Train a model based on thermal drone data

Hi guys, I’m new to this board. :slight_smile:

For the past few weeks I’ve been researching quite a lot concerning thermal drone data and how I could use them for object detection. All I could find were some papers discussing how to use thermal imagery to detect objects for self-driving cars and similar topics.

I’ve already trained a model that is capable to detect certain objects on cropland based on RGB drone imagery but before I try something similar with thermal drone data, I wanted to ask if that is even possible? Can a model be trained on thermal imagery (my research tells me it can but just to be sure) and can a model take into account the thermal data (pixel values) itself or does it only train itself based on the shape of the objects?

If you need to ask some questions to answer mine, just hit me up. :slight_smile:

Hi Chris, welcome to the forum!

Can you post some sample images? and maybe mark on them what the model has to see?

Regular object detection as far as I understand find shapes, I don’t know how it would do with color grading, but you can also run a color grading algorithm on the shape found (that information is in the image already) and compose the results with the predictions of the model
Maybe you may need to do some preprocessing on the images to make things easier to see for the model