5 Malaysian Fish Species in Turbid Water

Dataset Overview

This dataset is designed for underwater fish classification under challenging environmental conditions. It supports domain adaptation research by addressing domain shift caused by variations in water clarity, illumination, turbidity, and image quality degradation.

The dataset is specifically constructed for evaluating deep learning-based domain adaptation methods to improve robustness in real-world underwater environments.

Task Description

The primary task is multi-class underwater fish classification under both within-domain and cross-domain settings.

The dataset consists of two distinct underwater image domains captured under different environmental conditions. For cross-domain evaluation, the two domains are alternately assigned as the source and target domains, enabling bidirectional assessment of domain adaptation performance.

Fish Classes

• Catfish (Keli)
• Climbing perch (Puyu)
• Climbing perch (Kaloi)
• Iridescent shark (Patin)
• Tinfoil barb (Lampam)

Dataset Structure

The dataset is organized into training, validation and testing:

/Train /Catfish /Climbing perch /Climbing perch /Iridescent shark /Tinfoil barb 

/Eva /Catfish /Climbing perch /Climbing perch /Iridescent shark /Tinfoil barb

/Test /Catfish /Climbing perch /Climbing perch /Iridescent shark /Tinfoil barb

Dataset Statistics

• Total images: 5400
• Source domain images: 1080 (216 each class)
• Target domain images: 1080 (216 each class)
• Testing: 135
• Number of classes: 5

Domain Characteristics

This dataset introduces significant domain shift due to underwater environmental conditions:

• Water turbidity and scattering effects
• Illumination variation and color distortion
• Low contrast and blur effects
• Depth-dependent visibility degradation

These challenges make cross-domain fish classification difficult and suitable for domain adaptation research.

Experimental Setting

Two evaluation settings are used:

• Within-domain: Training and testing on the same domain
• Cross-domain: Training on source domain and testing on target domain 

Usage

The dataset can be loaded using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("mieraehh/5_Malaysian_Fish_Species_in_Turbid_Water")

print(dataset)

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Users are free to use, share, and adapt the dataset as long as proper credit is given.

Citation

If you use this dataset, please cite: https://doi.org/10.5281/zenodo.21219965

Acknowledgements

This dataset was developed as part of research on underwater domain adaptation for intelligent fish classification systems. It aims to support sustainable fisheries monitoring and marine biodiversity conservation.

Notes

• Intended for academic and research use only.
• Please report both within-domain and cross-domain results for fair comparison.