Underwater Image-Based Crab Detection Using Hybrid RetinaNet and YOLOv8
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Detecting crabs in underwater environments presents significant challenges, including low visibility, occlusions, and fluctuating lighting conditions.This study introduces a hybrid RetinaNet-YOLOv8 model aimed at improving the accuracy and robustness of crab species identification by combining the strengths of both object detection frameworks.YOLOv8 provides real-time detection features with impressive recall, whereas RetinaNet is particularly effective in addressing class imbalance through its focal loss function.The model under consideration is developed using the Crab Species Dataset, which includes a variety of crab images taken in their natural environments.A comprehensive series of experiments was carried out over 50 epochs, with performance assessed through critical metrics: Accuracy (0.94), Precision (0.96), Recall (0.92), F1-score (0.94), and Intersection over Union (IoU).The hybrid RetinaNet-YOLOv8 approach shows enhanced performance over standalone models, surpassing both RetinaNet and YOLOv8 in terms of detection accuracy and robustness under challenging conditions.The findings indicate that the combination of RetinaNet and YOLOv8 improves the detection of crab species, positioning it as an effective tool for ecological monitoring and studies on marine biodiversity.
Fields of Study & Subjects
Advanced Neural Network Applications
Water Quality Monitoring Technologies
Domain Adaptation and Few-Shot Learning
Publication details
- Journal / source
- Traitement du signal
- Volume
- 43
- Issue
- 1
- Pages
- 335–350
- Publication date
- 28 February 2026
- Language
- EN