Landsat decision tree land cover classification

[Cactuz]

Hi guys!

 

I`m working on land cover change detection project of Kyiv province in Ukraine using Landsat data and ancillary information. I want to find change in Land Cover from 1990 to 2014 of Kyiv province using Decision Tree approach in ENVI.

 

I have already done radiometric correction, atmospheric correction and apply cloud mask to whole Landsat`s scenes, and now I have question and request.

 

Do I need to do mosaicking before classification or mosaicking classified image ?

 

Does anybody have some decision rules for splitting pixels for appropriate land cover classes using Landsat data, DEM, VI?

 

Thanks

[mamadouba]

Cactuz,

 

Decision rules are arbitrary unless they are derived using a statistical function like recursive binary partitioning.  There's not a one-size-fits-all rule for specific cover types.  The more well known programs for classification and regression trees (CART) for remote sensing purposes are See5 and Cubist by RuleQuest.  Idrisi has a built-in Decision Tree classifier dating back to Andes edition.  If you are using ENVI, you can apply some of the built-in non-parametric classifiers such as Support Vector Machine (SVM) or Neural Net.  These are machine learning algorithms capable of handling both continuous and categorical data.

 

If you have your mind set on decision trees, then you can explore some of the free and good options.  A developer on Google Code implemented C45 (free source code of See5) in IDL.  This allows you to create decision rulesets using the See5 approach and the output is in the ENVI Decision Tree format.  Just choose the option in ENVI > Classification > Decision Tree and  "Execute Existing Decision Tree."  I've used this program myself and the results are impressive. The developer created a decent GUI if you aren't comfortable with IDL command line.

 

You can find the code here:

https://code.google.com/p/c45idl/

 

 However, my favorite free program is the R implementation of RandomForests.  This is an extremely robust classifier.  It requires knowledge of R, plus dependent libraries.  You can download the randomForests script here:

 

https://bitbucket.org/rsbiodiv/randomforestclassification

 

....and there are installation instructions and tutorials here:

 

https://bitbucket.org/rsbiodiv/toolsforr

http://www.whrc.org/education/indonesia/pdf/DecisionTrees_RandomForest_v2.pdf

 

Good luck