Research Internship Satellite Applications Catapult: Remote Sensing and Deep Learning for Polluting Cement Plant Detection
Date:
DISCnet Doctoral Training Consortium — Invited Presentation
This invited presentation summarised my research internship at the Satellite Applications Catapult, undertaken as part of my DISCnet scholarship. Alongside presenting the development of deep learning methods for detecting polluting cement plants using multi-temporal Landsat-8 imagery, I discussed how to scope and deliver a research project within a short industrial placement, collaborate effectively with non-academic partners, and translate academic methods into applied remote sensing workflows.
Event: DISCnet Doctoral Training Consortium
The presentation covered:
- What makes cement plants distinguishable in satellite imagery.
- Construction of a global database of cement production assets.
- Creation of 1 km2 Landsat-8 chip grids across multiple months.
- Use of thermal infrared (Bands 10 and 11) and short-wave infrared (Bands 6 and 7).
- Deep learning architectures: VGG13, ResNet, EfficientNet.
- Transfer learning and class weighting for imbalanced datasets.
- Multi-run evaluation across stratified splits.
- Results demonstrating the feasibility of automated cement plant detection.
The talk also reflected on the benefits of the research placement:
- Collaboration with industry and academia.
- Communicating technical work to non-experts.
- Applying deep learning to diverse remote sensing datasets.
- Building a research portfolio and producing publishable work.