Research
Academic & Applied Research
Blending scientific rigor with practical delivery across geoscience, computer vision, and automation labs.
R1
Geochemical rock classification powered by deep learning
University of Melbourne • Melbourne
Feb 2023 - PresentGeochemistry × AI
Fusing petrology expertise with transformer-based representation learning to unlock richer mineral insights.- Built a multi-branch Transformer that blends ICP-MS, XRF, and contextual stratigraphy for 14% accuracy gain over classical baselines
- Implemented uncertainty-aware labeling and active learning loops to prioritize expensive lab assays
- Released reproducible evaluation suite covering 40+ real-world deposits, ensuring trustworthy deployment reports
R2
Road crack detection via consumer-grade drones
Beijing University of Civil Engineering and Architecture • Beijing
Sep 2018 - Oct 2022Infra-vision ops
Designed an accessible drone+AI toolkit so municipalities can audit pavements without million-dollar rigs.- Engineered a DJI Mini-based imaging rig with adaptive flight plans, cutting survey labor by 60%
- Trained a lightweight MobileNetV3 crack detector with temporal filtering to keep on-device inference under 40 ms
- Curated and annotated 12k UAV frames, introducing pixel-level severity scoring for downstream maintenance teams
↑