Evaluating cow identification reliability of a camera-based locomotion and body condition scoring system in dairy cows

Dairy farms are growing in size, while the total number of farms is decreasing. As herd sizes grow, technology is increasingly adopted for monitoring and managing cows. However, for technology to replace subjective and labor-intensive tasks, accurate technologies are needed. Camera-based systems are being explored for improving management, particularly around lameness and body condition scoring. A key requirement a camera-based technology is its ability to accurately identify the animal being observed. The primary objective of this study was to evaluate the cow identification (ID) accuracy of a camera-based technology that locomotion scores and body condition scores dairy cattle. Secondary objectives were to determine the number of days required for initial identification and evaluate the frequency of animal recognition. Two dairy sites in Minnesota were visited twice and a total of 105 cows were enrolled from site A (n = 40) and site B (n = 65). Each cow had their cow ID and radio frequency identification recorded. All cows that had been in the pen for 5 or less days were enrolled in the study and had a combination of colors, letters, and numbers painted on their rumps as unique identifiers (paint ID; PID). The video feed from the camera-based technology was observed daily from the company website for 7 d. A cow was recorded as correctly identified if the website had an uploaded video for a cow's ID that matched its PID. Successful identification was defined as the proportion of cow ID for which video was uploaded to the user platform within 7 d from the start of the study, regardless of accuracy. Correct identification was subsequently calculated as the proportion of these successfully identified cows that had a PID corresponding to their PID. The days under the camera were calculated by including the time cows would have been exposed to the camera before our 7-d observation period. Additionally, days identified reflect the total number of days that cows were identified (correctly or incorrectly) and scored during the observation period. Of the 103 cows enrolled, 87 (84.5%; 95% CI: 76%–91%) of cows were successfully identified during the 7-d study period. One cow from those 87 was incorrectly identified, resulting in a correct identification of 98.9% (95% CI: 94%–100%). Of the 86 correctly identified cows, all cows were observed between days 4 and 11 under the camera. Of the cows identified, they were identified 1 to 7 times. This technology accurately identifies cows, but 16 cows were not initially identified and ended with a minimum and maximum of 7 and 11 d under the camera, respectively. To allow management decisions to be made early in lactation or for new cows entering the herd, the technology will need to accurately identify all cows within the first week of being exposed to the camera.