Real-Time Underwater Plastic Waste Detection Using an Enhanced YOLOv8 Pipeline with Learning-Based Image Restoration

Anshuman Nigam

Keywords: Underwater Image Restoration, Object Detection, YOLOv8, Marine Debris Detection, Real-Time Systems, Autonomous Underwater Vehicles (AUV) Computer Vision

TL;DR: Enhancing degraded underwater images improves real-time plastic waste detection accuracy using YOLOv8.

Abstract:

Marine plastic pollution poses a significant threat to aquatic ecosystems, necessitating scalable and autonomous monitoring systems. However, underwater visual perception remains challenging due to light attenuation, scattering, turbidity, and color distortion, which degrade image quality and adversely affect object detection performance.

This paper proposes a real-time underwater plastic waste detection pipeline that integrates a learning-based image restoration module with a YOLOv8 object detector. The degradation process can be modeled as $I(x) = J(x)t(x) + A(1 - t(x))$, where $I(x)$ is the observed image, $J(x)$ is the scene radiance, $t(x)$ is the transmission map, and $A$ represents ambient light. Instead of explicitly estimating these parameters, a neural network learns a direct mapping $f_{\theta}: I \rightarrow \hat{J}$ to restore degraded images.

The restoration model is trained on the UIEB dataset, enabling improved visual quality under diverse underwater conditions. The enhanced images are then used for detection, leading to improved feature representation and object localization.

Experimental results demonstrate that the proposed approach improves detection performance from 0.61 to 0.73 mAP@0.5, with corresponding gains in precision and recall, while maintaining real-time inference at 24 FPS. Additional evaluations under low-visibility conditions confirm robustness, showing significant improvements in challenging environments.

The proposed system provides an efficient and deployable solution for autonomous underwater monitoring, highlighting the importance of integrating image restoration with object detection in domain-specific perception tasks.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Paper Acceptance: No

Submission Number: 12