Following on from our article on how Ternan is developing machine learning tools and AI initiatives, we spoke to Dr Samarth Bachkheti again in greater depth about our AI tool for boulder detection on the seabed floor.
Thanks for your time Sam. Can you walk us through how the AI tool works in detecting boulders on the seabed? What makes this approach unique compared to traditional methods?
Our AI tool is designed to identify boulders on the seabed by analysing Side Scan Sonar mosaics. We start by providing it with training data – essentially, examples of what boulders look like. Once trained, it scans new datasets and picks out similar features, detecting boulders automatically.
What makes this approach unique are its efficiency, reliability, and speed. Traditionally, identifying boulders requires manual interpretation, which is time-consuming and subjective. This tool accelerates the process, providing consistent and reliable results in far less time. This means we can analyse large areas quickly, improving decision-making for offshore wind farm installations and other marine projects.
How did the collaboration with UKRI and academic partners contribute to the development and credibility of this AI tool?
Adopting cutting-edge AI technology in geoscience workflows as an SME requires collaboration, and that’s where our partnerships have been invaluable. UKRI funding and our work with academic institutions, including the STFC Hartree Centre, gave us access to world-class expertise in data science, machine learning, and high-performance computing.
These collaborations helped refine the tool’s accuracy and reinforce its credibility – showing that this isn’t just an in-house experiment but a rigorously developed technology backed by scientific research and innovation.
Could you elaborate on the process of training the AI? How do you select and use training data to improve boulder detection accuracy?
Training the machine learning model is a crucial step. We start with survey data where boulders have already been identified – this serves as our “ground truth.” The AI learns from these examples, understanding the shape, texture, and other characteristics that define a boulder on the seabed.
The model continuously improves as we feed it more data, learning to detect boulders in different seabed conditions and across various survey types. We also validate its predictions y comparing them against expert interpretations to refine its accuracy.
What challenges did Ternan face as a small to medium-sized company in developing this advanced AI technology, and how did partnerships help overcome these obstacles?
One of the biggest challenges was access to AI expertise and computational and human resources. Unlike large corporations with dedicated AI teams, we had to be smart with our approach. This is why we contacted UKRI and research institutions like the Hartree Centre, where we could leverage their expertise and infrastructure.
Through this process, we’ve developed a new competency within Ternan – AI-driven geological interpretation. We’ve built the knowledge and external partnerships necessary to push forward with more AI initiatives in geoscience. By collaborating, we’ve built an innovative AI tool without needing an in-house AI team. This proves that small and medium-sized companies can innovate AI by creating the right partnerships.
The tool shows green markers for training data and red for AI-identified boulders. Can you explain what this visualisation reveals about the tool’s learning and detection capabilities?
The green boxes indicate the boulders that were used to train the model, while the red markers represent predicted boulders. This visualisation helps illustrate the AI’s learning process. Additionally, the presence of predicted boulders in areas without training boulders suggests that the model has identified potential boulders that might have been overlooked in manual interpretation. This demonstrates the AI’s ability not only to replicate expert analysis but also to enhance it by uncovering additional features in the data.
How does this project demonstrate Ternan’s commitment to innovation and pushing the boundaries of AI in geological surveying?
Innovation is at the heart of what we do at Ternan. This project showcases our ability to take on cutting-edge challenges and develop practical solutions that improve seabed classification. We’re not just following industry trends – we’re helping shape them.
This is just the beginning. We’re building on our knowledge, strengthening our expertise, and pushing for even more advancements in AI-driven geoscience workflows. We speak to and listen to our clients about what they need and we go and work on the best solutions.
Data donations for improving the tool. What type of data are you looking for, and how can potential contributors participate?
Machine learning thrives on data – the more diverse, the better! We’re inviting contributions of seabed survey data to help refine our model. The data can be supplied anonymised and without geographic coordinates. All contributions will go toward improving detection accuracy across different seabed environments.
If you have datasets and are interested in contributing, we’d love to hear from you. Reach out to us info@ternan-energy.com, and let’s explore how we can collaborate to advance this technology together.
How do you see this AI tool complementing, rather than replacing, existing workflow processes in geological research and exploration?
We’re not here to replace workflows but to enhance them. AI is a powerful tool, but it works best alongside expert human interpretation. By integrating AI into existing processes, we can automate time-consuming tasks, allowing geoscientists to focus on higher-value analysis and decision-making.
Think of it as an advanced assistant – speeding up detection, reducing subjectivity, and providing another layer of validation to traditional workflows.
What are the next steps for this technology? Are there plans to expand its application to other geological or marine research areas?
While our focus has been seabed boulder detection, the principles behind this tool can be applied to other repetitive and time consuming interpretation challenges. In offshore wind, it could help with sediment classification and geohazard mapping. In CCUS, it could assist in monitoring CO2 storage sites. In oil and gas, it could enhance fault detection and reservoir characterization. AI-driven geoscience is an exciting field, and we’re eager to push the boundaries even further and talk to Universities and develop projects with their academics and students!
If you would like to know more about our projects please email info@ternan-energy.com we’d love to discuss this in more detail. To find out more about our services please visit www.ternan-energy.com