Datasets for hydrothermal plume detection

Published: 5 July 2022| Version 1 | DOI: 10.17632/dg2595f68b.1
Contributors:
,

Description

This repository contains following two dataset which were used for training, validation and testing of deep learning models to detect signatures of hydrothermal plumes from acoustic images. The dataset "20220110_signal_only.zip" is for one-class training of object detection models. Locations of only signals in images were annotated. The dataset "20220111_with_noise.zip" is for two-class training. Locations of signals and noises (similar color patterns with signals, but different shapes) were annotated. Both datasets contain 800 images and annotation information for training, 100 images and annotation information for validation, and 280 images and annotation information for test. Annotation information is formatted in two types. One is "via_region_data.json", which is applicable for utilizing Mask R-CNN model (https://github.com/matterport/Mask_RCNN). The other is "labels" directory which are applicable for utilizing YOLO-v5 models (https://github.com/ultralytics/yolov5).

Files

Steps to reproduce

Acoustic images were taken during research cruises YK14-17 and YK15-14 conducted by Japan Agency for Marine-Earth Science and Technology (JAMSTEC), using multi-beam echo sounder (MBES).

Institutions

Tokyo Daigaku, Chiba Kogyo Daigaku

Categories

Earth Sciences, Oceanography, Object Detection, Machine Learning, Image Database, Acoustic Imaging, Deep Learning

Licence