Jump to content

New Algorithm Reduces Noise


Recommended Posts

11 hours ago, jasong said:

AlphaZero has an open source equivalent called LeelaZero. Leela can run off a laptop computer ok.

Leela Zero is trained by a distributed effort, which is coordinated at the Leela Zero website. Members of the community provide computing resources by running the client, which generates self-play games and submits them to the server. The self-play games are used to train newer networks. Generally, over 500 clients have connected to the server to contribute resources.[7] The community has provided high quality code contributions as well.[7]

The above quote is from Wikipedia and is actually referring to the Go (game) version, not the chess version.  Apparently for this to work on a small platform the 'crowd learning' aspect would need to be carried over.  So users would share the learning their individual detectors experience in some kind of uploadable & downloadable database.  Interesting.  I hope there are enough users to make that useful.  Or in addition will the detector company have some kind of robot course where the software is trained and then incorporated??

When I was looking into artificial neural nets (ANN's) which I think is Alpha's artificial intelligence method, I was told it typically takes hundreds of thousands of datapoints to build a reasonable model.  Maybe metal detecting can be done with considerably less -- I think that's going to be a requirement.  But as you say, it will likely require "thinking outside the box" so conventional wisdom could put placed on its ear.

  • Like 1
Link to comment
Share on other sites


Another area of promise is computer enhanced digital signal processing and frequency analysis. Our detectors already do this, but they are limited in discriminating because there’s limited AI behind the signal processing. There are subtle sound differences a detector gives off on certain targets that experienced operators learn as they use their machines that help cue them to what may or may not be a good target, and if a detector could be programmed to learn this along the way, that would be something.

I used to work as a wildlife biologist and did bat surveys using a program that would learn different types of bat calls to help identify them from recorded ultrasonic call files. It would compare different frequency characteristics to known calls to find a best fit.  There is currently research being done on mine and UXO detectors to help them better discriminate between trash and hot targets. Part of the processing uses frequency analysis using polar plotting that compares signal amplitude, timing and frequency. It’s an interesting way to look at the signals and how they are  analyzed.

I tried this out in my yard once using my smartphone and a frequency analysis app with my SDC as I ran it over aluminum, iron and a bit of gold,  but there was too much EMI to do it well and I gave up.  Sometime I want to try it out in the field under real conditions.  It has some promise, and it would also be a great way for people with compromised hearing to be able to see exactly what their PI detector is saying  because instead of just a few LED lights turning on from a target, the display from the phone app gives a lot of information about the signal.

  • Like 4
Link to comment
Share on other sites

17 hours ago, Steve Herschbach said:

You know, when I was a teenager you could buy a pocket calculator that added, subtracted, multiplied, and divided with a couple other little functions for $500. I told my friend and partner Dudley that someday they would give them away free. He of course scoffed but in later days now he brings it up a lot that I was right. Never underestimate how quickly and how fast technology can advance giving us more power at lower prices. The expensive high tech toys of today are no different.

Just to add on what Steve said, today, you can buy a birthday or holiday card that plays music when you open it up. This card has more computing power than was available to the entire Allied armies of WW2. But today, when we're done with the card, we throw it away.

  • Like 2
Link to comment
Share on other sites

1 hour ago, GotAU? said:

There are subtle sound differences a detector gives off on certain targets that experienced operators learn as they use their machines that help cue them to what may or may not be a good target, and if a detector could be programmed to learn this along the way, that would be something.

That makes me wonder, what if a detector had a simple boolean algorithm that, after every hole you dug, it asks you if it was a "good" find or "trash." After a few hundred digs, I bet it could do a darn good job of being able to figure out if its user is likely to dig the next time the machine gets a signal.

  • Like 3
Link to comment
Share on other sites

46 minutes ago, mh9162013 said:

That makes me wonder, what if a detector had a simple boolean algorithm that, after every hole you dug, it asks you if it was a "good" find or "trash." After a few hundred digs, I bet it could do a darn good job of being able to figure out if its user is likely to dig the next time the machine gets a signal.

That may actually work.  There are several research papers out there on this very type of research, like this one using frequency analysis of detector signals to get more accurate target discrimination: https://www.ndt.net/article/ecndt2006/doc/Tu.4.5.4.pdf

Another interesting part of the above paper is how they are mapping out the shape of objects by using a small camera to determine the position of the detector as it sweeps across a buried object, the resulting image is like what one would get with a GPR but such a setup could be much simpler and less costly.  Something else to add to the GPX 6500!   🙂

  • Like 1
Link to comment
Share on other sites

When developing new detection technologies, I would focus more from my point of view on the elimination of erroneous or unnecessary data, especially the data that are unnecessary for certain types of detection. Here I would use AI.

In terms of the range and sensitivity of the detector, I would like to improve the quality of the S / N parameter of the received signal .. it can be improved in several ways ..

Link to comment
Share on other sites

1 hour ago, EL NINO77 said:

I would focus more from my point of view on the elimination of erroneous or unnecessary data, especially the data that are unnecessary for certain types of detection.

Would you give examples of these unwanted data?

  • Like 1
Link to comment
Share on other sites

The detector works more or less than a measuring device ... the signal from the coil goes to the electronics of the detector where it is processed through various stages of filters ... ..
Here are 2 simple examples:
 1 .Example .... - "detection of coins and jewelery" - you are only interested in non-ferrous signals .... If you use too open discrimination in such detection ... .or from the mineralization of the terrain ... it can strongly influence the correct final calculation of the target signal ID .... this means that it also affects the stability of detection ....
If you use a reasonable closure around the iron / non-ferrous border ... very often there is a significant improvement in the ID ...
 
I mostly use such a setting for detectors with Mix mode..a so I know that I will not pass any signal ..- even though I know that my discrimination is set reasonably high ..

2. The example is quite well known - it is a reduction of the signal or volume of iron, which is called "Iron Volume", which has a significant effect on the separation properties of the detector ... sometimes and extremely positive.

 

  • Like 1
Link to comment
Share on other sites

12 minutes ago, EL NINO77 said:

1 .Example .... - "detection of coins and jewelery" - you are only interested in non-ferrous signals .... If you use too open discrimination in such detection ... .or from the mineralization of the terrain ... it can strongly influence the correct final calculation of the target signal ID .... this means that it also affects the stability of detection ....
If you use a reasonable closure around the iron / non-ferrous border ... very often there is a significant improvement in the ID ...

Is this true when operating in simultaneous multifrequency or just in single frequency (or selectable frequency)?

  • Like 1
Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...