Unable to classify pixel using deep learning.

288
2
05-09-2022 07:05 AM
OgbajeAndrew
New Contributor

Hi Everyone,

I am working on Sentinel-1 using deep learning . The work is aimed at detecting flood water (pixel classification) using Unet Model and Resnet 18 backbone. Just wondering why (inference) pixel classification using deep learning has not been successful.

OgbajeAndrew_0-1652103610886.png

OgbajeAndrew_1-1652103773735.pngOgbajeAndrew_2-1652103852551.png

OgbajeAndrew_4-1652104833861.png

 

Unfortunately, after hours of model training, deep learning pixel classification turns up with an empty layer as attached.

OgbajeAndrew_3-1652104545873.png

Please, can anyone help?

Cheers

Andrew

0 Kudos
2 Replies
PriyaSharma_17
New Contributor

Hello Andrew, were you able to solve it?

0 Kudos
PrestonSenderoff
New Contributor

I'm encountering a similar issue with Landsat 8 data. I've created a composite image from bands 1 - 7 and am attempting to use the Classify Pixel Using Deep Learning tool with the Lansat 8 model deep learning package from ESRI Analytics. However, the resultant raster is always empty; it displays  the symbology in the contents panel, but no map-layout item is present. Also, when a copy of the raster is produced and the attribute table destroyed, the result is a raster with a checkerboard-grid of lowest and highest values in the original raster. Any fix?

0 Kudos