Field validation of a proximity-gated AI framework for mounting-based estrus detection in dairy cattle

Information from the abstract

Accurate estrus detection remains challenging because activity-based monitoring systems may misinterpret non-reproductive movements as estrus-related activity. This study hypothesized that mounting-based estrus detection can be improved by conditioning accelerometer-based motion classification on inter-animal proximity, where mounting is physically feasible. We developed and field-tested a proximity-gated AI framework integrating Bluetooth Low Energy (BLE) proximity sensing with tri-axial accelerometer-based motion analysis. Unlike conventional activity-only ML/DL approaches, the proposed framework treats estrus detection as an interaction-dependent behavioral event by first validating close-contact feasibility and then applying motion classification and event-level aggregation. Controlled evaluation using 120 manually annotated mounting events achieved an event-level F1-score of 93.2% with the LSTM model. A retrospective ablation analysis showed that BLE proximity gating reduced false-positive event candidates from 48 to 6 compared with ungated motion-only inference, corresponding to an 87.5% reduction. Precision increased from 70.0% to 94.8%, while recall decreased slightly from 93.3% to 91.7%. The system was subsequently deployed across three smallholder dairy farms using 71 collar-mounted devices. During field deployment, 31 confirmed estrus events were recorded, of which 23 were detected, corresponding to a field-level recall of 74.2%; no false-positive alerts were observed within this limited dataset. Among 13 system-detected cows that underwent artificial insemination, 11 did not return to estrus during the first expected post-insemination cycle follow-up, giving a first-cycle non-return proportion of 84.6% (11/13), used as a field proxy rather than confirmed pregnancy. These findings support the hypothesis that proximity-gated behavioral validation improves the biological specificity of mounting-based estrus detection, although broader multi-site and longitudinal validation remains necessary.

Why this record is monitored

This record has an Impact Signal of 72/100 based on recency, source, collaboration, and bibliographic signals. It prioritizes monitoring and is not a judgment of research quality.

Related topics: Reproductive Physiology in Livestock · Smart Agriculture and AI · Animal Behavior and Welfare Studies