DaNuMa: Data competencies in livestock farming - machine learning for automatic, robust behavioural classification in pigs
Overview
The project DaNuMa: Datenkompetenzen in der Nutztierhaltung -Maschinelles Lernen zur automatischen, robusten Verhaltensklassifikation bei Schweinen (english translation: "Data competencies in livestock farming - machine learning for automatic, robust behavioural classification in pigs"), aims to automatically identify and record behavioral patterns of livestock with the help of computer vision and machine learning models, using pigs as an example. The identified behavioral patterns can then be used for scientific research questions as well as for assistance systems in practical agriculture. DaNuMa aims to strengthen the key data science skills of young livestock scientists. For this, we offer a Summer School in the course of this project.
Project contents
Based on video footage from previous research projects on pig housing systems with a focus on animal welfare and tail biting, a catalogue of methods will be identified that includes the annotation of video images and the implementation of machine learning models for detecting and tracking individual pigs. This will be used to automatically classify specific behaviors and derive behavioral profiles that can be used for ethological research questions, automatic phenotyping and in assistance systems for animal observation.
Machine learning approaches such as image classification, object recognition, segmentation, pose estimation and tracking appear to be promising and are therefore the focus of the work. A critical evaluation and derivation of potentials and limitations for the use cases are also part of this project.
The technical objectives are to track individual pigs as seamlessly as possible over longer periods of time and at the same time to classify individual behaviors such as walking, lying or sleeping, eating or drinking, and social interaction. To our knowledge, there are currently no robust methods available for tracking and re-identification of individual pigs over longer periods of time, e.g., several hours or even days.
In addition, the project includes the provision of research data sets and guidelines for the application of methods, as well as the organization and implementation of a summer school. The innovative character of the project lies in the sustainable implementation of data science methods and the promotion of the competence profiles of future agricultural scientists. This will not only significantly promote future livestock science research, but also the digitalization of livestock farming.
Partners
The participating institutions combine their digital and data-related competencies for robust and real-time capable automatic behavioral classification in pigs based on innovative evaluation methods, use the IT infrastructure of the Göttingen Campus, and are well connected in the specialist community of livestock science through existing networks and cooperations.Behavioural Informatics in Livestock Husbandry
Institut für Tierzucht und Tierhaltung, Christian-Abrechts-Universität zu Kiel
The Behavioural Informatics in Livestock Husbandry team at the Institute of Animal Breeding and Husbandry at Kiel University, headed by Prof. Dr. Imke Traulsen, specializes in issues relating to livestock husbandry, animal welfare and digitalization.
Department für Volkswirtschaftslehre, Professur für Statistik
Georg-August-Universität Göttingen
The Statistics Division of the Department of Economics at the University of Göttingen, headed by Prof. Dr. Thomas Kneib, works in the field of flexible regression models at the interface of statistics and machine learning. It is part of the Campus Institute of Data Science (CIDAS).
Gesellschaft für wissenschaftliche Datenverarbeitung mbH, Göttingen
The GWDG has many years of experience in the integration and processing of various types of research data. One focus is to enable the processing of large amounts of data (big data) by means of dedicated workflows using HPC systems.
Publications
Henrich, J.; Post, C.; Kneib, T.; Yahyapour, R.; Traulsen, I. (2023): Systematische Evaluation der Detektion von Schweinen in Abhängigkeit von Strategien zur Auswahl der Trainingsdaten. DGfZ-Jahrestagung in Halle, 13.-14-09.2023.Henrich, J.; Post, C.; Kneib, T.; Yahyapour, R.; Bingert, S.;Traulsen, I. (2024): Entwicklung eines vielfältigen und anspruchsvollen Benchmark-Datensatzes für die Detektion von Schweinen in Bildern. GIL-Jahrestagung in Hohenheim, 27.-28.02.2024.
Henrich, J.; Post, C.; Kneib, T.; Yahyapour, R.; Traulsen, I. (2024): PigDetect: a diverse and challenging benchmark dataset for the detection of pigs in images. 11th European Conference on Precision Livestock Farming, 09.-12.09.2024.