09-23-2019 08:16 AM
ER Mapper returns faulty confusing matrix: only one column (out of 10 classes of >1200 points of the reference data) contains values. This applies to all kind of classified (ISOCLASS) images: ETM+, OLI, Sentinel. Reference data is a set of points digitized on a backdrop of high resolution aerial imagery. I checked all possible weak points: CRS, formats of the data, etc. What's interesting, ERM returns proper matrix for other areas and the same type of the data. Any ideas, suggestions, advise on how to address the problem will be highly appreciated.
09-24-2019 02:14 PM
We are not aware of a problem creating a confusion matrix in ER Mapper. Can you open a support ticket through the Hexagon Geospatial Support Portal with more information about your data and the exact workflow so that we can investigate this further?
Support Portal: https://hexagongeospatial.force.com/supportportal
Hexagon Geospatial Support
09-26-2019 12:42 PM
As I cannot log into support portal I'm attaching here my data (in .zip file): classified image S_ETM_ISO30.ers, reference dataset (points) pts.erv and confusion matrix obtained: conf_matr.xls.
The problem is that ERMapper returns values in the conf. matrix only in one column in this particular dataset. Conf. matrix for other datasets is produced properly.
Thanks for looking.
09-30-2019 12:27 PM
I found a "replacement" solution: I converted reference points (vectors) to raster, classified it (ISOCLASS; raster needs to be in RGB format) and applied as reference data for testing accuracy of the ISO classification in ERM. Works perfect!
Original problem still exists.
10-15-2019 01:32 PM
It looks like the vector as reference data had never worked in all previous versions of ER Mapper since last 12 years. may be nobody ever used / tried / complained about that.
I will report a software bug to our engineering to fix / start supporting vector as reference data in Confusion matrix tool.
Hexagon Geospatial Division