NRCan / geo-deep-learning

Deep learning applied to georeferenced datasets
https://geo-deep-learning.readthedocs.io/en/latest/
MIT License
150 stars 49 forks source link

Skip histogram equalization for already enhanced images #477

Open micpilon opened 1 year ago

micpilon commented 1 year ago

Although undesirable, there may be rare cases where the DigitalGlobe archive has already been enhanced before it enters the inference process. In fact, there is a optional parameter when ordering to apply a Dynamic Range Adjustment to the product. From my observations, applying our own in-house enhancement to this type of source greatly affects the quality of the image, making it unusable (see figure)

image (13)

A suggested solution would be to omit the enhancement step when it is already done by the supplier. Although the approach applied is probably a bit different from ours, I would expect better extractions than for a "double enhanced" image.

Here's how to find that information: In the METADATA asset of the STAC Item, see the tag <RADIOMETRICENHANCEMENT>DRA/ACOMP</RADIOMETRICENHANCEMENT> as opposed to expected <RADIOMETRICENHANCEMENT>ACOMP</RADIOMETRICENHANCEMENT>

remtav commented 1 year ago

As found by @micpilon : ACOMP = atmospheric corrections DRA = Dynamic Range Adjustment