Underwater Object Detection Enhancement via Channel Stabilization

Muhammad Ali,Salman Khan

2024-08-02

Abstract:The complex marine environment exacerbates the challenges of object detection manifold. Marine trash endangers the aquatic ecosystem, presenting a persistent challenge. Accurate detection of marine deposits is crucial for mitigating this harm. Our work addresses underwater object detection by enhancing image quality and evaluating detection methods. We use Detectron2's backbone with various base models and configurations for this task. We propose a novel channel stabilization technique alongside a simplified image enhancement model to reduce haze and color cast in training images, improving multi-scale object detection. Following image processing, we test different Detectron2 backbones for optimal detection accuracy. Additionally, we apply a sharpening filter with augmentation techniques to highlight object profiles for easier recognition. Results are demonstrated on the TrashCan Dataset, both instance and material versions. The best-performing backbone method incorporates our channel stabilization and augmentation techniques. We also compare our Detectron2 detection results with the Deformable Transformer. In the instance version of TrashCan 1.0, our method achieves a 9.53% absolute increase in average precision for small objects and a 7% absolute gain in bounding box detection compared to the baseline. The code will be available on Code: <a class="link-external link-https" href="https://github.com/aliman80/Underwater-" rel="external noopener nofollow">this https URL</a> Object-Detection-via-Channel-Stablization

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenge of underwater object detection in complex marine environments, especially how to accurately detect and locate underwater garbage. With the increase in modern plastic production and poor waste management, underwater garbage poses a serious threat to the marine ecosystem. Therefore, the development of effective underwater object detection techniques is crucial for alleviating this harm.

### Specific problems include:
1. **Complex marine environment**: Underwater images are usually affected by physical phenomena such as scattering and absorption, resulting in low - contrast, blurry images with color casts (such as being dominated by blue or green).
2. **Difficulty in detecting small - scale objects**: Underwater garbage is usually small in size and has a complex background, which increases the difficulty of detection.
3. **Limitations of existing methods**: Existing underwater object detection methods have limited effectiveness in dealing with these challenges, especially under low - light conditions.

### Main objectives of the paper:
- Propose a new Channel Stabilization technique to reduce fog and color casts in training images.
- Improve the performance of Detectron2 in underwater object detection tasks by enhancing image quality and using different backbones.
- Explore the optimal backbone network configuration to achieve higher detection accuracy.
- Verify the proposed method on the TrashCan dataset and compare it with existing methods.

### Main contributions:
- Propose LAB - Stretching and global stretching pre - processing methods for removing color fog and color casts in images.
- Introduce a channel stabilization module to reduce the dominant effect of specific colors (such as blue).
- Use sharpening filters and data augmentation techniques to highlight object contours and improve detection results.
- Experimental results show that in small - size object detection, the proposed framework has a significant improvement compared to the baseline method. For example, in the instance version of the TrashCan 1.0 dataset, the average precision is increased by 9.53%.

Through these improvements, the paper aims to provide more effective and accurate technical means for underwater garbage detection, thereby helping to protect the marine environment.