Farrowing is a critical stage in pork production, however barn workers may not be present for the onset or duration. Use of detection systems may improve health and welfare in this period. Over 1,000 annotated and labelled images were collected from surveillance cameras and used as dataset for training YOLOv5 model variants (artificial intelligence technology). To train the model, three classes of posture and locomotion behavior (standing/walking, lying and sitting) were used as initial indicators of activity level of pigs in a group-housed system. Mean average precision (mAP) and number of parameters employed were used as evaluation metrics to assess the overall performance of the models. The most promising model for monitoring and detection of posture and locomotion behavior was YOLOv5S, which was capable of achieving a mAP of 83% and more efficient than the other model variants. The YOLOv5S model performed poorly in detecting sitting behavior compared to standing/walking and lying behaviors, which could be due to the large disparity in the number of images used to train the mode. Additional datasets with relatively equal proportion of standing/walking, sitting and lying behaviors are being compiled, and different hyperparameter combinations are adjusted to determine the best weights for the final object detection model. In addition, collection of farrowing videos is currently underway to detect the onset of farrowing and identify sows in distress.
