Computer Vision (Analyst) · Level 301
Computer Vision Analyst 301 — Your Brief
Design a change-detection study, audit subclass reliability, or scope field validation — deliverable is a defensible memo, not a leaderboard score.
Computer Vision (Analyst) · Level 301
Computer Vision Analyst 301 — Your Brief
Design a change-detection study, audit subclass reliability, or scope field validation — deliverable is a defensible memo, not a leaderboard score.
Computer Vision Analyst 301 — Your Brief
Prerequisite: Computer Vision Analyst 201.
No step-by-step ML cells. You choose a narrow monitoring question and deliver a memo a ranger or Landscape Recovery partner could use.
Want to train and compare models in code? → Computer Vision Developer 301.
Module 1 — Your Brief
Manager email
Pick one project below. Deliver
report/analyst301_brief.md(500–700 words) and one chart or diagram (can be hand-drawn workflow, confusion matrix from 201, or a map sketch). Write for rangers, not ML researchers.— Andy
Choose one
Option A — Change detection scoping (recommended)
Design a 2010 vs 2020 land-cover change study (Turing DSG brief).
Which classes would you track? How handle different flight dates? What accuracy
makes results actionable for Landscape Recovery? Optional: sketch a validation plan.
Option B — Subclass reliability audit
Which subclass confusions (heather moor vs rough grass, wet flush vs pasture)
would most affect grazing vs woodland expansion monitoring? Use published
project accuracy ranges (72–92% subclasses) and the 2023 press release.
Option C — Field validation plan
A colleague has a 90% accurate pasture map. Design which fields you would visit
and what you would record to verify it before a grazing decision.
Option D — Fragmentation memo
Why does wet grassland fragmentation matter for conservation (Remote Sensing paper)?
Propose one simple metric rangers could understand (e.g. number of patches above a size threshold).
Download the brief template.
What Andy will look for
- A narrow, answerable question tied to park monitoring.
- Honest limitations (EuroSAT ≠ APGB, season, geography).
- A recommendation someone could act on with stated checks.
- Clear prose — no jargon without explanation.
Module 2 — What You Have Learned
You have completed the Computer Vision Analyst track:
- Analyst 101 — Reading the map
- Analyst 201 — Trusting the model
- Analyst 301 — Your brief
Related developer path (implement convolutional neural networks (CNNs) in Python):
Operational data: github.com/pdnpa/cnn-land-cover.