Abstract:
This research presents a foaling detection system based on a deep learning-based image recognition algorithm for pregnant mares in precision agriculture. The system aims to predict water breaks and critical events before foaling and provide real-time monitoring to improve animal welfare. The experiment was conducted in a stable, using two camera devices capturing thermal images. The dataset of 5,618 infrared images was carefully selected to represent events before and after water breaks. The chosen deep learning model achieved an overall accuracy of 95.40% and an F1-score of 76.77%, indicating its effectiveness. However, challenges were identified, such as misidentifying urine and heat as water breaks. Rule-based corrections were introduced to address this, resulting in an improved F1-score of 79.02%. The foaling detection system's practical applications in precision agriculture include labor-saving benefits for ranchers and enhanced animal health during the foaling process. Furthermore, integration into existing farming practices could lead to timely interventions during horse births, improving overall farm productivity.
Keywords:
deep learning
,
Precision Agriculture
,
Horse farming
,
Agriculture management