Comparative Study of Deep Learning Techniques in Cattle Identification

Accurate identification of individual livestock is extremely important for proper care and herd disease management. Livestock identification contributes to breed management, productivity monitoring and reduce labour cost through automation. Cattle identification with deep learning using biometric and facial recognition is more accurate, consistent than traditional branding or ear tags which may go missing or get damaged. Due to the enormous amount of data that is sent through quickly, an electromagnetic technique like RFID may be able to marginally speed up the process over hand sorting, but it will never be able to attain the same levels of analysis as Deep Learning. This adds up for behavioural analysis, continuous health monitoring and automated decision-making that will eventually lead to increased yields and herd management. This study compares the capabilities of state-of-the-art deep learning systems and looks ahead at future research in this area. In this study, opencows2020, dataset is used, which comprises of 4243 images of Holestian frestian cattle, that are divided into 46 categories are used to compare the performance of the deep learning models AlexNet and ResNet-50 with the DeiT transformer. The DeiT transformer performed better than the other two versions, according to the study.