10-05-2018 02:40 PM
Hello. Can anyone offer advice for an unsupervised classification accuracy assessment? After recoding and completing the assessment, the report looks like the attached - with 'phantom' zeros. They aren't in the attributes (also attached)...but if I look at Recode via the Thematic tab, they are there and I cannot recode them. Suggestions much appreciated.
10-08-2018 11:00 AM
I assume Classes.PNG is a screeenshot from the Atttribute Editor? All three programs Accuracy Assessment, Thematic tab > Edit group > Recode, and the Attribute Editor should be using the same resource from the input file, that being the image's descriptor table.
You mention blank classes are seen in Recode but can't be recoded. Can you provide more details of what 'cannot recode' means?
It seems that Classes.PNG shows the attributes of a recoded image where blank classes have been removed. Are you sure the same image from the Attribute Editor screenshot is used with Accuracy Assessment?
If this doesn't explain what you're seeing, can you attach the classified image?
10-08-2018 11:19 AM
I appreciate the reply. Using Thematic > Edit > Recode, I see 256 values instead of three - still only three in the Attribute Table - and even if I try to recode them all to 0, it doesn't work. Have uploaded a screenshot of Thematic section. Also attempting to upload the reclassified image.
10-08-2018 11:51 AM
What version of ERDAS IMAGINE are you running? I ran your data through Accuracy Assessment in ERDAS IMAGINE 2018 Update 1 and got this
10-08-2018 01:32 PM
2014. I have redone the classification and recode and it keeps giving me 256 values with 253 zeroes. What you have is, basically, there but at the bottom in mine. I can't figure it out.
10-08-2018 01:37 PM
Short of updating to ERDAS IMAGINE 2018, have you tried outputting your data as Unsigned 2-bit (rather than the u8 you're currently using)? You only have 4 classes so it should fit fine in u2. You should be able to use the Subset tool to write the existing classification out to u2.
I don;t have an old copy of ERDAS IMAGINE 2014 lying around to test, but I suspect limiting the data to u2 will also limit the accuracy assessment report to only the DN range that's available to the data type.