S1 & S2 Land use/land cover classification using deep learning

[RDAltarez]

Hello, I'm working on deep learning land use and land cover classification with Sentinel 1 and Sentinel 2 data.
I used the composite bands to stack my data, and I used the standard deep learning workflow.
I'm doing a binary classification right now ( Forest - a class value of 0, Non-forest - a class value of 1).

When I used my deep learning model to classify the pixels, the result was a black image.

Anyone have any suggestions for how I could solve this? I will surely appreciate.

[andrei_ste]

Are you using GPU? If yes do you have CUDA installed? I used the pretrained deep learning model developed by Esri for Land Cover Classification from Sentinel-2 images and had this issue. Solved it installing CUDA.

[RDAltarez]

Hi @andrei_ste, sorry for this late response. I installed CUDA toolkit 11.7, but I am unsure if each package component was correctly installed.

I tried to do the deep learning pixel classification again and here's what I am getting. Do you know any solution for this problem?

Thank you!

[andrei_ste]

Hi @RDAltarez ,

I also used the Deep Learning Libraries Installer for my specific version of ArcGIS Pro. It is possible that you do not have all the required libraries installed.

You can download for free the Deep Learning Libraries Installer for your specific version of ArcGIS Pro from here:

https://github.com/Esri/deep-learning-frameworks/blob/master/README.md?rmedium=links_esri_com_b_d&rs...

[RDAltarez]

Hi @andrei_ste  thank you for your reply. I did follow this link before doing my deep learning I believe, I have installed 347 packages. Please see below. I also found other way to get away with the "Unbound error...". I used Jupyter notebook, and used this script (see attached).

Unfortunately, when I used the created model, the resulting image is clear. There's no classified image. See image below.

[andrei_ste]

I read your first post and you said that you used composite bands to stack your data. I think you should use instead the metadata of your satellite image. Also are you using the pretrained model from Esri or are you using a trained model created by you or someone else? You could use the pretrained model from Esri as a test to check if everything is set up ok: https://www.arcgis.com/home/item.html?id=afd124844ba84da69c2c533d4af10a58

There are detailed instructions on how to use the model in a pdf that you can find in the description. If the steps provided work than it must be something related to your trained model or input parameters.

[RDAltarez]

Thank you @andrei_ste  for your response. I followed the steps provided in this link https://community.esri.com/t5/arcgis-pro-questions/help-with-deep-learning-and-pixel-classification/.... I was able to generate my classes, at last! I used a TIFF file of my Sentinel 2 (RGB) as my input. Yahoo!