10MinimumDistanceandTexturalClassificationLectureNotes

10MinimumDistanceandTexturalClassificationLectureNotes -...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon
GISc 4037 Digital Image Analysis Dr. Charles Roberts The Minimum Distance to Means Classification A Review of the Supervised Classification Process 1. An appropriate classification scheme must be adopted. This is a function of the particular goals of a particular project. 2. Representative training fields must be selected, including an appreciation for signature extension factors if possible. -Training fields should be nearly homogeneous in tone -Should be typical, not unusual, sample of land-cover -Number of pixels in each training field should be upwards of 10 x the number of bands being used. -Training fields should be representative of every spectral signature on the image. Spatio-temporal considerations Temporal: wet, dry, winter, summer: how do spectral signatures change through the seasons, and does this consideration affect your project? Space: Do subtropical hammocks in south Florida resemble those in the Caribbean? Do mangroves in Palm Beach County resemble those in Australia? 3. Statistics must be extracted from the training field spectral data. 4. The statistics must be analyzed to select the appropriate features to be used in the classification process. 5. The appropriate classification algorithm must be selected. 6. Classify the image into X classes. 7. Statistically evaluate the classification accuracy. An Ideal Sequence for the Selection of Training Fields when you can ground truth your training fields: Not to be confused with ground truthing your classified image 1) Assemble ancillary data: maps and aerial photographs 2) Examine the image, and get a sense of landmarks that you can
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
find on the ground during your ground truthing visit. 3) Run an unsupervised classification on the image, specifying many clusters, and small cluster radiances. 4) Take and evaluate training fields 5) Design and conduct field work in the areas where you will take training fields. 6) Proceed with Classification Minimum Distance to Means Classification Advantages of this classification program: Slower than parallel piped, faster than maximum likelihood Good to use when the number of pixels in each training field is small, or when the training fields are not very well defined Unlike parallel piped, will allow user to classify entire image, no holes. How Minimum Distance Classifier Works The User provides a mean vector in the form of a training field for each class. The program takes the mean vectors, and goes through the image, pixel by pixel, calculating a
Background image of page 2
Image of page 3
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 12/21/2010 for the course GIS 4037c taught by Professor Roberts during the Fall '10 term at FAU.

Page1 / 6

10MinimumDistanceandTexturalClassificationLectureNotes -...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online