Output raster of Classify pixels using deep learning is blank

[MelodyYu]

Dear all, 

When I tried to use the tool ' ClassifyPixelsUsingDeepLearning', I got the result raster in the contents list and the folder. But it's compeletely blank in the map workspace.

Could anyone can help? 

[DanPatterson]

I presume the map wasn't empty when you ran the tool, so right-click on the layer in the table of contents and select "zoom to layer".  It may be there but doesn't have a defined coordinate system so it is appearing out of the current bounds.

[MelodyYu]

Dear Dan, 

Thanks for your advice. I tried zoom in it's still blank, and the coordinate system is the same as the map. But when I changed the symbology from ‘unique values’ to ‘stretch’, it showed up as black-and-white checkerboard. I guess it's because of my software environment...

[ShivaniPathak]

Hi Melody,

Can you please share some information:

• Which ArcGIS Pro version are you using and how you created the deep learning environment.
• Which model architecture have you trained i.e. UnetClassifier, DeepLab, etc.
• Share screenshot of Classify Pixels Using Deep Learning tool parameters.

[MelodyYu]

Hi Shivani, 

Here are the details:

• I'm using ArcGIS Pro 3.5.0, and downloaded the newest packege 'deep-learning-essentials'. Also in the clone environment (screen shot 1)
• this is the model I applied: Land Cover Classification (Aerial Imagery) - Overview 
• the parameters are as shown in screenshots 2-3, the imput raster is https://tiles.arcgis.com/tiles/nSZVuSZjHpEZZbRo/arcgis/rest/services/HR_Luchtfoto_2024/MapServer

Best,

Melody

[ShivaniPathak]

Hi @MelodyYu,

Thanks for sharing the information. I have few suggestion which I would like you to try:

1. With the output raster which you are not able to see and seems blank:

• In the below screenshot we can't see "HR_Luchtfoto_2_80cm" raster in the map and seems blank raster:

• But when I use explore and click on the map I can see the rasters class values:

• By adding the raster in a new map we can see the raster in the map:

• You can try this for your output raster.

2. Changing cell size in Classify Pixels Using Deep Learning tool:

• The Land Cover Classification (Aerial Imagery) - Overview  pretrained model works best with 10 cm cell size.
• Use 0.1 value for Cell Size parameter in Environments tab of Classify Pixels Using Deep Learning tool:

3. By visual interpretation the raster which you shared doesn't seem like it's spatial resolution (cell size) is 10 cm so you can also try High Resolution Land Cover Classification - USA (https://www.arcgis.com/home/item.html?id=a10f46a8071a4318bcc085dae26d7ee4) which works for raster with cell size between 80cm - 1m.