Automated computer vision analysis of pen-level feeder occupancy in post-weaning piglets under two pen configurations and by sex

The nursery phase is a challenging period in swine production, and automated monitoring of feeder use may provide additional information about piglet activity during early post-weaning. A YOLOv8n detector trained using whole-frame bounding boxes was used to classify video frames as at_feeder or not_at_feeder. These classifications were used to quantify pen-level feeder occupancy, defined as the percentage of analyzed frames in which at least one piglet had its head inserted into the feeder. Eighty PIC piglets weaned at 21 d were housed in 16 single-sex pens. Eight pens contained four piglets (0.39 m²/pig; 18.75 cm feeder length/pig) and eight contained six piglets (0.26 m²/pig; 12.5 cm feeder length/pig), resulting in two pen configurations that differed simultaneously in group size, floor-space allowance, and feeder-space allowance. Sex and pen configuration formed a 2 × 2 factorial arrangement with four pens per cell, and the 16 pens were distributed across four spatial blocks. Video was analyzed during three 24-h windows beginning at 0900 h on Day 1, the day of weaning, and on Days 4 and 7. On a balanced 2,000-frame within-recording validation set, the model achieved mAP@0.5 = 0.995 and an F1-score of 0.993 for both classes. A mixed model with spatial block as a fixed effect and pen as a random intercept identified an observation-window effect (P < 0.001) and a pen configuration × window interaction (P = 0.014). Sex and interactions involving sex were not significant. Occupancy was higher in six-piglet pens during the later observation windows. An illustrative sensitivity analysis showed that the later differences between pen configurations were compatible with the expected mathematical effect of group size on a presence-based occupancy measure. Average daily gain and feed disappearance did not differ by sex, pen configuration, or their interaction. Feed conversion ratio showed a sex × pen-configuration interaction (P = 0.010). In a secondary within-pen analysis, each additional hour of feeder occupancy was associated with 0.023 kg/pig greater feed disappearance over the temporally matched 24-h interval (95% confidence interval, 0.010 to 0.037 kg/pig; P = 0.003). Automated video analysis provided a high-throughput measure of pen-level feeder occupancy across the early post-weaning period, establishing a foundation for subsequent development of individual-level feeding measures.