I participated in the Reflection Connection: Bringing New Algorithms to Old Data challenge organised by ThinkOnward, an innovation lab focused on accelerating energy innovation. The competition centred on applying machine learning to seismic data analysis, specifically using computer vision techniques to understand and match features within seismic images.
This challenge was situated at the intersection of geophysics and computer vision and required developing vision models capable of learning meaningful representations from seismic imagery. The objective was to explore how far modern classification and image-matching algorithms could be pushed when applied to legacy seismic data, a critical resource in oil and gas exploration and subsurface interpretation.
A key requirement of the competition was the development of a one-shot learning solution. The model needed to generalise beyond the training data and correctly identify and match seismic features it had not previously seen. To assess this, solutions were evaluated on additional categories of seismic features outside the original training set and reviewed by a geoscience judging panel, placing strong emphasis on robustness and generalisation rather than memorisation.
Out of 147 participants, I developed the top-performing one-shot solution and achieved first place in the competition. The resulting model demonstrated the ability to extract and generalise structural patterns from seismic images, highlighting the potential of computer vision techniques to support seismic interpretation workflows and improve understanding of subsurface features using existing datasets.
Project Link:
ThinkOnward Reflection Connection – Computer Vision for Seismic Data Analysis in the Energy Sector
Leave a Reply