Hi,
I'm using Idrisi Selva to do an unsupervised (ISOCLUST) classification of Landsat TM images. My study area is at the juncture of two Landsat scenes and it spans two UTM zones. I radiometrically corrected each band of each image in R, then mosaicked each band in ArcGIS v10, then clipped the mosaicked bands to the boundary of my study area.
I exported the data to ASCII, then imported three bands (R, NIR, MIR) to Idrisi Selva. Using ISOCLUST, I specified that I wanted 10 classes with a minimum class size of 30 pixels.
The problem is that every few minutes, I would get an error that said "Variance/covariance matrix is singular. Check for perfectly correlated variables or variables with zero variance."
This message was immediately followed by: "Inverse matrix process failed."
I found that if I sat there and clicked through the errors when they popped up every few minutes, the classification worked (at least from a quick visual assessment).
My questions are:
1- Why is this error occurring?
2- How is the variance/covariance matrix generated?
3- Did the unsupervised classification work in spite of the error messages?
Any help would be really appreciated.