Underwater Object Detection Enhancement via Channel Stabilization
This work addresses the problem of detecting marine trash in underwater environments, which is crucial for mitigating harm to aquatic ecosystems, but it appears incremental as it builds on existing Detectron2 methods with enhancements.
The paper tackles underwater object detection for marine trash by proposing a channel stabilization technique and simplified image enhancement model to reduce haze and color cast, achieving a 9.53% absolute increase in average precision for small objects and a 7% absolute gain in bounding box detection compared to the baseline on the TrashCan Dataset.
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: https://github.com/aliman80/Underwater- Object-Detection-via-Channel-Stablization