Which tool do you prefer to use for classification of remote sensing data, e.g. classifying land use, and why?
Which other tools have you tried, and why did you decide against them?
1
-
for a dedicated remote sensing stackechange: area51.stackexchange.com/proposals/59346/remote-sensing– WAFCommented Aug 28, 2013 at 10:14
Comments disabled on deleted / locked posts / reviews
|
9 Answers
I use a number of tools depending on the type of classification I am trying to perform.
For general unsupervised/supervised classification I use ENVI, which has many options for classification methods (including some more advanced methods using neural networks and support vector machines). It is very easy to extend ENVI using the IDL programming language, and I have found that this often simplifies post-classification analysis (as you can write your own code to do this if needed).
If I want to perform object-based classification (which involves segmenting the image into objects and then classifying these objects, the benefits being that you can use aggregated properties of the objects such as means of the bands, shape and texture) I use eCognition, although I have also heard that ENVI EX is good if you don't need the power of eCognition.
If you're looking for free software then Opticks has a number of options for classification, although I've never got on very well with Opticks. Also, Spectral Python is a very nice tool that allow you to load images into NumPy arrays in Python and then process them. It includes a module containing various classification methods, and is very easy to extend.
For an open source GIS solution, see here: http://grass.osgeo.org/wiki/Image_classification It includes a small tutorial as well.
My favorite discovery this year has been the Orfeo Toolbox and the associated program: Monteverdi.
http://orfeo-toolbox.org/otb/monteverdi.html
Lots of options for Remote Sensing work and very helpful documentation. Oh, did i mention it is free and o-s
Enjoy, sa
I just saw this post on QGIS forum and thought I would place here.
Hi all.
Sorry for crossposting. As some of you know, the r.li suite of GRASS commands allows landscape analyses. Its interface is rather complex, and is still in TclTk, not ported to either wxpython or qgis. As such, it is now more difficult to use than it should be, and it will become unusable when TclTk support will be dropped. The possible solution (thanks Radim) is to rewrite the interface as a qgis python plugin. It should not be a huge work (we provisionally estimate 2-3 weeks).
The question is: is there anybody willing to invest either his/her time, or some money, to write such a plugin?
We (Faunalia) would be happy to help if necessary.
All the best.
Qgis-user mailing list [email protected] http://lists.osgeo.org/mailman/listinfo/qgis-user
answered Nov 12, 2010 at 18:10
-
I know, this is an old post. But anyway... The statement that the
r.li.*suite will become unusable when TclTk support will be dropped isn't quite true! One can, and will be able to, use the tools via the new -- actually current? -- (wx)GUI as well as via GRASS' shell. Yes, in both the current version (6.4) as well as in the upcoming GRASS-GIS 7. Commented Nov 15, 2012 at 20:37
I tried Erdas Imagine and ENVI softwares, and feel unable to say which one is the best. Both can classify your imagery using supervised and unsupervised methods.
Also take a look in SPRING software, made by Brazil's National Institute for Space Research (INPE). Not sure if it's open source but It's definitely free.
answered Jun 1, 2012 at 13:35
I have used Erdas Imagine,ENVI ITT,Idrisi Selva,PCI Geomatica. ENVI has IDL extensions which provide you to drive advanced classification algorithms such as SVM,ANN,DT,etc.Idrisi Selva has quite good classification algorithms on both supervised and unsupervised,especially on neural networks(SOM,MLP,RBF,FuzzyART) .I have also a little experience on Monteverdi,Orfeo Toolbox.It is very user friendly software. MultiSpec has also classification algorithms for images
I have no preference yet (haven't tried any FLOSS alternatives), but I've tested Feature Analyst, a plugin for Arc*. While inferior to e-Cognition, it has a low entry barrier. It's simple to use and offers a nice interface for supervised classification. You can use various "brushes" as the main detection unit, but that doesn't affect the result as much as one would expect. It also has a batch mode, but in my case it was useless, since the rasters needed individual training sample tweaking to give good results.
answered Jun 1, 2012 at 12:55
-
I am not an advocate of eCognition of Feature Analyst. However, your statement of FA being "inferior" is completely unsupported and subjective. Given that FA is a feature extraction algorithm and eCognition is focused on image segmentation they are completely different models with different applications. It may be that FA did not perform in your given application but this does not mean that it would not perform well in a different analysis. We have had good performance with FA in situations that eCog would perform poorly. Commented Jun 16, 2014 at 15:51
-
Unsupported? The question is about classification, for which FA has or had far fewer knobs and options than the other. Clearly things could have changed in these five years, but such a magnitude would be unlikely. Commented Jun 18, 2014 at 21:45
I tried Erdas imagine and has done classification. But if the rulesets are given correctly in e-cognition it produces a better output than erdas. But the development of rulesets are a bit complex in e-Cognition Developer.
