Manzanilla-Pech et al.: HERITABILITIES USING 3-DIMENSIONAL CAMERAS 9007 netic variation among lactation stages and parities. However, measuring full lactations in commercial data is not feasible, due to its costs and labor involved. In contrast, BW is recorded by different methods such as traditional weighting scales, walk-through scales, and scales attached to the automatic milking systems (e.g., Lely A3). Additionally, BW is also predicted by body size measurements as heart girth (Branton and Salisbury, 1946; Yan et al., 2009); however, there is not a consistent and continuous measurement method for BW, as we see in milk recording. Recently, several technologies have been developed to facilitate precision phenotyping in livestock (Brito et al., 2020; Neethirajan and Kemp, 2021) for welfare, pre- diction, and diagnostic purposes. With these technolo- gies, a new era of precision phenotyping has emerged. One of these technologies is the use of 3D cameras to predict individual feed intake and BW in commercial herds on a large scale. The Cattle Feed Intake System ( CFIT ; Lassen et al., 2018, 2023; Thomasen et al., 2018; Viking Genetics, 2022), which combines the use of 3D camera recordings and artificial intelligence, is an alternative tool for predicting individual DMI and BW compared with scale-based systems. The CFIT 3D cameras, located in barns, can identify individual cows and record individual DMI and BW for the entire herd using artificial intelligence algorithms (Borchersen et al., 2014, 2017; Lassen and Borchersen, 2020). One of the advantages of this new technology is that it allows us to predict feed intake and BW of every cow in the barn during the entire lactation and for all lactations. In comparison to traditional methods (feed bins) to record feed intake, recording with 3D cameras, located on the barn’s roof, does not affect, or limit the feeding behavior of the animal. Furthermore, feed intake and BW are mainly recorded (fully or partially) on the first and second lactation, as most of the research on feed intake and feed efficiency, has been done using only early lactations. This has led to neglecting later (3+) lactations, whereas all commercial farms have cows in later lactations. Finally, this methodology has been previously vali- dated with crossed experiments using traditional meth- ods and 3D cameras, the resulting squared correlation between traditional feed bins and 3D cameras had a value of 0.9 (Lassen et al., 2022). Additionally, Gebr- eyesus et al. (2023) presented prediction accuracies of 0.94 between the contour data (used to predict BW) and the actual BW measurements, with low prediction errors. However, as with every new phenotype, the genetic variation in these 3D traits (DMI and BW) needs to be quantified. This is the first study to report the heritability of 3D camera phenotypes for DMI and BW in dairy cattle. The objective of this study was to estimate the genetic parameters for DMI and BW (ob- tained by 3D cameras) from 6,000 Danish dairy cows from 3 different breeds, Holstein, Jersey, and Nordic Red, in commercial herds. MATERIALS AND METHODS Because this study was based on data collection without handling of animals, no ethical approval was needed. Recording of the 3D Camera Phenotypes The data were collected from 3D cameras installed in 17 commercial farms in Denmark during 2019 to 2021. The cameras were located above the feed bunk (attached to the roof) and the cows were recorded while eating (Lassen et al., 2018; Thomasen et al., 2018). The 3D camera was based on time-of-flight technology (Mi- crosoft Xbox One Kinect v2) to create a 3D image, and the ear tags were read using a radio frequency identifi- cation reader (Agrident Sensor ASR550). A Dell T630 128-GB RAM server was installed in each herd with a 3090 RTX graphics card (NVIDIA) and used for data analysis. An algorithm based on artificial intelligence identifies cows and translates their 3D images into their phenotypes (DMI and BW). Lassen et al. (2018) and Thomasen et al. (2018) presented a description of the 3D camera methodology used to identify cows and mea- sure individual feed intake. Feed intake was assigned after each cow visited the feed bunk. The cows were fed a total mixed ration diet consisting mainly of maize silage, grass silage, and concentrates. The last image of the feed before a cow begins a visit was stored. The cow puts down its head and is identified. The position in the barn and the starting point of the visit are saved. When the cow takes out its head again, the visit is over, and a new image of the feed is available. Differences in the 2 feed surfaces were determined, and the amount of feed removed was saved along with the end time of the visit. The position of the cow’s head was used to distribute feed between 2 cows that could share feed and visits. A cow’s head is in a virtual window and can eat feed from this window as well as from 2 corresponding windows to the left and 2 to the right, from the window in which its head is. If a cow eats alone from these 5 virtual windows, it will have all the feed distributed to her. If a virtual window is shared with another cow, the feed is distributed equally between the 2 cows. Body weight is also predicted using 3D images of the back of the cow while they pass a corridor leaving the milking parlor, from which the contours of the cow’s back were obtained (Lassen et al., 2022; Gebreyesus et al., 2023). The images were standardized in terms of width and Journal of Dairy Science Vol. 106 No. 12, 2023
Download PDF file