Assessing the Accuracy and Reliability of the Swine Sense Hub® Camera for Estimating Body Weights of Individual Growing Pigs

Routinely monitoring body weight (BW) in growing pigs provides insights into their growth rate, feed efficiency, and health status. However, manual collection of scale BW requires significant labor. The objective of this study was to evaluate the accuracy and reliability of a novel computer vision-based system (Swine Sense Hub® Camera) able to estimate the BW of pigs in a commercial swine barn. Pen cameras were installed at ceiling height in front of the feeder at a commercial research finishing barn in Indiana. During the period of June to October of 2025, camera-based BW estimates were compared to manually collected scale BW on the same day. The full analysis consisted of two different groups of pigs over six different collection periods. In Turn 1, 443 observations were collected from six pens (17 to 19 pigs per pen) over four different dates, with a mean initial age of 119 days (BW of 82 kg) and a final age of 161 days (BW of 124 kg). In Turn 2, 461 observations were obtained from twelve pens (20 to 21 pigs per pen) on two different dates, with a mean initial age of 84 days (BW of 47 kg) and a final age of 100 days (BW of 62 kg). The agreement between the camera and scale weights was assessed using a repeated-measures concordance correlation coefficient (CCC), where a value of 1 indicates perfect positive agreement and 0 indicates no agreement beyond random chance. In both turns, the agreement between weights was high (CCC > 0.98). In Turn 1, the CCC at the individual pig level was 0.98 (95% confidence interval = 0.975-0.983). In Turn 2, the CCC was of 0.99 (95% confidence interval = 0.986, 0.990). The absolute percent error was calculated as the absolute percent difference between the scale and camera weight, divided by the scale weight. The mean (± standard deviation) absolute percent error was 2.39% (± 2.31%) for Turn 1 and 2.10% (± 1.78%) for Turn 2. Reliability was used to determine the proportion of actual observations captured by the camera compared to the total number of expected observations. The mean reliability of the cameras estimating individual pig BW for each of the days collected indicated that 85% of the expected observations were captured by the cameras during Turn 1 (6 pen cameras) and 86% in Turn 2 (12 cameras). Results from this analysis suggest that the computer vision-based system tested has a strong predictive ability for scale BW (CCC > 0.98, and mean error < 2.39%), as well as relatively high reliability of capturing individual pig BW (>85%) during the test period.