Research

Research and
technical experiments.

Academic research, conference work and the experiments behind my software, mostly on sonar, the seabed and applying machine learning to survey data.

Academic research

2024

Automated boulder detection on side-scan sonar mosaics

Master’s capstone at Memorial University’s Marine Institute: training a YOLO-based deep-learning detector on boulders in a NOAA side-scan backscatter mosaic. Supervised by Dr. Sean Mullan and Paul Elliott.

  • Computer vision
  • YOLO
  • Side-scan sonar
2015–2016

Machine learning for coastal flood risk and shoreline change

Research at the Department of Marine Science & Technology, Federal University of Technology Akure: Python machine-learning models built from satellite imagery and elevation models, with inundation and risk maps produced in ArcGIS and QGIS.

  • Machine learning
  • Remote sensing
  • GIS

Conference presentations

  • Canadian Hydrographic Conference 2024, St. John’s Convention Centre. Poster: “Automated Boulder Detection on Side-Scan Mosaics using YOLO-based Deep Learning Model”. Read the research summary.

Applied research in software

Much of my current research happens inside Dolphin Explorer, the marine survey software I build:

  • Shadow-based boulder detection. A rule-based detector that models a boulder by its acoustic shadow and measures the face in front of it, working at a cell size tied to the smallest boulder wanted so speckle averages out.
  • Trainable detection. Project-local RetinaNet models trained from labelled datasets in YOLO or COCO format, currently being evaluated against the rule-based detector.
  • Bottom tracking. A single seabed detector shared by the waterfall, the processing graph and slant-range correction, with a confidence score on every pick for quality control.

Technical writing and experiments

Research interests

  • Automated detection and classification of seabed objects in side-scan and sub-bottom data.
  • Processing, quality control and evidence trails for marine survey data.
  • Machine learning that supports, rather than replaces, the interpreter’s review.
  • Geospatial data infrastructure, from survey data to property and location data.

For research collaboration or to discuss any of this work, get in touch.