If you intend to classify different surfaces within an urban body based on images with a spatial resolution of 30 meters, you have limitations due to mixed pixel information. That means, many pixels contain more than one distinct class (vegetation, roof/house, sealed [roads and open places]...).

One effective way to overcome this is the spectral unmixing method. It is comparable to a supervised classification but instead of assigning one class to each pixel, the algorithm analyzes the signatures of each pixel, compares it to the signatures of trained classes, and gives a fractional value of each trained class within this pixel, e.g. 0.5 water means that the pixel is covered by water by 50%.

You can use ESA SNAP software for training and unmixing your images (http://step.esa.int/main/toolboxes/snap/). If you have ERDAS IMAGINE, you can use the subpixel classification module which does basically the same.

• https://pdfs.semanticscholar.org/8d5a/3a134e3340b1754d5608080d9b213c56dd8b.pdf
• http://kirj.ee/public/Ecology/2007/issue_1/bio-2007-1-2.pdf

Hi dear

If I were you I would use ASTER (15m) or Hyperion images instead of Landsat. There are more accurate. But remember that ASTER doesn’t have any band in blue region. After atmospheric correction use pan-sharpening method to make pixels more useful.

Hope it can be useful for you.

Have a good time.

maximum method in Arc gis

As already stated, it is difficult to achieve urban classification satisfactorily using 30m. But pan sharpening would improve the resolution and may also improve the result of classification when combined with spectral unmixing method.

Best of luck

Thank you Andreas Braun and Glory Enaruvbe for your detailed and effective recommendations. I will check these out and hope that it will find some satisfactory result. :)

You may also considering super-resolution method.

I think it depends a lot on what you are trying to get as a result..If you are trying to make fine distinctions btw. different land covers within urban areas, then you'll definitely want to try to spectral mixture analysis approaches; but if you're just trying to differentiate urban from non-urban areas, I suggest starting by experimenting with scenes from different times of year, and using several of the urban/built- up area indices to see how much confusion there is, and which land covers are being confused. That may lead you to combining approaches, eg. identifying non-urban areas first with NDVI, by using imagery from the middle of the growing season, and then masking out those areas when applying the urban area index.................Al ot depends on the color of bare soil (eg. how similar to the built up areas) and on the nature of vegetation at the time you are studying the urban scenes...

If you have the option to use ENVI you can try the Decision Tree classification method. It gives you plenty of options to control the thresholds of band ratio for each land use class. You can select a few training polygons and know what the spectral properties are for your polygons, then set up a decision tree to cluster the land use classes which match your own defined thresholds. Whenever you want to carry out an unsupervised classification I believe decision tree algorithm's are very useful and give convenient upshots.

Thanks.

Dear, there are many algorithms to do classification of urban area and the results will be great such as SVM, KNN and other but my question to you what is your data type?

regards

@Hayder Dibs

I have both freeware Landsat and Santiniel-2 satellite data. Can you help me with some basic tutorials (e.g. video, documents) to step by step understand the algorithms.

Dear Haque, Pixel based classification by using SVM algorithm has the best results.

Article Landsat 8 vs. Landsat 5: A comparison based on urban and pen...

You can use ENVI for the classification with SVM algorithm.

Best of luck

Thank you Mr Susanta Mahato.

Haque: here are some references regarding different index-based urban classification:

Zha, Gao, and Ni. Use of normalized difference built-up index in automatically mapping urban areas from TM imagery. Int'l Journal of Remote Sensing , 2003, Vol 24, No 3, 583–594 ;

Abd. Rahman As-syakur 1,*, I Wayan Sandi Adnyana 1,2, I Wayan Arthana 1 and I Wayan Nuarsa . Enhanced Built-Up and Bareness Index (EBBI) for Mapping Built-Up and Bare Land in an Urban Area 1 Remote Sens. 2012, 4, 2957-2970

My experience is that neither supervised nor unsupervised classification is able to differentiate well btw. the varied surfaces of urban areas from similar looking areas outside of the urban zones. A lot depends on time ofyear, local soil color, type of vegetation, etc. Best wishes.

Have you tried object based classification yet? Though mainly used with high resolution satellite images, object based image analysis can be used with medium resolution satellite images from sensors such as Landsat 8 OLI to classify urban areas with fair amount of accuracy.

Dear it is based on what u need , and there are a huge classifiers but do you need to use pixels or object classification? any way you can use SVM, KNN classifies to get a good result

My regards

Dear Md. Inzamul Haque this what you request step by step just follow these videos

https://www.youtube.com/watch?v=S456WtK-tyg

https://www.youtube.com/watch?v=HSWLvSOWefw

you are so welcome

Hello, Haque

You can use several indices for the classification of urban areas in Landsat images. About subject is add the attechment file.

  https://www.researchgate.net/publication/326634695_the_relationship_of_urban_expansion_landscape_patterns_and_ecological_processes_in_denizli_turkey

Interesting and good Discussion