Epsilon Archive for Student Projects

Abstract

The aim of this thesis was to evaluate the use of a 3D camera as a tool to estimate body weight and growth in dairy heifers. Data collection lasted from October 2022 to January 2023 and was performed at the Swedish Livestock Research Centre in Uppsala, Sweden. Data collection included a total of 165 dairy heifers of two breeds: 96 Swedish Red and 69 Swedish Holstein. Body weight, 3D images and a set of nine different body measurements: body length, chest girth, hip width, backside width, ischial width, hip ischial width, withers height, hip height and external width between the hip joints, were collected at six different data collection occasions. All heifers with a full set of manual body measurements and BWs from the scale (n=46) were used in the statistical analysis.
Pearson correlations were used to investigate the relationship between each body measurement and body weight. The highest correlation was found between body weight and chest girth (r = 0.94). The correlation between the body weight and the external hip width (r = 0.91), hip ischial width (r = 0.82) and hip height (r = 0.79) were also among the highest. Body measurements with a correlation ≥ 0.75 (external hip width, hip ischial width, hip width, backside width, hip height, chest girth) were used in the model development together with Point cloud images collected by the 3D camera. Three models, based on data from the 46 heifers with a full data set, were created to predict body weight: 1) a regression model using the manual body measurements as input, 2) a regression model based on the manual body measurements together with the Point cloud image data, 3) a machine learning conventional neural network using the Point cloud image data as model input. The performance of the prediction models were assessed using R2 and root mean square error (RMSE).
Model 1 showed the best performance among the three models (R2 = 0.81, RMSE = 17.04 kg). Combining the image data with the body measurements (Model 2) did not improve the model, in fact, lowered the R2 value (0.41) and increased RMSE (27.13 kg). Model 3 was slightly better than Model 2 with an R2 value of 0.53 (RMSE = 22.77 kg). Despite the small dataset, the results show potential in creating a model extracting the body measurements from the Point cloud image data rather than only using the point cloud image information. However, not possible to extract body features partly due to the distance between the camera and the heifer, especially for younger heifers with not yet pronounced body features. Several of the previously described and commonly used body measurements were shown to be useful in estimating body weight. Furthermore, the hip ischial width, not described previously, showed a considerably high correlation to body weight and could thus be used in future automatic feature extraction using 3D imaging technology.