An efficient YOLOv11-based approach for marine waste detection

View at publisher Summarise with VIRAT
Authors: Minchong Shen, Youfu Jiang
Cited by: 0
References: 6

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Marine debris detection is crucial for ocean environmental protection and ecosystem preservation. This paper presents an enhanced YOLOv11 architecture specifically designed for marine waste detection, incorporating three key improvements to address the challenges of detecting small-scale debris in complex marine environments. First, we replace the standard convolutional downsampling layers with Pinwheel-shaped Convolution (PConv), which utilizes asymmetric padding to capture multi-directional features more effectively. Second, the backbone C3k2 modules are enhanced with Efficient Multi-scale Attention (EMA) mechanism, forming C3k2_EMA blocks that improve feature representation through adaptive channel and spatial attention. Third, we substitute the traditional nearest-neighbor upsampling with Dynamic Sampling (DySample), which learns content-aware sampling offsets for more precise feature reconstruction. Experimental results on the Flow-Img and Trash-ICRA19 datasets show the model achieves 60.90% mAP@0.5:0.95 on Trash-ICRA19 and improves small object detection accuracy by 1.51% on Flow-Img, confirming its effectiveness over the baseline.

Fields of Study & Subjects

Advanced Neural Network Applications Microplastics and Plastic Pollution Remote-Sensing Image Classification

Publication details

Journal / source
International Conference on Algorithms, High Performance Computing and Artificial Intelligence (AHPCAI)
Pages
25–25
Publication date
19 March 2026
Language
EN