Yertegin Nurlan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yertegin Nurlan has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Linrong Li
Shanxi Jiaokong Fenshi Expressway Co., Ltd., Lvliang 033000, China
1 shared publication - Mengyuan Zeng
College of Transportation, Tongii University, Shanghai, China
1 shared publication - Yadi Sang
College of Transportation, Tongji University, Shanghai 201804, China
1 shared publication - Yahor M. Zhukouski
Faculty of Transport Communications, Belarusian National Technical University, 220013 Minsk, Belarus
1 shared publication
Latest Publications
- Open Access
- 21 July 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (14) , 7287
Distributed Fiber-Optic Sensing Data-Based Vehicle Event Recognition
Distributed optical vibration sensing (DOVS) provides dense spatiotemporal measurements for pavement and traffic monitoring, but nonstationary background noise, spatially confined responses, and data-quality anomalies complicate vehicle-event detection. This study presents a deterministic, training-free, and interpretable detector for single-lane highway DOVS matrices. The algorithm forms a detrended absolute-amplitude representation and combines percentile-based temporal candidate detection, robust background estimates based on the median and median absolute deviation (MAD), a candidate spatial-width fraction derived from channel-specific thresholds, track-direction evidence, and explicit decision rules. Evaluation on 3085 manually labeled matrices acquired from 2023 to 2026 yielded 87.23% accuracy, 87.42% precision, 87.31% recall, and an F1-score of 87.36%. In a secondary analysis, excluding 98 quality-flagged matrices increased precision to 91.28% and F1-score to 89.20%; the exclusion removed 67 of 196 false positives and no false negatives. Relative to the diagnostic-refinement configuration, the final rule set increased F1-score by 2.173 percentage points, with a matrix-level bootstrap 95% confidence interval of 1.437–2.963 percentage points. The exact McNemar test for paired correctness differences gave p = 6.60 × 10−9. A sensitivity configuration changed only four classifications and produced no meaningful gain. These results quantify performance at the tested site; narrow responses, upward-like tracks, data-quality anomalies, single-annotator labels, and post hoc rule selection limit broader inference.
- 21 July 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (14) , 7287
Open AccessDistributed Fiber-Optic Sensing Data-Based Vehicle Event Recognition
Distributed optical vibration sensing (DOVS) provides dense spatiotemporal measurements for pavement and traffic monitoring, but nonstationary background noise, spatially confined responses, and data-quality anomalies complicate vehicle-event detection. This study presents a deterministic, training-free, and interpretable detector for single-lane highway DOVS matrices. The algorithm forms a detrended absolute-amplitude representation and combines percentile-based temporal candidate detection, robust background estimates based on the median and median absolute deviation (MAD), a candidate spatial-width fraction derived from channel-specific thresholds, track-direction evidence, and explicit decision rules. Evaluation on 3085 manually labeled matrices acquired from 2023 to 2026 yielded 87.23% accuracy, 87.42% precision, 87.31% recall, and an F1-score of 87.36%. In a secondary analysis, excluding 98 quality-flagged matrices increased precision to 91.28% and F1-score to 89.20%; the exclusion removed 67 of 196 false positives and no false negatives. Relative to the diagnostic-refinement configuration, the final rule set increased F1-score by 2.173 percentage points, with a matrix-level bootstrap 95% confidence interval of 1.437–2.963 percentage points. The exact McNemar test for paired correctness differences gave p = 6.60 × 10−9. A sensitivity configuration changed only four classifications and produced no meaningful gain. These results quantify performance at the tested site; narrow responses, upward-like tracks, data-quality anomalies, single-annotator labels, and post hoc rule selection limit broader inference.