DL image numerical target Question

[ghost]

Hi there

Thanks so much for your repo. As a beginner, it has been instrumental for my learning.

I have a question regarding deep learning that I hope you can help me with.

I have numerical features (e.g. precipitation, temperature, slope) map as rasters (tif) files, and numerical target (e.g. soil moisture) map as raster tif files also. I need to use the futures to predict the target using DL (CNN or Unet). Most resouces I looked at only show DL with images for segmentation and classification where the target is CATEGORICAL (binary or multi classes) but is not numerical. I am struggling particularly with the following:

• How to build dataloarder
• How to code the output layer after building the hidden layers of the network

Could you please help with this? or point to resources (preferably, tutorials) that show how to tackle problems like this?

Big thanks in advance,

[robmarkcole]

I suggest to also establish a baseline using binary mask - i.e. high/low moisture per pixel.

For a % per pixel see this approach https://github.com/tha-santacruz/BayesianUNet

Do let me know how you get on