I am new to deep learning and trying to see if it is useful for land cover classification. I am currently specifically looking into canopy cover classification. I was able to run the "Classify Pixels Using Deep Learning" tool however, most of the canopy cover was not classified. I think this is because I am not creating good training data. I would very appreciate if anyone provides me with suggestions/tips for the training data.
I am using locally stored NAIP imagery with 1m resolution (6403 x7659) for testing. Please see the steps I took below:
Step 1: I classified canopy cover was creating 30m fishnet layer and classified whether there is canopy cover or not. Then export the feature layer to raster and reclassified canopy cover value (green) to be 1 and unclassified value (pink) to be 0 (Image 1).
Step 2: Exported training data using the "Export Training Data for Deep Learning" tool (Image 2.1). Using the "Input Mask Polygons" parameter didn't output images and labels. I was not able to figure out the reason not creating chips so I run without it. 503 images and labels were created, but as you see in Image 2.2, most of them aren't classified. I think I need to use the "Input Mask Polygon" parameter to train chips to be only classified to canopy cover. Is there a particular specification for "Input Mask Polygon" data? I tried a few polygon feature layers but none of them didn't work for me.