The computer vision (CV) can be summed up as an image matcher. The CV only knows what it knows, so if there is no or very little representation of those types of crickets in the CVs training data. The CV will suggest the next closest looking thing.
Here you can see with the geomodel included. There isn’t really a good match in the CV system.
When ignoring location, you can see other crickets mostly from other regions suggested.
Really the only way to fix it is to teach it those crickets without representation in the CV exist by identifying them enough the taxa get eligible. If they can’t be identified due to lack of identifiers, difficulty, they are undescribed, etc. There’s not much that can be done.
There are many other forum posts about CV issues. There are also multiple feature requests requesting changes or revamps to how the system works.