Author Identifiers

Noman Md Kislu

https://orcid.org/0000-0001-6281-1464

Islam Syed Mohammed Shamsul

https://orcid.org/0000-0002-3200-2903

Seyed Mohammad Jafar Jalali

https://orcid.org/0000-0002-2169-4350

Jumana Abu-Khalaf

https://orcid.org/0000-0002-6651-2880

Paul Lavery

https://orcid.org/0000-0001-5162-273X

Publication Date

2024

Document Type

Dataset

Publisher

Edith Cowan University

Faculty

School of Science

Funders

This research work is funded by the Australian Government Research Training Program (RTP) scholarship and supported by School of Science, Edith Cowan University, Australia.

Description

The "ECU-MSS-2' dataset contains four different habitats, 'Amphibolis' spp (hereafter 'Amphibolis"), 'Halophila' spp (hereafter 'Halophila'), 'Posidonia' spp (hereafter 'Posidonia') and 'Background'. We compiled this image dataset from different sources. The 'Halophila' images were collected by the Centre for Marine Ecosystems Research, Edith Cowan University, Western Australia.

The 'Amphibolis' 'Background' and 'Posidonia' images were collected by the Department of Biodiversity, Conservation and Attractions (DBCA), Australia. The 'Amphibolis' class includes the seagrass species Amphibolis griffithii and Amphibolis Antarctica. The 'Halophila' class includes Halophila ovalis, while the 'Posidonia' class includes Posidonia sinuosa, Posidonia coriacea, and Posidonia Australis. The 'Background' class includes coral, sand, sponge, seaweeds, fish and other benthic debris.

The dataset contained total 5,201 images, where the 'Amphibolis' class has 1304 images, the 'Background' class has 1237 images, the 'Halophila' class has 1315 images and the 'Posidonia' class has 1345 images. The total images were divided into training and test sets. The training set has 4,161 images and the test set has 1,040 images. All four classes have the same 260 images in the test set.

DOI

https://doi.org/10.25958/q8s6-5p21

Research Activity Title

BAOS-CNN: A novel deep neuroevolution algorithm for multispecies seagrass detection

Research Activity Description

We developed a novel Deep Neuroevolutionary (DNE) model that can automate the architectural engineering and hyperparameter tuning of CNNs models by developing and using a novel metaheuristic algorithm, named 'Boosted Atomic Orbital Search (BAOS)'. This proposed BAOS-CNN model achieves the highest overall accuracy (97.48%) among the seven evolutionary-based CNN models. The proposed model was tested and evaluated on this multi-species seagrass dataset.

Methodology

Centre for Marine Ecosystems Research, Edith Cowan University, Western Australia collected the Halophila images. They captured the images by a Fujifilm X-T30 mirror-less digital camera with wXF 18-55mm F 2.8-4 R LM OIS, all under natural light condition. DBCA collected the Amphibolis, Posidonia and background images. These images were captured by a downward-facing camera held 1-metre above the seagrass canopy with a field of view of 75 x 60 cm along transects used for measuring seagrass density/height. Ten downward-facing images were collected from each of the three transects per sites (30 images per site in total).

Start of data collection time period

2020

End of data collection time period

2022

File Format(s)

Image dataset

File Size

Total of files 12.9GB

Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

Article Location

 
COinS