Answers are grounded in this paper and cite their sources.
Acoustic monitoring offers a non-invasive route to dairy-cow estrus recognition, but sparse target vocalizations and continuous-recording storage burdens limit practical use. We evaluated an acoustic detection framework that combines an enhanced Symbolic Aggregate Approximation (SAX) strategy with heterogeneous ensemble learning. The SAX-derived stage converts each 2-min audio window into candidate image representations and retains one image using a joint structural-similarity and gain-based criterion. The study monitored 116 Holstein cows at one commercial farm; complete 24-h estrus recordings were available for 27 cows, five complete non-estrus records served as controls, and 32 same-farm cases were used for onset-time comparison. High-pitched-audio extraction and redundancy-filtering accuracies were 94.85% and 94.99%, respectively, while retained-data storage decreased from 18 GB to 0.104 GB. The ensemble correctly classified 782 of 790 pooled test images (98.99%; Wilson 95% confidence interval, 98.01-99.49%). Predictions preceded reconstructed veterinary onset in 22 cases (mean lead, 88.64 min) and followed it in 10 cases (mean absolute delay, 27.90 min). These results support internal feasibility under the studied single-farm conditions; cow-independent generalization, external-farm validity, and embedded-hardware performance remain to be established.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: