using a classified raster as input in Deep Learning model for UNET but no .ecd file?

504
0
08-04-2020 07:27 PM
Labels (2)
AnnaMikkelsen
New Contributor

Hello,

I am trying to classify aerial images using the Deep learning toolbox with UNET, but I am having trouble using training data that is already annotated and with no .ecd file.

I have about 40 images that were classified in ENVI using an ISODATA classifier which I subsequently manually edited by merging several classes together, resulting in a .dat file, and a .tif file with three 'values/classes'. I would like to use these images as training data for a Deep Learning model in ArcGIS Pro, but when complete the step to "Export  Training data for Deep learning" it will only take an .ecd file or a shapefile. From what I can read the .ecd file is created in ArcMap by training another model (supervised or unsupervised). I have not had good luck with unsupervised classification alone and would prefer not to redo the work I already did. The same issue occurs if i attempt to create a 'formal' classified raster from my training data using the "Classify Raster" tool with no .ecd.

I tried to convert the classified .tif file to a shapefile, which lets me "Export Training data for Deep Learning" AND it lets me "Train Deep Learning Model", but my problem here is that is only allows me to train an object detection model (Single shot detection or Feature Classifier) and not any pixel classifier. out of curiosity I trained the feature classifier and ran it, but it fails before creating any model..

Apologies for the long question. If anyone has any inputs or ideas how to either convert the training data I have now to an .ecd file or any other workaround, that would be greatly appreciated.

Thank you! Have a wonderful day

0 Kudos
0 Replies