Select to view content in your preferred language

Possible to use new training data for Land Cover Classification model?

349
4
04-24-2024 01:29 PM
rescobar
New Contributor III

Hello, 

 

I'm trying to use the following model:

Land Cover Classification (Aerial Imagery) - Overview (arcgis.com)

And following this guide to "fine-tune" a pretrained model: Fine-Tune a Pretrained Deep Learning Model (esri.com)

 

The "ModelName" in the .emd file is "UnetClassifier". I don't have this option to export training data into this format from "Export Training Data for Deep Learning (Image Analyst)" Is there a way to create a training file from ArcGIS pro?

 

 

 

0 Kudos
4 Replies
PavanYadav
Esri Contributor

@rescobar here are detailed steps - https://doc.arcgis.com/en/pretrained-models/latest/imagery/finetuning-the-land-cover-classification-...

You should be looking for  the Classified Tile metadata format in the Export Training Data for Deep Learning (Image Analyst)"

In the Train Deep Learning tool tool - you don't have to set UNET either. 

Pavan Yadav
Product Engineer at Esri
AI for Imagery
Connect with me on LinkedIn!
Contact Esri Support Services
0 Kudos
rescobar
New Contributor III

I try to use the "Export Training Data For Deep Learning (Image Analyst)" but I get this: 002088: Feature class does not contain enough information for supervised training.—ArcGIS Pro | Docu...

I do have "ClassName" and ClassValue fields in the feature class I'm trying to use.Error_exporting.PNG

0 Kudos
MuhammadElbagoury
New Contributor II

Try to assign the {Class Value Field} Parameter in Export Training Data Tool to the name of the field which has these values which is in your data {ClassValue}

rescobar
New Contributor III

If I use a shapefile as the "Input Feature Class Or Classfied Raster or Table". The {Class Value Field} doesn't populate - I have a "ClassValue" field in the shapefile. If I use the output of a previous operation to test this function, it tells me the raster is not thematic (Error 002090 Raster is not thematic.)

0 Kudos