May I know for Landsat 8 I have to do the Pan-sharpening using its 8th panchromatic band? I have to do this segmentation using ecognition Software, and my ultimate aim is to segment and classify the urban features especially buildings when we are doing segmentation in ecognition it will ask layer weightages, Scale, Compactness, Shape values. Can anyone help me to do this process?
Dear Safwan,
Why do you have to use ecognition? Ecognition is better used with high and very high resolution images (IKONOS, Quickbird, Geoeye...etc), Landsat 8 and 7 panchromatic image is only 15 meters.
Ecognition uses different segmentation methods (Quadtree, multiresolution, nearest neighbor ...etc). But, as it shows and based on several experiments that it is not better than any known pixel based segmentation technique for Landsat 8 and 7. In addition, it depends on how the use is supervising the segmentation process.
I suggest to use other techniques such as Fuzzy C-means.
A little precision: TM sensor was on board of Landsat 4 and 5.
In the rest suggestions I agree to Mr Awad.
As such pan sharpening is not essential. Segmentation in eCognition is possible for multi-source data. Multispectral and pan data can be input for the segmentation. You can give equal weightage for all the bands, however, higher weightage to SWIR band is preferred separate spectral mixing of urban and fallow lands.
Dear Safwan,
In eCognition software generally ideal parameters such as compactness, shape factor, and band weight are detected using a small sample plot with user experience. Just try in a small part of your study area to find the best. However, I suggest you to use the NDVI and surface texture data in addition to the landsat wavebands. Because, NDVI refers the vegetative lands and urban cover is not vegetative so that you can use to seperate built up land from other land use cover.
Hi – I suggest to be a bit more careful with statements such as “… band weight are detected using a small sample plot with user experience.” This is what used to be done. A lot has been happening in OBIA research, and methods exist now to help determine ideal segmentation parameters objectively. See for example the following papers:
- On ESP (estimation of scale parameter):
o Dragut L, Tiede D, Levick SR. ESP: a tool to estimate scale parameter for multiresolution image segmentation of remotely sensed data. International Journal of Geographical Information Science. 2010;24(6):859-71. PubMed PMID: ISI:000276747400004.
o Dragut L, Csillik O, Eisank C, Tiede D. Automated parameterisation for multi-scale image segmentation on multiple layers. ISPRS-J Photogramm Remote Sens. 2014 Feb;88:119-27. PubMed PMID: WOS:000331921200011
- On Plateau Objective Function:
o Martha TR, Kerle N, van Westen CJ, Jetten V, Kumar KV. Segment Optimization and Data-Driven Thresholding for Knowledge-Based Landslide Detection by Object-Based Image Analysis. IEEE Trans Geosci Remote Sensing. 2011;PP(99):1-16
- On use of machine learning (random forests):
o Stumpf A, Kerle N. Object-oriented mapping of landslides using Random Forests. Remote Sens Environ. 2011;115(10):2564-77.
In those papers you can find out how the process of segmentation can be optimized without having to reply on trial-and-error.
Good luck, Norman
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing