A Large Scale Side-Scan Sonar Dataset of Seafloor Sediments for Self-Supervised Pretraining

This dataset serves as an extension to the dataset part of "A convolutional vision transformer for semantic segmentation of side-scan sonar data" published in Ocean Engineering, Volume 86, part 2, 15 October 2023, DOI: 10.1016/j.oceaneng.2023.115647 for self-supervised pretraining. This dataset consists of patches of side-scan sonar waterfalls collected along the coast of Catalunya during an extensive survey. The waterfalls were partitioned in batches of 384 lines to generate images of size 384 × 384 with a 192 pixel-overlap along-track and across-track. This resulted in a total of 434,164 images capturing various seafloor types including rocky bottoms, sand ripples, detrital funds, posidonia, cymocea, mud, corals, artificial reefs etc. Additional tools for using the data for self-supervised pretraining can be found under https://github.com/DeeperSense/deepersense-seafloorscan Acknowledgements The data in this repository were collected by Tecnoambiente SL as part of the project DeeperSense that received funding from the European Commission. Program H2020-ICT-2020-2 ICT-47-2020. Project Number: 101016958.

本数据集作为发表于《海洋工程》（Ocean Engineering）2023年10月15日第86卷第2期、DOI为10.1016/j.oceaneng.2023.115647的论文《用于侧扫声呐（side-scan sonar）数据语义分割的卷积视觉Transformer（convolutional vision transformer）》中数据集部分的扩展，用于自监督预训练（self-supervised pretraining）。

本数据集包含在加泰罗尼亚沿岸大规模勘测期间采集的侧扫声呐瀑布图切片。研究人员将这些瀑布图按384条测线为一批进行划分，生成尺寸为384×384的图像，且沿航迹与跨航迹方向均存在192像素的重叠。最终共得到434164张图像，涵盖多种海底类型，包括岩质底、沙纹、碎屑基底、波西多尼亚海草（posidonia）、仙掌藻（cymocea）、泥质底、珊瑚以及人工鱼礁等。

用于该数据集自监督预训练的配套工具可于https://github.com/DeeperSense/deepersense-seafloorscan获取。

致谢
本仓库中的数据由Tecnoambiente SL公司作为DeeperSense项目的一部分采集，该项目获得欧盟委员会H2020-ICT-2020-2 ICT-47-2020计划资助，项目编号：101016958。

搜集汇总

背景与挑战

背景概述

该数据集是一个大规模侧扫声纳数据集，包含434,164张384×384像素的图像，覆盖多种海底沉积物类型，专为自监督预训练设计。数据采集自加泰罗尼亚海岸，是已发表研究的扩展部分，并提供额外工具支持使用。