Application of depth sensor to estimate body mass and morphometric assessment in Nellore heifers

摘要

The potential of an RGB-D sensor as a tool to estimate Nellore heifers' body mass and morphometric measurements through image analysis was evaluated. An Intel RealSense D435i depth sensor was used to acquire dorsal images of 260 animals aged between 8 and 18 months. Images were acquired from March to December 2019 in six-time points. The best quality images were selected using a multi-layer perceptmn neural network (nn = 547). The images were then manually associated with each animal's electronic ear tag. Morphometric measurements were manually acquired using a hipometer and measured on the images using the OpenCV library's graphical interface. These values were acquired in pixels converted to meters. The adjusted linear regression analysis between body mass measured with a scale and estimated body volume presented a high coefficient of determination of R-2 = 0.97. The mean absolute percentage error was 3.13%, the absolute error was +/- 8.85 kg, and the mean squared error was 10.07 kg. The mean absolute error, mean squared error and mean absolute percentage error between manually acquired and digitally acquired morphometric measurements were: 4.23 cm, 5.34 cm, and 18% for chest width (R-2 = 0.56); 4.4 cm, 5.1 cm, and 13.9% for croup width (R-2 = 0.86); 6.0 cm, 8.0 cm, and 19.3% for croup length (R-2 = 0.75); 4.7 cm, 6.6 cm, and 3.8% for croup height (R-2 = 0.9); and 3.5 cm, 5.1 cm, and 2.9% for withers height (R-2 = 0.92). This study showed that it is possible to estimate body mass in Nellore heifers using a depth sensor and has good potential for application on morphometric evaluations.