In Shift and In Variance: Assessing the Robustness of HAR Deep Learning Models Against Variability

Sensors, Journal Year: 2025, Volume and Issue: 25(2), P. 430 - 430

Published: Jan. 13, 2025

Deep learning (DL)-based Human Activity Recognition (HAR) using wearable inertial measurement unit (IMU) sensors can revolutionize continuous health monitoring and early disease prediction. However, most DL HAR models are untested in their robustness to real-world variability, as they trained on limited lab-controlled data. In this study, we isolated analyzed the effects of subject, device, position, orientation variabilities HARVAR REALDISP datasets. The Maximum Mean Discrepancy (MMD) was used quantify shifts data distribution caused by these variabilities, relationship between model performance drawn. Our results show that different types variability significantly degraded performance, with an inverse performance. compounding effect multiple studied further underscores challenges generalizing conditions. Analyzing impacts highlights need for more robust generalize effectively settings. MMD proved valuable explaining drops, emphasizing its utility evaluating

Language: Английский

Rumination Detection in Sheep: A Systematic Review of Sensor-Based Approaches

Animals, Journal Year: 2023, Volume and Issue: 13(24), P. 3756 - 3756

Published: Dec. 5, 2023

The use of sensors to analyze behavior in sheep has gained increasing attention scientific research. This systematic review aims provide an overview the developed and used detect rumination Moreover, this provides details that are currently commercially available describes their suitability for based on information provided literature found. Furthermore, lists best sensor performances terms achieved accuracy, sensitivity, precision, specificity detection, detailing, when applicable, position epoch settings were achieve results. Challenges areas future research development also identified. A search strategy was implemented databases PubMed, Web Science, Livivo, yielding a total 935 articles. After reviewing summaries 57 articles remaining following filtration (exclusion) repeated unsuitable articles, 17 fully met pre-established criteria (peer-reviewed; published between 2012 2023 English or German; with particular focus detecting sheep) included review. guidelines outlined PRISMA 2020 methodology followed. results indicate sensor-based systems have been utilized monitor behavior, among other behaviors. Notably, none identified specifically designed sheep. In order meet specific needs sheep, customized solution is necessary. Additionally, further investigation optimal Implications: utilization such significant implications improving welfare enhancing our knowledge various contexts.

Language: Английский