexing. A metadata analysis has been conducted with the objective of identifying common usage patterns and determining whether concept coverage in AGROVOC is adequate for annotation and indexing in AGRIS.
- KonferenzbeitragThe impact of expected data transparency, misuse, and ownership on the perceived ease of use of AI-camera systems in animal husbandry(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Kühnemund, Alexander; Recke, GuidoThis study investigates factors influencing the perceived ease of use (PEOU) of AI-camera systems among German pig farmers. AI-based surveillance systems support tasks such as animal detection, tracking, behavior analysis, and disease diagnosis, but their adoption is hindered by concerns over data privacy and usability. Using the Technology Acceptance Model (TAM) as a foundation, the study explores three factors: perceived risk of data abuse (RI), perceived property rights of data (PR), and perceived transparency (TR). Survey data from 185 pig farmers were analyzed using partial least squares structural equation modeling (PLS-SEM). Results indicate that TR significantly enhances PEOU, while RI negatively impacts it, aligning with prior studies linking trust and usability. Higher PR also boosts PEOU, suggesting that clearer data ownership rights could improve AI adoption. These findings highlight the importance of transparent systems and defined data ownership to foster AI integration in agriculture.
- KonferenzbeitragRecognition of phenological development stages of apple blossoms using computer vision(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Nguyen, Xuan Khanh; Braun, Bastian; Heider, Nico; Schieck, MartinDeep learning-based computer vision is increasingly supporting precision agriculture in orchards, reducing reliance on manual monitoring by trained specialists. This work presents an approach for automated monitoring of apple blossom growth stages, an important task for optimizing yield and quality in orchard management. We (1) construct an annotated dataset of hourly images capturing apple blossoms across BBCH stages 53 to 71, (2) develop convolutional neural networks (CNNs) for growth stage classification, and (3) validate model performance using explainable AI (XAI) to ensure interpretability. Our best-performing model achieves a classification accuracy of 93.1%, demonstrating strong potential for integration into Farm Management Information Systems for data-driven orchard management. Model interpretability analysis further reveals that, with adequate training data, the network predominantly relies on features within the blossom itself to inform predictions, suggesting robustness in real-world application scenarios.
- KonferenzbeitragMultistage Eartag Detection – Entwicklung eines KI-basierten Ansatzes zur automatischen Einzeltier- identifikation mittels Ohrmarken beim Schwein(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Wutke, Martin; Debiasi, Damiano; Höne, Ulrike; Probst, Jeanette; Dirksen, Neele; Lieboldt, Marc-Alexander; Kemper, Nicole; Traulsen, ImkeObwohl bildbasierte KI-Methoden wie die Lokalisation und die Erkennung von Einzeltieren mittlerweile verstärkt Anwendung im Bereich der Nutztierhaltung finden, stellt die Tieridentifikation, vor allem in Bereichen wie der Schweinehaltung mit homogenen visuellen Tiermerkmalen, nach wie vor große Herausforderungen an bestehende Systeme. Das Ziel der vorliegenden Studie ist diesbezüglich die Entwicklung einer optischen Methode zur Lokalisation und Identifikation von Einzeltieren im Kontext der Schweinehaltung. Durch die modulare Verwendung von vier State-of-the-Art Objektdetektionsmodellen werden die vorhandenen Bildinformationen stufenweise analysiert und lesbare Tier-IDs anhand handelsüblicher Ohrmarken automatisiert bestimmt. Im Ergebnis erreicht der vorgestellte Ansatz eine Erkennungsgenauigkeit von 0,993 bei einer Fehlerrate von 0,007. Im Rahmen weiterer Untersuchungen soll die entwickelte Methode zur Verbesserung von Trackingansätzen und im Kontext spezifischer Anwendungsfälle implementiert werden.
- KonferenzbeitragMöglichkeiten und Voraussetzungen zum Einsatz digital-technischer Systeme in der Pferdehaltung(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Speidel, Linda Thurid; Winter, DirkDie fortschreitende Digitalisierung und Mechanisierung lassen weitreichende Veränderungen in der Tierhaltung erwarten. Dennoch werden die meisten Arbeitsprozesse in der Pferdehaltung noch manuell durchgeführt und die Voraussetzungen zum Einsatz der digital-technischen Systeme in Pferdebetrieben sind bisher noch wenig erforscht. Daher wurden verschiedene digital-technische Anwendungen in Pferdebetrieben in den Bereichen Entmistung, Fütterung, Kommunikation sowie Gesundheits- und Sicherheitsüberwachung und das Tierwohl mittels Onlinebefragung bei 451 Akteuren untersucht. Die Ergebnisse zeigen, dass die Nutzung digitaler Systeme maßgeblich von der Haltungsform (Gruppen- oder Einzelhaltung), einer stabilen Internetverbindung sowie der Investitionsbereitschaft der Betriebsleitenden beeinflusst wird. Es ist von essenzieller Bedeutung zur Etablierung digital gesteuerter Technik auf den Pferdebetrieben, den Nutzen der Systeme für alle Beteiligten – Pferde sowie Betriebsleiter und Mitarbeiter – deutlich herauszustellen und zu kommunizieren hinsichtlich ökonomischer Vorteile, zur Verbesserung des Tierwohls und zur Optimierung der Gesundheit der Pferde.
- KonferenzbeitragSingle-image-based georeferencing for unmanned aerial vehicles(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Stoop, Ralph L.; Sax, Markus; Seatovic, Dejan; Anken, ThomasGeoreferencing is important for many applications in precision farming, in particular those based on unmanned aerial vehicles (UAVs). In this context, georeferencing typically relates the optical features of UAV images to their actual position in the 3D world, creating a grid map of the area of interest. Although state-of-the-art georeferencing methods are very accurate, these methods rely on multiple-view geometry reconstruction, which requires largely overlapping images of high quality. Acquiring such images can be difficult in practice, given the low-cost requirements for precision farming. In this paper, we study the practical applications and challenges of a simple, computationally inexpensive and fast method for georeferencing that is solely based on single images. Our method only uses an affine transformation where the UAV’s height is adjusted by a digital terrain model and does not require overlapping images. We find that our single-image-based method can be used for smart farming applications, where spatial accuracies of around 25 cm are sufficient.
- KonferenzbeitragDeep-learning-based quantification of sow activity during farrowing(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Wahmhoff, Johann; Wutke, Martin; Traulsen, ImkeThe activity of sows after farrowing significantly influences the occurrence of crushing losses. For this reason, sow activity is often quantified in scientific research through video analysis. To improve practicality, this study developed an automated approach based on the YOLOv10 object detection framework, addressing the big data challenge. The core idea of this approach was to calculate an activity score based on the number of postural changes. Therefore, the model detects the sow’s posture for each frame in the videos. The raw data from the model is then filtered, and the number of postural changes is calculated, from which an activity score is derived for each video. For comparison with a human-generated activity assessment, the videos were finally categorized into three classes based on the activity score. The final performance evaluation across activity classes shows an accuracy of 0.945. Based on these results, it can be concluded that this approach provides an automated alternative to manual video analysis for determining sow activity.
- KonferenzbeitragMonocular ground surface estimation for precision farming(45. GIL-Jahrestagung, Digitale Infrastrukturen für eine nachhaltige Land-, Forst- und Ernährungswirtschaft, 2025) Stollmeier, Frank; Will, Jens Christian; Homann, HannoIn precision farming, detecting the ground surface is crucial for optimizing different processes, such as sowing, weeding, and growth monitoring. This research introduces a method to estimate the ground surface using a single monocular RGB camera, which is both cost-effective and simple. The camera captures overlapping images as it moves along the field, and feature points from these images are matched to create a 3D point cloud. Points are classified as ground or plant using the Triangular Greenness Index (TGI), allowing the filtering of plants to isolate the ground. The proposed method is validated quantitatively in comparison with a stereo camera.