Real time identification of individual dairy cows through computer vision

In livestock management, computer vision enables automated systems to interpret and understand visual information in animal environments. By drawing on methodologies from image processing and machine learning, computer vision facilitates the analysis, interpretation, and extraction of valuable information from visual data. This paper explores the application of a deep learning-based computer vision system for the identification of individual dairy cows. The initial phase involves training a YOLO neural network, focusing on recognizing cows based on unique features like coat patterns. Subsequently, this trained network is applied to a dataset collected in a different position of the barn to verify the capability of network to recognize the individual cow in different contexts. Performance evaluation of the network involves the use of precision-recall curves, coupled with the application of data augmentation techniques to enrich the dataset and optimize detection efficiency. This process provides insights into the system's generalizability and robustness across various contexts within the agricultural landscape. The findings not only contribute to advancing precision livestock farming but also illuminate the adaptability of the developed model across different farm locations and scenarios, particularly in recognizing cows collected in various positions. The application of the technology tested here can provide real-time recognition and the results of the paper shows the valuable insights into the monitoring of behaviours and movements of individual cows in a barn.