313 Estimating Body Condition Score of Dairy Cows from Color Depth Images Using a Mobile Device

Abstract Body condition score (BCS) is a commonly used tool to monitor body tissue mobilization in dairy cows and manage potential health issues associated with negative energy balance. However, BCS is a subjective measurement that requires a trained evaluator and is time-consuming. Consequently, the development of a computer vision system to assess BCS in real-time can have a crucial role to optimize management decisions. Recently, smartphones started to be equipped with depth sensors, which allow the acquisition and analysis of depth images in real-time. In this study, our objective was to estimate BCS of dairy cows using color depth images collected from a mobile device. A total of 276 images were collected from 31 Holstein dairy cows containing, acquired at the UW-Madison Dairy Cattle Research Center (Madison, WI). The images were collected with an iPhone 13 Pro using the advanced Dual-camera system from a rear view of the animal. The distance between the animal and the evaluator was kept constant. Two trained evaluators were asked to assess BCS of each cow. The score provided by each evaluator was used to train a deep learning algorithm. The analysis was implemented in Python, and Keras and TensorFlow were used to train the deep neural network. A pre-trained network based on the Xception architecture was used, and transfer learning was adopted by extracting the weights trained on Imagenet. A leave-one-out cross-validation strategy was used to split the data into training and testing sets. Our results showed that the model presented an average accuracy of 60% and 40% for color and color depth images, respectively, in the testing set. Our results indicate that it is possible to use a mobile device to develop computer vision tools to predict BCS of dairy cows.