The Classification Of Cattle Behaviors Using Deep Learning

Cows have different behaviors that act as indicators of health and well-being. Monitoring the behaviors of cows is an essential component of livestock management since it offers the help needed by farmers to make decisions that contribute to the overall welfare of cattle. Traditional manual monitoring of cows can be costly, inefficient, and subjective to each human observer. In this paper, we propose an approach for the classification of two specific cow behaviors which are Drinking and Grazing. The dataset we used consisted of labelled videos containing 3 classes which are "Drinking", "Grazing", and "Other". The number of videos in each class were 360,413, and 941 videos respectively. We used a CNN-RNN architecture to handle both the temporal and spatial information of the videos. We carried out 2 experiments. Firstly, we inputted the imbalanced data as it is to the model. Due to the imbalanced nature of the classes, we used proper evaluation metrics which were weighted average precision, recall, and F1-Score in addition to the testing accuracy. The results were a testing accuracy of 73.24 %, a precision of 72.72 %, a recall of 73.18%, and an F1-Score of 72.94%. Secondly, we used data augmentation techniques to increase the size of the dataset. We used rotation using different angles combined with horizontal flipping to make the number consistent across all classes. The final data contained 942 videos in each class. This produced a testing accuracy of 84.88%, a precision of 85.92 %, a recall of 84.89%, and an F1-Score of 85.5%.