Related Experiment Videos
Toward high-accuracy underwater object detection in optically challenging environments via dynamic adaptive
Applied Optics
|August 13, 2026
Summary
This study introduces a dynamic adaptive object detection model for challenging underwater environments. The novel approach enhances feature representation and uses a specialized loss function, achieving state-of-the-art accuracy with real-time efficiency.
Area of Science:
- Computer Vision
- Robotics
- Oceanography
Background:
- Underwater object detection is hindered by poor visibility, low light, and spectral distortion.
- Existing vision-based systems struggle with reliability in these complex aquatic conditions.
Purpose of the Study:
- To develop a novel dynamic adaptive object detection model for improved underwater perception.
- To enhance feature representation and localization robustness for small and blurry objects.
Main Methods:
- Implemented a collaborative optimization framework with dynamic feature fusion (DySample) and dynamic convolution (DyC2f).
- Designed a tailored loss function (FWNWD) incorporating WIoU, Focaler-IoU, and normalized Wasserstein distance (NWD).
- Evaluated the model on the large-scale DUO dataset.
Main Results:
- Achieved state-of-the-art mAP@0.5 of 91.7% with low computational cost (6.6G FLOPs).
- Demonstrated real-time performance at 23 FPS on an embedded platform (Orange Pi Aipro).
- The FWNWD loss improved recall for small objects by 2.3%.
Conclusions:
- The proposed model offers a practical and efficient solution for high-precision underwater object detection.
- The dynamically tunable architecture provides insights for vision systems in other degraded environments.