Neural network in a metal detector

[ivconic]

Neural networks can find patterns in fuzzy data that are difficult to see using other methods, but the information must be there in the first place.
The training data is also very important, otherwise the network will have difficulty converging. One of the problems with NNs is that they can converge on a local minimum, so you may not get the results you expect.
Personally, I'm not sure an NN will generate any significantly better results in a metal detector than the methods already used. I've seen papers where NNs have been used to analyze data from mine detectors, but the advantages (or otherwise) were questionable.
Anyway, it's always good to experiment with new alternative ideas.

Ha, ha, ha!
How God arranges for some things to happen at the same moment! That's amazing!
I have just had an extensive and detailed correspondence with a colleague in China.
I explain to him how the Fisher 1265 is a "cornerstone" and a very important project in the history of metal detectors.
Of course, don't be offended, I don't expect 99.99% of people on this forum to understand the point and what I'm talking about.
But I'm just trying to explain to my buddy in China why we should go back to basics and start developing a fundamentally good detector.
And as an educational example, I proposed the model 1265 because it is everything that represents the essence of successful work in the real field.
1265 is the essence of everything. 99.99% of people won't understand this and I don't mind.
Because understanding this requires a lot of knowledge and even more real experiences in the field.
Abstract simulations and simple schematic analysis are not enough to even touch the tip of the iceberg.
But why am I now mixing this story into this ongoing topic?
Because Qiaozhi just said what I said earlier in that uncharacteristically short post of mine.
No matter how good and advanced the software is; it will ONLY be able to process what the hardware is capable of delivering!
And nothing more than that!
Analogous to that; 1265 hardware is one of the few hardware that delivers the maximum of what is possible to deliver.
Again, not intending to hurt anyone's ego... just so you get this; you have to lick a lot of salt in life! Kids!
I am very sorry that Dave is not here with us now. I'm sure he would have peed himself laughing!

[moodz]

You guys are looking at the wrong end of the stick …. The real application of neural nets and metal detectors is solving the complex equations
Involved in the settings of the detector. For instance when you adjust the ground balance control your meat ware in your head is “solving” the ground balance equation implemented in the detector.

If there’s a whole bunch of controls … parameters if you like … the NN adjusts these better than you could and converges on a solution that might be maximum target signal or target type or optimal ground balance etc etc.

eg the solution to these equations were found by a NN but could not be solved by matlab.

[Taktyk]

Yes here parameter optimization, ground balance or ground tracking can also be based on the model.

[ivconic]

Makes sense.
I completely misunderstood the idea, my fault.


This gave me the idea to write a recent experience... so as not to spoil this topic; I'll refer you to my post #52 on this topic: https://www.geotech1.com/forums/foru...anticore/page4

[CrizzyD]

The has been some work done on convolutional neural networks when looking at the Magnetic Polarisability Tensor (MPT). I'd suggest looking at some papers concerning analysis of coinage (it seems you've already gone down the rabbit hole of demining). But MPT can take into account things like rotation and orientation as applied to within the magnetic field

[Taktyk]

The has been some work done on convolutional neural networks when looking at the Magnetic Polarisability Tensor (MPT). I'd suggest looking at some papers concerning analysis of coinage (it seems you've already gone down the rabbit hole of demining). But MPT can take into account things like rotation and orientation as applied to within the magnetic field

Interesting example I need to read more. It's surprising how much more there is to do. I believe that years of experience when it comes to analog technologies, analog methods, then digital processing have given the right foundation to enter the next more intelligent level.

[Rafael]

Your project is very interesting. I also think that you should explore using penetrating radar (GPR) technology to search for anomalies in the subsoil and that would give you the advantage of being able to detect non-metallic mines.

[Taktyk]

Your project is very interesting. I also think that you should explore using penetrating radar (GPR) technology to search for anomalies in the subsoil and that would give you the advantage of being able to detect non-metallic mines.

GPR brings new data it certainly gives more analysis but the cost also increases. I think AI may be some cheaper alternative.

The idea was to make the project cheap and easy to implement. In general, UXO detectors are hellishly expensive.

Rather, it will remain an amateur project until it finds investor interest. Projects of this type consume a lot of time and money.

For the time being, I simply wanted to combine my considerable experience and knowledge of MD with the subject matter I'm currently pursuing, which is programming and AI.