Hello, world (and what's coming)

Starting this blog as a public notebook. Partly for myself, since writing forces clarity. Partly because I’m in the final stretch of an M.S. in Remote Sensing at NCU and looking for fully-remote ML / CV / Earth Observation roles.

What I’m working on right now

My thesis sits at the intersection of synthetic aperture radar (SAR) image processing and deep-learning object detection. Specifically, I’m comparing how different speckle-noise reduction techniques (Lee, Refined Lee, Frost filters) affect ship detection accuracy. The Lee-filter integration I proposed for torchgeo (PR #3655) is the first step toward making these preprocessing tools routinely available in the PyTorch geospatial ecosystem.

Posts coming in May and June

• A walkthrough of the Lee filter as a Kornia transform: math, edge-preservation behavior, and why I wrap it as IntensityAugmentationBase2D.
• Building a local-LLM RAG pipeline: how stock-rag glues TWSE data, news, and Ollama qwen2.5 into per-stock weekly reports without burning OpenAI credits.
• Gesture controls with MediaPipe Hands: the cooldown and pinch-distance heuristics behind ninjapause.

How to reach me

GitHub: box1401 · LinkedIn: kade-chen-rs · email: abcd140224 [at] gmail.com

Happy to chat about SAR, geospatial PyTorch, or remote-first engineering.