Marine Environmental Monitoring Via Attention Enhanced Yolov10 For Debris Detection

Abstract

Marine debris severely threatens aquatic ecosystems and biodiversity. Automated underwater debris detection is essential for scalable environmental monitoring. This paper proposes a lightweight real-time debris detection framework based on YOLOv10n optimized for edge deployment. The proposed system integrates adaptive preprocessing, feature pyramid fusion, and anchor-free detection mechanisms. Mathematical formulations of detection loss, bounding box regression, and evaluation metrics are presented. Experimental evaluation demonstrates 94.8% mAP@0.5 with 32 FPS inference speed on edge hardware. Comparative analysis confirms reduced computational complexity while maintaining high accuracy. The proposed model provides a scalable and energy-efficient solution for marine environmental surveillance.

DOI: https://doi-ds.org/doilink/03.2026-35676848