Improving Marine Ecosystem Monitoring through Transfer Learning and Deep Multisensor Data Integration

Authors

• Jie Wang, Anhui University
• Shujing Qin, Wuhan University
• Rana Muhammad Zulqarnain, Saveetha School of Engineering
• Qaiser Abbas, Korea University
• Frank Juma Ong'ondo, Mississippi State University
• Mitiku Badasa Moisa, Wollega University
• Habib Kraiem, Northern Border University
• Nazih Y. Rebouh, RUDN University

ORCID

Ong'ondo: https://orcid.org/0000-0002-3550-2110

MSU Affiliation

College of Arts and Sciences; Department of Geosciences

Creation Date

2026-06-01

Abstract

For sustainable resource management, environmental monitoring, and assessing geomorphological changes in dynamic coastal zones, it is essential to classify coastal and aquatic habitats correctly. Recent improvements in remote sensing technology, together with deep learning, have made it much easier for us to analyse complex land cover patterns in high resolution satellite images. This study utilizes the NWPU RESISC45 dataset and a transfer learning technique based on the ResNet50 convolutional neural network to develop a robust deep learning framework for classifying images of China's coastline and aquatic bodies into multiple classes. We employed a carefully chosen set of six land cover classes: beach, island, lake, river, sea ice, and wetland, to focus on elements that are found in water and on the coast. The suggested technique combines the latest image preprocessing, stratified data splitting, and fine-tuning of ResNet50. After that, the model is thoroughly evaluated using precision, recall, F1-score, and confusion matrix analysis. The experimental findings show that the test set had a high classification accuracy of 95.7%, combining with all categories doing well. The confusion matrix analysis shows that the model can distinguish between classes that look similar, with just a few mistakes. The current study uses high-resolution optical images, but the suggested framework may easily be expanded to include data from many sensors, such as synthetic aperture radar (SAR) and multispectral or hyperspectral sensors. The results show how deep learning and transfer learning might improve the monitoring of coastal areas. They also set the way for future research that combine several remote sensing sources to improve coastal environmental analysis even further in coastal zones in an environmentally sustainable manner.

Publication Date

11-13-2025

Publication Title

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Publisher

Institute of Electrical and Electronics Engineers

First Page

1

Last Page

17

Creative Commons License

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

Recommended Citation

J. Wang et al., "Improving Marine Ecosystem Monitoring through Transfer Learning and Deep Multisensor Data Integration," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, doi: 10.1109/JSTARS.2025.3632334.