Classifications Flashcards
What is the process of orthorectification?
Process of removing geometric errors within images
Projection is as if every point looks as if an observor were looking straight down at it to earth
Influenced by camera, perspective, and topography
What does orthorectification rely on?
Digital Surface Model
Quality significantly affected by the uncertainity in the DSM
What is image classification?
All pixels assigned to a particular class
A digital image classification may use _ to classify each individual pixel
Spectral based classification (spectral information)
Explain unsupervised classification
Spectral classes grouped first based solely on spectral information in the data (numerical)
Analyse will specify how many classes are to be looked at
Unsupervised classification is an…
Iterative process
Looks for seperation distance among and variation within each cluster
ISODATA algorithm is a refinement of the K-Means algorithm. What are these 4 refinements?
1) clusters with too few members = discarded
2) clusters with **too many **members = split into new groups
3) clusters too large to disperse = split into new groups
4) two clusters centres close together = merged
Explain supervised classification
Sample areas are selected for identifying different surface cover of interest
Analyst familair with the area and knowledfe of the actual surface cover types
Why is supervised classification good?
An analyst can supervise the categorisation of a set of classes, training the computer to recognise spectrally similiar areas for each class (in signatures) with the spectral pattern of each pixel compared to the signature
What are two common classifers?
Minimum Distance to Means
Maximum Likelihood
What is the first value to be determined in each band for each each category?
Mean spectral value
What happens to unknown identitifed pixels?
Minimum distance to means
Assigned to closest class
Computing distance between the value of the unknown pixel and each of the category means
Not used if the spectral classes have high variance
What happens to unknown identified pixels?
Maximum Likelihood
Assigned to the most likely class (indicative of the highest probability value)
(unknown if the probability values are below a threshed set by the user)
For Data Validation
Overall error =
Omission (test result wrongly indicates absence) error =
Commission error (test result wrongly indicates presence) error =
1 - overall accuracy
1 - producers accuracy (perspective of map producer)
1 - users accuracy (perspective of map user)
Row for Users Accuracy
Coloum for Producers Accuracy
Kappa statistic serves as…
An indicator of the extent to which the percentage correct values of an error matrix are due to TRUE agreement v CHANCE agreement
