As one of the most biologically complex and valuable ecosystems on the planet, coral reefs also extremely susceptible to environmental and climate changes. The investigation, reservation, and restoration of coral reefs require accurate classifications of the specific species of corals that form those reefs through examining the underwater coral reef images. Conventionally, such classifications are performed by experts in a manual fashion. This paper proposes a cross-domain coral image classification approach using a novel structure of dual-stream hierarchical neural networks. Our extensive experiments demonstrate that the approach does not only achieve state-of-the-art performances in classifying coral species but also has the capability to realize comparable and robust performances in dealing with coral image classification tasks with data from different oceanic regions and different coral life-cycle without labelling the target domains. Therefore, our approach has the potential to reduce ocean scientists' manual efforts in recognizing different species of corals.
- Linux Ubuntu 20.04
- python==3.7
- pytorch==1.10.1
- numpy==1.21.6
- CUDA=11.7
- GPU = RTX 3090Ti
- easydict==1.9
- tqdm==4.51.0
- Pillow==9.1.0
Please click the link to download the following datasets.
- Run DHDAN_train.py.
If you find the code useful in your research, please consider citing:
Hongyong Han, Wei Wang*, Gaowei Zhang, Mingjie Li, and Yi Wang. 2024. Cross-Domain Coral Image Classification Using Dual-Stream Hierarchical Neural Networks. In Proceedings of ACM SAC Conference (SAC’24). ACM New York, NY, USA, Article 4, 9 pages. https://doi.org/10.1145/3605098.
Please create a new issue