Enhanced underwater vision: a refined Faster R-CNN approach for object detection
The underwater environment presents unique challenges for robotic vision due to low light conditions and medium interference, which often result in images with low contrast and indistinct textures. To address these challenges, we developed an enhanced object detection method for underwater vision, leveraging the refined Faster R-CNN framework augmented with image enhancement techniques. During preprocessing, a Multi-Scale Retinex (MSRCR) algorithm with color restoration is used to minimize the effects of underwater interference and restore the original color, contrast, and feature details of the images. Moreover, the model’s capability is enhanced by substituting the conventional VGG-16 network with a more robust ResNet-101 deep neural network, aiming for superior feature extraction and recognition accuracy. Comprehensive testing on an underwater dataset revealed that our modified algorithm not only improves the accuracy and mean Average Precision (mAP) by 3.3% and 2.9%, respectively, over the standard Faster R-CNN but also outperforms other leading algorithms in recognition accuracy.
Authors
- Tao Yin (ORCID: https://orcid.org/0009-0003-8927-6838)
- Xiantao Jiang (ORCID: https://orcid.org/0000-0002-2226-482X)
- Zhongdai Wu
- Jin Liu
- Bing Han
Institutions
- Shanghai Ship and Shipping Research Institute (CN)
- Shanghai Maritime University (CN)
Publication Details
- Journal
- Journal on Image and Video Processing
- Published
- 2026-11-19
- DOI
- https://doi.org/10.1186/s13640-026-00711-9
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00