A deep learning-based detection model and illumination-adaptive behavioral analysis for soldier crabs in the intertidal zone
Soldier crabs (Mictyris spp.) are widely recognized as ecological indicators in intertidal ecosystems. Their behavioral rhythms respond sensitively to environmental fluctuations, especially variations in sediment composition and illumination. However, due to their small size and tidal-driven activit...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1574954125003474 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Soldier crabs (Mictyris spp.) are widely recognized as ecological indicators in intertidal ecosystems. Their behavioral rhythms respond sensitively to environmental fluctuations, especially variations in sediment composition and illumination. However, due to their small size and tidal-driven activity patterns, conventional behavior detection methods suffer from low efficiency and considerable observer bias, particularly under dark conditions where detection errors and omissions are prevalent. To address these limitations, we propose a lightweight deep learning model, SFS-YOLO, built upon the YOLOv11n framework, for accurate behavior detection of soldier crabs under varying illumination. The model integrates an SPD-Conv module for efficient downsampling, a fine-grained channel attention mechanism for enhanced feature representation, and a robust SIoU loss function to improve the localization of small targets. Experimental results demonstrate that SFS-YOLO improves mean average precision from 92.4% to 94.8%, increases recall from 93.0% to 93.6%, reduces the number of parameters by 8.33%, and compresses the model size by 0.4 MB. The model achieves mAP scores of 95.4% and 93.2% under bright and dark conditions, respectively. Beyond detection performance, this study investigates sediment-specific behavioral rhythms and identifies light-dependent temporal patterns that may serve as early ecological indicators of habitat perturbation. The proposed approach offers a practical solution for intelligent monitoring of small benthic organisms in intertidal zones. Moreover, the rhythm-based indicators developed herein show strong potential for applications in biodiversity assessment and sustainable ecosystem management within dynamic coastal environments. |
|---|---|
| ISSN: | 1574-9541 |