Cattle-CLIP: A Multimodal Framework for Dairy Cattle Behaviour Recognition from Video
Cattle-CLIP is a vision-language model that recognizes cattle behavior by aligning images and text representations across time, rather than just looking at single frames. It achieves 96.1% accuracy on six behaviors in supervised settings. The framework uses temporal integration to understand video sequences and employs customized prompts to improve accuracy in real farm environments.