Using AI for map georeferencing

How does this work on non-standard maps? How about to extract features? I have a client that permits utilities to use their right of way, and the permit holders have to submit a map in their permit application showing where they ran their utilities at. The maps are of various qualities. I do have a milepost for every application that I’ve fed into Azure’s AI, but it has been pretty miss to do the georeferencing and feature extraction.

How can we get this scenario / workspace to use in our process ?

​@dijkdjv Here is the workspace, the input data, and the results. Please keep in mind a few things:

• you’ll need your own Gemini connection
• the process is built for very particular naming convention, which allows building projection name, and it might be not available for other sources.
• my prompts are not necessarily the best. I went through a few iterations, and other prompts looked very different yielding similar results, but you’ll need to find your own that work best for your data
• I had to adjust the prompt to each scale in my series of examples because their neatlines looked a bit different, so having three prompts for each map type is better than one prompt describing three different situations.

https://www.dropbox.com/scl/fi/g8x6su6aozwvcj5mq47d8/MapGeoreferencingV2.fmwt?rlkey=xlfmu0xnkksdn6kj5qmvz2g77&dl=0

Dmitri

​@runneals my process was specifically made for Soviet map sheet nomenclature which allows easily calculate where a map belongs in the world. Other maps may require some other methods - maybe reading (OCR) information outside the map frame or using other clues. For feature extraction, I would go with a more specialized tools such as Picterra, but Gemini (with FME, of course) is actually not bad with finding/extracting features. Here you can see a couple of examples:
 

Dmitri