Defining objects in image chips for Deep Learning

ElianeMcCarthy — Mon, 20 Apr 2020 06:52:31 GMT

Hi all,

I am currently trying to train a deep learning framework for detecting bats from thermal imagery taken from a drone. You will see in the below image the bats are the detectable as the bright circles against a dark background:

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 good results:

You will see in some instances, completely dark areas are classified as bats, or just the edge of a bat is detected.

I suspect that the issue here is how I am defining the bats in the training samples. I am thinking about trying this process again with rectangular training samples.

Is anyone able to offer advice on the best way to define a bat in my training images?

Thanks in advance for your help!

Elie

topic Defining objects in image chips for Deep Learning in ArcGIS Pro Questions

Defining objects in image chips for Deep Learning