The Marine Debris Forward-Looking Sonar Datasets

2025-03-28

AI Summary

The development of underwater AI is hindered by the scarcity of publicly available, high-quality forward-looking sonar (FLS) data. Method: This paper introduces FSSonar-ML—the first open, multi-scenario FLS dataset specifically designed for marine litter detection. It encompasses three representative underwater environments: water tank, turntable, and submerged quarry, comprising thousands of high signal-to-noise ratio sonar images with fine-grained annotations for classification, object detection, semantic segmentation, image patch matching, and unsupervised pretraining. The entire pipeline—including sonar imaging acquisition, multi-platform sensor calibration, and meticulous manual annotation—is fully reproducible, and a benchmarking framework is open-sourced. Contribution/Results: Hosted on Zenodo, FSSonar-ML enables cross-scenario generalization studies. Baseline evaluations across five fundamental computer vision tasks confirm its effectiveness, significantly enhancing both trainability and generalization capability of sonar image understanding models.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Remote Sensing / Geospatial AI

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Sonar sensing is fundamental for underwater robotics, but limited by capabilities of AI systems, which need large training datasets. Public data in sonar modalities is lacking. This paper presents the Marine Debris Forward-Looking Sonar datasets, with three different settings (watertank, turntable, flooded quarry) increasing dataset diversity and multiple computer vision tasks: object classification, object detection, semantic segmentation, patch matching, and unsupervised learning. We provide full dataset description, basic analysis and initial results for some tasks. We expect the research community will benefit from this dataset, which is publicly available at https://doi.org/10.5281/zenodo.15101686

Problem

• Lack of public sonar datasets for AI training
• Need diverse marine debris datasets for sonar tasks
• Providing multi-task sonar data for underwater robotics

Innovation

• Diverse sonar datasets in three settings
• Supports multiple computer vision tasks
• Publicly available for research community