2026-LU06 AI-Driven Computer Vision for Automated Monitoring of Marine Biodiversity and Climate Impacts

This project will develop AI/deep learning–based computer vision tools to automatically identify and count barnacles, macroalgae, and other invertebrates. Training will use 3000 images per year from the 20-year UK MarClim survey, supplemented by public datasets. Advanced models—including CNNs, instance and panoptic segmentation, vision transformers, attention, and self- or few-shot learning—will address complex intertidal scenes. To handle challenges such as similar species, subtle differences, variable life stages, irregular, incomplete or unseen specimens, contrastive learning, anomaly detection, and synthetic data augmentation will be explored. The resulting AI framework will enable high-resolution, large-scale monitoring, providing robust abundance data and novel insights into species’ responses to anthropogenic and climate-driven stressors, supporting coastal ecosystem conservation and management. 

Domain Knowledge: MBA experts will validate datasets, interpret outputs, identify biases, and ensure AI tools are accurate and reliable, providing actionable insights for marine biodiversity monitoring, climate impact assessment, and evidence-based policy.  

DRs will be awarded CENTA Training Credits (CTCs) for participation in CENTA-provided and ‘free choice’ external training. One CTC can be earned per 3 hours training, and DRs must accrue 100 CTCs across the three and a half years of their PhD.  

The student will be trained in advanced deep learning methods—including CNNs, panoptic segmentation, vision transformers, attention mechanisms, and self- or few-shot learning—alongside practical skills in image annotation, dataset curation, and model evaluation. Training by AI specialists and MBA experts will integrate ecological knowledge with computational methods. 

Experts from MBA will train team members in species identification, life-stage recognition, and image annotation, guide AI output interpretation, identify biases, and advise on integrating AI tools into monitoring workflows to produce reliable and actionable data for evidence-based policy and management. 

Loughborough provides AI modules, high-performance GPU-A100 access and ethical AI training. 

Professor Baihua Li, an AI expert at Loughborough University with over 20 years’ experience in computer vision and machine learning, has published more than 120 papers and supervised 20+ PhDs. She leads AI research for environmental protection, including an EPSRC-funded £2.5M project, as well as several UKRI- and Innovate UK–funded initiatives. 

Dr. Nova Mieszkowska at the Marine Biological Association leads large-scale MarClim, the world’s most extensive rocky intertidal dataset. Her research combines experimental ecology, genomics, and long-term monitoring to reveal rapid climate-driven species shifts, providing critical insights for marine biodiversity conservation and national and international policy. 

Year 1: Foundations & Data Preparation 

• Conduct literature review on marine biodiversity monitoring and deep learning methods. 

• Receive training from MBA in species identification, life-stage recognition, and annotation. 

• Curate, clean, and annotate MarClim datasets. 

• Develop baseline computer vision models (CNNs, object detection, segmentation). 

• Present initial results at workshops/conferences. 

Year 2: Model Development & Refinement 

• Implement advanced architectures (e.g. vision transformers, attention, panoptic segmentation). 

• Address challenges: species similarity, subtle textures/shapes, life-stage recognition, unseen/incomplete specimens. 

• Explore contrastive learning, anomaly detection, and synthetic data augmentation. 

• Iteratively train, validate, and refine models  

• Publish initial findings in an AI or ecological journal. 

Year 3: Integration, Impact & Dissemination 

• Develop operational AI framework for large-scale, automated biodiversity monitoring. 

• Integrate ecological knowledge with AI outputs to generate actionable abundance and distribution insights. 

• Apply framework to long-term MarClim data to assess climate- and anthropogenic-driven changes. 

• Share findings and AI tools with policymakers, conservation organizations, and relevant stakeholders. 

• Write thesis, publish high-impact papers, present at international conferences. 

Year 4 (if needed): Improve results, prepare, submit, and revise journal publications, defend the PhD thesis (viva). 

1. MarClim project & reports. Marine Biodiversity and Climate Change | Marine Biological Association,  https://www.mba.ac.uk/marclim-mieszkowska/?utm_source=chatgpt.com  
2. Mei, S., Lian, J., Wang, X., Su, Y., Ma, M. and Chau, L.P., 2024. A comprehensive study on the robustness of deep learning-based image classification and object detection in remote sensing: Surveying and benchmarking. Journal of Remote Sensing.  
3. M. Sonmez, et al. 2023. Deep learning-based classification of microalgae using light and scanning electron microscopy images. Micron, 172.  
4. Li, Z., Wang, Y., Zhang, N., Zhang, Y., Zhao, Z., Xu, D., Ben, G. and Gao, Y., 2022. Deep learning-based object detection techniques for remote sensing images: A survey. Remote Sensing, 14(10).