hariasa Posted May 7, 2012 Report Share Posted May 7, 2012 Hi, Does anyone know how to perform a RandomForest classification? Or which programs support it? Are there example scripts for R? Or plugins for ERDAS, ENVI or something? I use shapefiles for training areas and tif files for the actual data to be classified. All help would be awesome. Thanks, -hariasa Quote Link to comment Share on other sites More sharing options...
3dbu Posted May 7, 2012 Report Share Posted May 7, 2012 http://www.google.com.ni/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0CHkQFjAA&url=http%3A%2F%2Fwww.webchem.science.ru.nl%2FPRiNS%2FrF.pdf&ei=WCGoT7KLIImu9ASm6fzBAw&usg=AFQjCNFI8epM3ssKUuH3yuOuOq9yZFzrxw may be usefull Quote Link to comment Share on other sites More sharing options...
nomlas Posted June 27, 2012 Report Share Posted June 27, 2012 Hi you can run randomforest in R , using the randomForest library or alternatively if you prefer you can download the rattle GUI (rattle.togaware.com), your shapefile containing your training data should have the values from your tiff file extracted then you can use the dbf file in rattle. there is a tutorial on the togaware webpage. There is also a library that creates maps based on your RF analysis called modelmap Recently, I came across a stand alone IDL based remote sensing software called EnMAP-Box that allows you to carry out a random forest classification/regression , you check out the software at Environmental Mapping and Analysis Program Documentation also on the website The paper below explains the use of the software imageRF – A user-oriented implementation for remote sensing image analysis with Random Forests Environmental Modelling & Software, Volume 35, July 2012, Pages 192-193 Björn Waske, Sebastian van der Linden, Carsten Oldenburg, Benjamin Jakimow, Andreas Rabe, Patrick Hostert or you can try--> openmodeller.sourceforge.net Thanks Quote Link to comment Share on other sites More sharing options...
sijooss Posted June 28, 2012 Report Share Posted June 28, 2012 i think Feature space based classification will be far better and will give better outputs since i t will account for each pixel values of the image. Quote Link to comment Share on other sites More sharing options...
nomlas Posted June 28, 2012 Report Share Posted June 28, 2012 @ sijooss Feature space based classification are rather simplistic when compared to ensemble methods like random forest, boosting or bagging. A number of recent multispectral and hyperspectral studies have confirmed this. Due to the nature of the random forest , the pixel values are considered are number of times as a result of the bootstrapping aggregation, making the algorithm very robust and accurate. Additionally, by considering different variables (i.e bands) for creating the trees, an additionally measure of randomness is introduced. hence the more diversity thta is introduced into the algorithm the more powerful it becomes. Thanks Quote Link to comment Share on other sites More sharing options...
hariasa Posted June 30, 2012 Author Report Share Posted June 30, 2012 Eventually, I performed the RandomForest method with R. It was quite easy to program, with the help of a few tutorials on the R help. I have not tried EnMAP yet, I will do so now and post here how it went. Having a lot of variables in R with very big imagery does not always go well... sometimes randomforest in R 'forgets' to classify certain things the first time. It's also quite RAM intensive, so if anyone else wants to do it, you need at least 4 GB of RAM to do a decent random forest classification with R for a decent image. Thanks nomlas for your help! +1 Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.