AI Archeology

Before any detection model can be created we need to feed the model with a knowledge about the domain, where the model will be applied. Let’s call this feeding phase learning. Our domain is archeology and detection of sites in LIDAR with a historical value. There are different strategies how to approach the problem of learning. The most basic one is to label the object in the sample data. That is, with a labeling tool, the maker of the model, the teacher, marks the very specific locations in the training data, where the object of interest is. The marks in the training data are called annotations. Please check the previous part related to collecting the images of locations with our object of interest: the Roman watchtowers.

In the screenshot above I am using a labeling program called RectLabel¹, but there are different applications that can help you with the task: 3D Slicer, Labelimg, VoTT (Visual Object Tagging Tool – developed by Microsoft), VIA (VGG Image Annotator), LabelMe, and Label Studio.²

All of them support annotation and even if I am using this RectLabel tool, I stick through the process to the basic labeling, or annotation functionality.

Let’s dive more into what we actually see in the screenshot. There is the LIDAR sample image in the screenshot’s left side. That picture we have collected from the LIDAR sources.³ In the center of the image you can see a small annotation rectangle. It’s borders annotate the image with almost exact coordinates where the object of interest (in our case a Roman watchtower) lies.
On the right side of the screenshot, there is a pane that shows a small letter w. It is

1. RectLabel
   https://rectlabel.com/ ↩︎
2. 9 Best Image Annotation Tools for Computer Vision [April 2023 Review]
   https://encord.com/blog/9-best-image-annotation-tools-for-computer-vision/#:~:text=CVAT%20is%20one%20of%20the,%2C%20LabelMe%2C%20and%20Label%20Studio. ↩︎
3. Hessisches Landesamt für Naturschutz, Umwelt und Geologie
   https://windrosen.hessen.de/mapapps/resources/apps/windrosen/index.html ↩︎