Original Research Articles
Automated Detection of Nocturnal Blue Tears Using YOLO12n and CutMix Data Augmentation
Received 18 Nov 2026, Published online: 18 Nov 2026
Published in Volume 6, Issue 2, 2026, pages 69-73
Shahbaz Nazeer, S. N., Abdul Qadeer, A. Q., Rabia Saleem, R. S., Saher Asghar, S. A., & Faiq Hassan, F. H. (2026). Automated Detection of Nocturnal Blue Tears Using YOLO12n and CutMix Data Augmentation. Journal of Magnetism and its Applications, 6(2), 69-73.
Open access
Abstract
Blue Tears are transient coastal bioluminescent events that are challenging to detect in low-light images. This study evaluates YOLO12n with CutMix augmentation on BT3.8k, a benchmark of 3,827 annotated images extracted from 241 user-uploaded videos. Training used 260 epochs, 640 × 640 inputs and CutMix probability 0.5. The reported evaluation yielded 95.09% mAP@50, 72.03% mAP@50–95, 92.45% precision, 89.89% recall and a 91.15% F1-score. Compared with YOLO12n without CutMix, mAP@50 improved by 7.86 percentage points. The detector contains 2.56 million parameters versus 22.39 million reported for BT-YOLO, although this cross-study comparison did not use a unified retraining protocol. These findings support augmentation-assisted detection on BT3.8k, while clip-level separation, geographically independent evaluation and deployment latency still require validation.
Keywords
- Blue Tears; bioluminescence; YOLO12n; CutMix; object detection; BT3.8k; coastal monitoring
Metrics
- Article views
- 0
- Downloads
- 0
Licensing
© 2026 Shahbaz Nazeer, Abdul Qadeer, Rabia Saleem, Saher Asghar, Faiq Hassan. This article is published under CC BY-SA 4.0.
How to cite
Shahbaz Nazeer, S. N., Abdul Qadeer, A. Q., Rabia Saleem, R. S., Saher Asghar, S. A., & Faiq Hassan, F. H. (2026). Automated Detection of Nocturnal Blue Tears Using YOLO12n and CutMix Data Augmentation. Journal of Magnetism and its Applications, 6(2), 69-73.