I was on a team led by a guy who claimed to have a viable design for GPR in a consumer level metal detecting like machine. He claimed to have built a prototype and was looking for production when he passed away. I never saw the design, not sure if anyone else did either. I'm a "believe it when I see it guy", so who knows? I personally always thought sonar/ultrasound might be a better bet, but I'm not smart enough to know which frequencies work best. I do know ultrasound is pretty amazing at short range anyway.
The bottom line, IMHO anyway, is that VLF is dead. I'm doing just as well with my ancient E-Trac as all these new machines are doing. I'm sure the new machines are better, but only on the margin. All they seem to be doing is tweaking around the edges and coming up with new buzzwords and new ways to look at old returns with faster processors and better software, but no new game changing technology.
The next paradigm will come from out of the blue, IMHO, either GPR, sonar, or something we haven't thought of yet. I know people are working on this stuff, and have been for years. Perhaps the fact that it hasn't come yet suggests that it is more difficult or costly in practice that I think it is.
As for using GPR to find depressions, LIDAR works pretty well for that purpose now, and LIDAR returns and software to read them are all available in the public domain for free. You would be amazed at what you can see on LIDAR, like old roads and stuff, that you simply can't see in the field, or at least not see easily without an old map.