图像分类精度评价.pptVIP

Accuracy Assessment: Quantifying Example: Total accuracy was 71%, but User’s accuracy for hardwoods was only 48% Class types determined from reference source User’s Accuracy Class types determined from classified map # Plots Conifer Hardwood Water Totals Conifer 50 5 2 57 88% Hardwood 14 13 0 27 48% Water 3 5 8 16 50% Totals 67 23 10 100 Producer’s Accuracy 75% 57% 80% Total: 71% * Accuracy Assessment: Quantifying What to report? Depends on audience Depends on the objective of your study Most references suggest full reporting of error matrix, user’s and producer’s accuracies, total accuracy, and Kappa * Accuracy Assessment: Interpreting Why might accuracy be low? Errors in reference data Errors in classified map * Accuracy Assessment: Interpreting Errors in reference data Positional error Better rectification of image may help Interpreter error Reference medium inappropriate for classification * Accuracy Assessment: Interpreting Errors in classified map Remotely-sensed data cannot capture classes Classes are land use, not land cover Classes not spectrally separable Atmospheric effects mask subtle differences Spatial scale of remote sensing instrument does not match classification scheme * Accuracy Assessment: Improving Classification Ways to deal with these problems: Land use/land cover: incorporate other data Elevation, temperature, ownership, distance from streams, etc. Context Spectral inseparability: add spectral data Hyperspectral Multiple dates Atmospheric effects: Atmospheric correction may help Scale: Change grain of spectral data Different sensor Aggregate pixels * Accuracy Assessment: Improving Classification Errors in classified map Remotely-sensed data should be able to capture classes, but classification strategy does not draw this out Minority classes swamped by larger trends in variability Use HIERARCHICAL CLASSIFICATION scheme In Maximum Likelihood classification, use Prior Probabilities to weigh minority classes more * Accuracy Assessment: S