Automatic counting and positioning of slaughter pigs within the pen using a convolutional neural network and video images

• Automatic monitoring of the positioning behaviour slaughter pigs within a pen. We make use convolutional neural networks with linear regression output. The trained models generalize well to unseen data. Pre-trained can be fine-tuned situations data one day Pen fouling is an undesired pigs, which increases labour costs for farmer, worsens hygiene and welfare has negative environmental consequences. Previous research suggests that grower/finisher their pen potential used in early warning systems alert farmer impending event 1–3 days advance. For such system feasible, pigs’ must automated. To this end, we present novel yet relatively simple method, namely network (CNN) single output node. proposed CNN takes partial images pen, corresponding different areas outputs estimated count number image. By inputting three model estimate each area. generally performs when applied from test pens, mean absolute errors less than 1 pig coefficients determination between observed counts above 0.9. In cases where underperforms on fine-tuning by transfer learning applied; show initially underperforming day’s worth set (26 labelled images), after it will produce near-perfect estimates all subsequent same set.