.png){kind=icon}
.png){kind=icon}
Return
Swimming behavior of fish across habitats: A trajectory based analysis using YOLO–DeepSORT and Markov chain models
.png){kind=icon}
OA
AI
N
Y
N
H
DOI：10.1016/j.ecoinf.2026.103967.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Computer vision offers a powerful tool for analyzing the swimming behavior of fish and understanding how it varies across natural habitats. We developed a video based behavioral analysis framework to compare damselfish (Pomacentridae) across five environments: submerged tree root zones, shallow coastal zones, coral reefs, bleached coral reefs, and seagrass meadows. The pipeline integrates image enhancement, fish detection and tracking, and behavioral modeling. Contrast-limited Adaptive Histogram Equalization (CLAHE) is applied to enhance underwater images, enabling the YOLOv12 model to reach a detection accuracy of 93.0%, a 3.5% improvement over unprocessed images. Fish trajectories are tracked using DeepSORT, from which kinematic features such as swimming speed and turning angle are extracted to construct behavioral descriptors. Then, a Markov chain model is used to characterize behavioral state transitions. The results indicate that coral reefs and shallow coastal zones promote faster swimming with frequent deceleration and directional changes, while bleached coral reefs exhibit unstable behavioral transitions, suggesting unfavorable conditions. In contrast, structurally complex habitats such as seagrass meadows and submerged tree root areas support slower, more stable swimming patterns associated with resting and foraging.
Keywords:
Fish behavior
Habitats
YOLOv12
Markov model
Tracking
Journal
IF:
7.3
Papers:
3.7K
Citations:
1.3W
Organization
Cited Papers