Anyone Using Artificial Intelligence With Their Gold Detectors?

[Ben201000]

17 hours ago, GotAU? said:

As for accurately mapping where a detector coil has been, here’s a paper describing using a PDA and a camera to precisely map out the coil position and it creates a low resolution outline image of the target as the coil scans it. The primary use for the system was intended for UXO detecting, but it may be useful for relict detecting as well.  They even use the system for object recognition, but it would be useful for small targets though.  As for hardware, I’d think any new smartphone has the capability to do this, as the researchers who made the system were using a basic PDA and camera to do it.

Thanks for the paper! 

11 hours ago, BrokeInBendigo said:

You're most welcome. I've a keen interest in this, so please excuse if I go on a bit long about this stuff.

Thank you all of that post. There was so much in it. I'm new to everything to do with detecting so all of it was extremely helpful. What are your ideas on other methods of acquiring the signal? Is multi frequency analysis something that could be used?

20 hours ago, Chase Goldman said:

To do this accurately you need to move away from GPS as it can only be accurate at best to within 6 feet.  Fractions of an inch matter in coil coverage.  I'm thinking a portable local solution.  Perhaps consisting of laser tracking of a coil target monitored and recorded with precise positioning data coupled/integrated with visual information using a drone or drones hovering over the search area.  Think golf ball flight tracking and virtual first down marker lines across the video display of a football field.

I was thinking an approach to this could be to use a similar approach to photogrammetry or biometrics. Only using image and no location data. The phone/drone/AR glasses would map key points of each image and an AI detects the location of the coil against the image.

I'm doubtful if this could work right now, but I 100% think it will sometime in the future. 

In regards to you mentioning that induction is maxed out, are there any new alternatives? Maybe technologies that pick up so much that they're practically impossible to turn into a human interpretable audio signal, but could all be feed into some kind of neural network? Is multifrequency something in this direction? 

[Steve Herschbach]

6 hours ago, Valens Legacy said:

Chase, What if they keep induction based technology, add a full color screen and assign a color to each type of metal. That would give a person more of an idea of what they have come across.

Since I am still new to detecting I have often wondered about this, and would like to know if it is possible to do such a thing. I could see cases where silver and gold would show up at the same time, or copper and iron together. If the display would show the colors of each item then that would be the next step in detecting for me and a lot of others.

Just a thought.

The problem with a lot of this theorizing about machine learning, is that metal detectors know less than most people think. For instance, they have no idea what metal is under the coil. Any metal can read almost anywhere, depending on the size of the target. Target id only works for items that never vary, like a silver dime, yet many trash items can mimic a silver dime. Still, you can build a detector that can cherry pick coins rather effectively for use in the U.S.

But go almost anywhere else, or detect for anything else, and it rapidly turns into “dig all non-ferrous targets”. Nugget detecting, jewelry detecting, relic hunting, and detecting anywhere with history, like the U.K., you better be digging all non-ferrous. That’s because the range for metals is based on target size, not composition, and good targets exist at all target id numbers. Eliminate anything, you potentially eliminate good targets. In a nutshell, for any given metal, the target id range goes from small targets to large targets. The idea target id identifies type of metal is an illusion.

And then you discover that all detectors misidentify some non-ferrous as ferrous, so you end up just digging everything. Target masking also motivates people to remove trash targets. All nugget hunters eventually dig all targets, until an area is devoid of targets, period. This is extending to both beach and relic hunting as time goes by.

Depth? Max VLF depth has not changed in decades. My old Compass Gold Scanner Pro went as deep as any VLF made today. We have got an improvement in target id accuracy at depth, but sheer depth hit a wall ages ago in VLF. PI is where the depth advances came, and we are at the wall there also. I can imagine some improvements in target id in PI by rapid mode switch analysis, but due to the nature of PI you will always have severe overlap of ferrous and non-ferrous, because, again, target id is based more on size, than composition.

Target ID More About Size, Than Type Of Metal

The target id range for one special type of metal, gold, goes from deep in the ferrous range, for gold in mineralized ground, all the way to silver dollar and higher, if you are detecting gold bars. Pick your metal, and the result is the same. It’s all about the size of the target, with infinite overlap between metal types. For every gold target, an equivalent aluminum target can exist.

Finally, a detector really does not even know the size of a target. It only knows strong signals, and weak signals. So a large target at the edge of detection depth can deliver the same result as a small surface target. Usually only small gold nuggets give a ferrous reading in mineralized ground. However, every nugget, no matter the size, if buried deep enough in mineralized ground, can read like a small ferrous target. At the edge of detection depth the ground signal overwhelms the target signal, flipping non-ferrous items to ferrous. If you pass on ferrous targets, you can pass on a two ounce gold nugget at depth (for example). This applies just as much to silver coins as to gold nuggets. Any non-ferrous target can read ferrous under the right conditions in mineralized soil, and the more mineralized the soil, the more severe the problem. I hunt ground where coins read ferrous after just a few inches, and recently saw a detector identify a large brass shell casing as ferrous at two inches.

Tune Out Nails, You Will Miss Gold!

Long story short detectors are as dumb as stumps, and good luck making them smarter. The only thing that can get smarter is the operator, and the smart ones know that as long as targets remain in the ground, good finds can still be made. Reliance on discrimination is the main reason why good targets continue to exist, waiting to be found by those who will dig what everyone else thought was trash.

 

[Norvic]

12 hours ago, phrunt said:

While I've never owned any model of GPX below the 4500 both my 4500 and 5000 can turn the Ground Balance off.

Aye, very rarely used a GPX in ground tracking (other than the 6K) same with the GPZ 7000. Manuals been the go, auto tracking probably is part of the evolution of our detectors to AI, Geosense probably another part, hope I`m around to see AI compete with what we have between the ears.

[phrunt]

Yea, I don't think I've ever used tracking on the 4500, but in saying that I'm the person least likely to need to use it, I was always on fixed and manual is my go-to on the 7000.  The more detector manufacturers can improve ground balance the better depth the users will get, I think people in hotter soils would get a surprise the depth difference between having any ground balance and ground balance disabled.  For example, the GPX 5000 with GB completely turned off, not fixed, off completely gives so much more depth than the same detector with Ground balance enabled and balanced in mild soils where the detector gets no reaction from the soil with it either on or off.    The QED was the same, in any mode other than Mode 11 (GB Disabled) in mild soils even with a perfectly balanced detector the depth is killed just by enabling the ground balance circuit.  I guess in a way Geosense is working towards this, I would still like a way to disable it on the 6000 just to see how it works for me.

In some of my prospecting areas I can run both the QED and GPX 4500/5000 with ground balance completely disabled and they remained perfectly balanced, no reaction from the ground at all, the down side is by disabling it the hot rocks really come alive so the area has to be selected carefully to take advantage of it.

The better manufacturers can improve the detectors ground balance the better the depth will be and if they can use some sort of AI technology to do this or just faster processing or whatever it will be the next big improvement in PI's.  I always thought that if I used ground balance and balanced the detector I'd get the same depth as if I had ground balance turned off entirely in very mild soil, and this is simply not correct.    My mild soils are the perfect example of this as even though I can run with no ground balance at all, just by enabling it I am hindering depth.

[BrokeInBendigo]

2 hours ago, Norvic said:

Aye, very rarely used a GPX in ground tracking (other than the 6K) same with the GPZ 7000. Manuals been the go, auto tracking probably is part of the evolution of our detectors to AI, Geosense probably another part, hope I`m around to see AI compete with what we have between the ears.

Simon’s referring to no ground balance at all - not fixed/manual GB - which is a setting I’d wager nobody uses on Australian goldfields. 

[BrokeInBendigo]

1 hour ago, phrunt said:

Yea, I don't think I've ever used tracking on the 4500, but in saying that I'm the person least likely to need to use it, I was always on fixed and manual is my go-to on the 7000.  The more detector manufacturers can improve ground balance the better depth the users will get, I think people in hotter soils would get a surprise the depth difference between having any ground balance and ground balance disabled.  For example, the GPX 5000 with GB completely turned off, not fixed, off completely gives so much more depth than the same detector with Ground balance enabled and balanced in mild soils where the detector gets no reaction from the soil with it either on or off.    The QED was the same, in any mode other than Mode 11 (GB Disabled) in mild soils even with a perfectly balanced detector the depth is killed just by enabling the ground balance circuit.  I guess in a way Geosense is working towards this, I would still like a way to disable it on the 6000 just to see how it works for me.

In some of my prospecting areas I can run both the QED and GPX 4500/5000 with ground balance completely disabled and they remained perfectly balanced, no reaction from the ground at all, the down side is by disabling it the hot rocks really come alive so the area has to be selected carefully to take advantage of it.

The better manufacturers can improve the detectors ground balance the better the depth will be and if they can use some sort of AI technology to do this or just faster processing or whatever it will be the next big improvement in PI's.  I always thought that if I used ground balance and balanced the detector I'd get the same depth as if I had ground balance turned off entirely in very mild soil, and this is simply not correct.    My mild soils are the perfect example of this as even though I can run with no ground balance at all, just by enabling it I am hindering depth.

Exactly. Your experiencing running with GB off illustrates how much detecting depth depends on the ground handling. 
 

Think I’d still rather have our Aussie sized nuggets and hotter ground than your mild soils and fine gold ? hope you can make it over here some time for a prospecting adventure and smash some personal bests!

[Gold Catcher]

I always use semi auto GB, manual GB only when homing in on a target, but not all the times (usually only to avoid tracking out faint targets). IMO the GPZ ground processing algorithms are unmatched by any other detector that I have used (including 6000), but it will require a proper GB for it to work.

Considering how AI has improved the technology in so many technology sectors, I would not be surprised if it would eventually also be used for gold detectors. There ought to be a way to differentiate lead and iron from gold via decay patterns. Just put Watson on it. But please without adding 2000 lbs to the detector.

GC

[BrokeInBendigo]

10 hours ago, Ben201000 said:

Thanks for the paper! 

Thank you all of that post. There was so much in it. I'm new to everything to do with detecting so all of it was extremely helpful. What are your ideas on other methods of acquiring the signal? Is multi frequency analysis something that could be used?

I was thinking an approach to this could be to use a similar approach to photogrammetry or biometrics. Only using image and no location data. The phone/drone/AR glasses would map key points of each image and an AI detects the location of the coil against the image.

I'm doubtful if this could work right now, but I 100% think it will sometime in the future. 

In regards to you mentioning that induction is maxed out, are there any new alternatives? Maybe technologies that pick up so much that they're practically impossible to turn into a human interpretable audio signal, but could all be feed into some kind of neural network? Is multifrequency something in this direction? 

Here's a moderately technical paper from Minelab (who, at this time, makes the best pulse induction prospecting detectors): https://www.minelab.com/__files/f/11043/KBA_METAL_DETECTOR_BASICS_&_THEORY.pdf

More basics (see Multi Period Fast, Multi Period Sensing, Smart Electronic Timing Alignment, Zero Voltage Transmission): https://www.minelab.com/anzea/knowledge-base/key-technologies

Some good info in there. Minelab does use multi-frequency TX and RX but their analysis is magic sauce. Imagine they have some sophisticated analysis. As you well know, if there was AI involved anywhere, their marketing would ensure we know allll about it.

Alternative avenues of improvement include ZVT (as mentioned above, this was the main innovation in the GPZ 7000, released 2015, new updated model on same platform expected in the relatively near future) and innovative coil design (for example, the recent rise of concentric coils for the GPZ, some recent patents for coils in which a flat-wound coil is twisted at front and rear to be vertically oriented, to reduce saturation from ground mineralisation and other designs).

[phrunt]

6 hours ago, BrokeInBendigo said:

Exactly. Your experiencing running with GB off illustrates how much detecting depth depends on the ground handling. 

Think I’d still rather have our Aussie sized nuggets and hotter ground than your mild soils and fine gold ? hope you can make it over here some time for a prospecting adventure and smash some personal bests!

Just imagine how many of your big Aussie nuggets are left in the ground as current technology can't get deep enough, so if they improve ground balance at some point in the future the possibilities are great.   NZ is a great training ground, finding little dinks at depth with small variations in threshold, the ever slightest of target response is very good training for when better ground is encountered.    The depth I see Aussies get the little nuggets and carry on they were deep in videos is quite funny, the fact they call them little, and the fact they think they were deep.  It really demonstrates the differences in soil conditions and detector performance in different conditions though.