PS2-15. Multimodal Video and Audio Signal Processing for Estimating Cattle Forage Intake.

Abstract Accurate quantification of forage intake and rumen digesta during grazing is important for nutrition research but remains difficult to measure at scale. This pilot study evaluated whether contactless video and audio recordings collected during grazing contain kinematic and acoustic signatures that can be related to forage disappearance and rumen digesta dry matter measured via cannulation. Two Angus cattle were recorded across three days during controlled feeding trials in which cereal rye was presented in ground-level pots (three pots per animal per day) and consumed while a fixed-position action camera captured video with onboard audio. Jaw and head keypoints were tracked from video using SLEAP, and grazing-associated audio features were extracted from the video soundtrack using standard digital signal processing implemented in librosa. Features linked to biting events were used as inputs to a linear regression model to estimate forage dry matter disappearance and rumen digesta dry matter recovered after each trial. In this pilot dataset, recovered rumen digesta dry matter was highly correlated with forage disappearance (R² = 0.99), supporting the use of disappearance as a closely aligned intake-related endpoint under these controlled conditions. The results motivate expanded validation across more animals, forage types, and grazing conditions to assess model generalizability and the potential for contactless intake-related monitoring using invasive ground truth for calibration. For image description, please refer to the figure legend and surrounding text. For image description, please refer to the figure legend and surrounding text.