Remote Sensing
Photo-interpretation
Human visual reading of imagery to classify land cover and change, using trained judgement rather than automated rules.
Photo-interpretation relies on a trained analyst examining imagery for texture, shadow, shape and context. It captures things algorithms miss, such as distinguishing a plantation from natural regrowth.
Its weakness is subjectivity. Good practice requires written interpretation keys, multiple interpreters on a subsample, and a documented disagreement rate so the resulting labels can be defended.
Frequently asked questions
Why use people instead of machine learning?
Because reference labels for training must come from somewhere, and expert reading remains the benchmark for ambiguous classes.
How is consistency checked?
By re-reading a subsample and reporting agreement rates between interpreters.