I am new to weka and am trying to make landuse / landcover maps of a particular city through data mining, so that I can automate the process. and can make landuse / landcover maps automatically. Could someone please suggest how to go about it?
hello Carola, I want to build an automatic classification system which can produce maps of landuse and landcover automatically after providing some satellite image as an input, I want to build this system using some open source software of image classification, for which i have identified WEKA software. so in this regards i hope i have made myself clear, could you please suggest some way to do that as manual doing classification for over 100 satellite images is a tedious work. if possible please help
with the 'Knowledge Flow' you can automate your process - I don't know in which format your data were avaible, you have to decide yourselve. And I hope, it's clear for you, that you have to convert your existing dataset into .arff format
The main challenge in automating the land use classification is the labelling process. When you produce a classified map, you will need to label each class in your map. for example, the system you build (unsupervised system) produces a number of classes from the input image without knowing what those classes are. you need to determine suitable labels for them based on some classification schemes (i.e. Anderson scheme). In addition, the other challenge would be how you determine the exact number of classes you have in your image. based on the above discussion, I suggest to go for semi-supervised systems. in this system part of the process is automatic and the others are manual process. this system is widely used and suggested in the literature.
coming to your question, Weka software can be used to classify attribute data not satellite images. but you can built some systems based on decision tree algorithm then you can use with satellite images in other software such as envi or ArcGIS. however, all these depends on the data you want to use. for some data other methods are suggested. in general, urban and very high resolution, SVM method is recommended.
I think building a classification system for general purpose is quite complex and the literature suggests developing systems for specific applications. Hope the question is answered.