We are trying to map land cover classes on a watershed. We have selected training sites (during a field campaign in early 2017) and extracted their spectral profiles based on a Landsat 8 image acquired at the time of field surveying.
In order to assess the land cover changes, we wanted to map the same cover classes at a previous year. Since our training sites might not be relevant, we wanted to perform supervised classification using endmembers spectra instead of ROIs. When importing those spectra inside ENVI's Endmember Collection toolbox, it appears that only Spectral Angle Mapper and Spectral Information Divergence classifiers could be used. Common algorithms such as Maximum Likelihood or Mahalanobis distance fail, returning the following error message :
Problem: the selected algorithm requires that the collected endmember spectra all contain an associated covariance. ENVI is unable to continue because some of the endmembers collected to not have their covariance.
Could anyone help here ? Actually is our method relevant ? How can we possibly perform supervised classification using Maximum Likelihood/ Mahalanobis classifiers on some older satellite images ?
Dear Roland Yonaba, methods using ROI and methods that uses spectral end members may be different and applied for different aims. I really don't see why you should use ML or MD classifiers instead of using another one. I think you should put your efforts to have an accurate map. You can use SAM, however since you're using OLI images, probably you don't have enough spectral features to do a very good classification. On the other hand, are you going to assess the accuracy of the classifiers? Is yes, you may not use your in situ spectra, and then use these spectra information do to the validation.
Point taken, Enner Alcantara. However, we were focusing on ML/Mahalanobis classifiers just because they were the classifiers that work the best (higher Cohen's kappa values) when doing supervised classification for 2017. As such, for the sake of consistency, we wanted to rely on the same classifiers for the previous dates.
I would highly recommmend you to implement SVM classification that is generally accepted of outperforming than conventional classifiers like ML or MD. Have a look my paper below.
@article{doi:10.5721/EuJRS20154823,
author = {Mustafa Ustuner and Fusun Balik Sanli and Barnali Dixon},
title = {Application of Support Vector Machines for Landuse Classification Using High-Resolution RapidEye Images: A Sensitivity Analysis},
journal = {European Journal of Remote Sensing},
volume = {48},
number = {1},
pages = {403-422},
year = {2015},
publisher = {Taylor & Francis},
doi = {10.5721/EuJRS20154823},
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing