I am performing end-to-end deep learning in ArcGIS Pro 2.5, following a similar procedure to the one set out in this tutorial: https://learn.arcgis.com/en/projects/use-deep-learning-to-assess-palm-tree-health/
I am trying to train the model to recognise bats in thermal orthomosaics taken from a drone. The bats are much warmer than their background, so are detectable in the images as bright circular objects with a diameter of approximately 5 pixels, against a dark background (see image 1).
So far, I have been defining training samples by drawing a circle from the center of each bat to just outside the edge of the animal.
However, this is not giving me great results.
Is anyone able to offer advice on the best way to define a bat in my training images?
Should I exclude all of the dark area surrounding the bat?
Should i use the square training sample drawing tool instead?