[D] Which model/algorithm should I use to associate the legend of a geographical map (or a chart or a process diagram) to the actual elements in the map?

I'm somewhat familiar with augmenting real-life images to get a larger and more generalisable dataset for building a ML model. But when it comes to maps or engineering diagrams, how can I build a model such that it looks at the legend to know the existing elements and identify them in the image?

Say, I have a map like this image, what model/tool should I use that can look at the legend, learn that there are 5 components (or …

algorithm machinelearning map process