I am attempting to run a Maximum Likelihood classification using ENVI Classic and pixel-based ROIs. However, whenever I run the classification several (but not all) of my classes consistently are placed as 'unclassified' pixels. I have tried manually setting the threshold for all of the classes to 0, thinking it was possibly a problem with that, but the issue still persists. I thought maybe it was a problem with the scale factor, but why would it only be affecting 2 of the 12 classes? And if it is a scale factor that for some reason is only affecting a few classes, I'm not sure of the best way to resolve it. Would manually changing the scale factor from what ENVI selects cause problems with the other classes that are currently being classified with decent accuracy? If it's not the scale factor and it's not the threshold, I'm not sure what other variables to check. Other algorithms, such as SVM, do not have this same issue occur. Any ideas/suggestions would be appreciated.
|