Thesis subject

MSc thesis topic: Object Detection Hycaniths as Proxy for Plastics

Water hyacinths are floating plants that accumulate plastics in rivers, as shown in the attached images. Last year, a GRS Master Students travelled to Bangkok, Thailand, to take pictures and on-site measurements of the concentration and extent of plastic pollution in the Chao Phraya river and their connection to plastic pollution in the area.

This thesis is about processing the data with a deep segmentation model to create an automated pipeline to detect Hyacinths and Plastics on both ground images, taken from bridges, and UAV images. Finally, the detections from the acquired image data should be compared to satellite-based detections of hycanths using Sentinel-2 and Planetscope images. In terms of deep learning, it involves using and fine-tuning panoptic segmentation models (both ResNet-based and ViT-based) and comparing different model architectures and pre-trained models.

Relevance to research/projects at GRS or other groups

This Thesis is supervised by Marc Rußwurm who will provide expertise in machine learning and computer vision and Tim van Emmerik with expertise in river plastic monitoring

Research Questions

  • How accurately can we identify hyacinths and plastics from ground and UAV imagery?
  • How accurately can we estimate hyacinth density from Sentinel & PlanetScope imagery

Objectives

  • Train a panoptic segmentation model to identify both hyacinths and plastics from ground (bridge) and UAV images
  • Compare the results with a remote-sensing classifier that monitors rivers in Bangkok with PlanetScope and Sentinel-2

Requirements

  • required: deep learning course
  • recommended: interest reducing plastic pollution in marine environments

Expected reading list before starting the thesis research

  • Schreyers, L. J., van Emmerik, T. H., Bui, T. K. L., Biermann, L., Uijlenhoet, R., Nguyen, H. Q., ... & van der Ploeg, M. (2024). Water hyacinths retain river plastics. Environmental Pollution, 356, 124118.

Theme(s): Sensing & measuring; Integrated Land Monitoring