How get small object detection in 4k images DetecNet

[mjack3]

I need to train a 4k dataset of airplanes and make inferences in similar data.
This is how the images look
https://ibb.co/QvbL7kp
(yes there is one plane in the bottom-left corner)

The images have a 5472x3648 shape and DIGITS is unable to learn something in the training phase because the airplane is too small

I have been looking for some documentation to modify the detectnet prototxt but I founded nothing.
Could you give me some advice about what should I do?

:)