EQUAINE: Asking machines to understand equine data and keep learning from it
Synopsis
This thesis explores how sensor signals combined with AI models can quantify equine locomotion and respiration, finding that a single limb sensor can accurately classify terrain type even at low sampling rates, that a combination of head, withers, and pelvis sensors can discriminate between sound and lame strides with performance varying by front or hind limb lameness, that vertical ground forces are better estimated from upper-body sensors than limb sensors, that dynamic respiratory rate can be computed from downsampled audio signals, and that explainable AI confirms models rely on kinematic landmarks similar to those used by veterinarians, while also presenting a data collection platform for harness racing horses and publishing an audio dataset.
Interpretation
A single limb sensor can accurately classify the type of terrain onto which a horse walks or trots, even at low sampling rates. Prior terrain classification often relied on multiple sensors or high sampling rates; this work shows a single limb sensor at low sampling rates suffices, lowering hardware and computational barriers. Based on sensor signals and AI models, the thesis reports accurate classification, but no specific sample sizes or statistical metrics are provided.
A combination of sensors placed on the head, withers, and pelvis can discriminate between sound and lame strides, with performance varying according to whether the lameness stems from a front or hind limb. Applies multi-site sensor fusion to lameness detection and reveals differing model performance for front versus hind limb lameness, offering a new perspective for clinical monitoring. Based on multi-sensor combination and AI models, the thesis reports discrimination ability, but no specific performance values or sample sizes are given.
Vertical ground forces can better be estimated from upper-body sensors compared to limb sensors, and dynamic respiratory rate can be computed from downsampled audio signals. Highlights the advantage of upper-body sensors for estimating vertical ground forces and demonstrates that downsampled audio can be used for respiratory rate calculation, paving the way for less computationally expensive tools. Based on sensor signal and audio signal processing, the thesis reports estimation and computation results, but no error metrics or validation details are provided.
Explainable AI shows that lameness detection models rely on kinematic landmarks similar to those used by veterinarians, while identifying additional movement phases not perceptible to the naked eye. Uses explainable AI to verify that model decisions align with veterinary clinical judgment and discovers new movement phases, enhancing transparency and clinical acceptability of these tools. Based on explainable AI analysis, the thesis reports feature selection and movement phase identification, but no quantitative explanation metrics are provided.
Perspective
This work targets equine locomotion and respiration monitoring, applicable to settings requiring regular monitoring of lameness and respiratory status, such as training and clinical environments. Its methods can inspire other animal or human motion analysis, but require re-validation on corresponding data. The public dataset and collection platform provide a foundation for the community, but application scope is limited by the types of data collected and horse breeds.
The thesis does not provide specific sample sizes, statistical metrics, or error values; readers may wonder about model generalization across different horses and ground conditions, and the clinical significance of movement phases identified by explainable AI. Additionally, the scale and diversity of the public dataset, the practical effect of real-time feedback in training, and long-term performance in real-world deployment are directions worth further observation.