Artificial Neural Network - ANN and detectors

[sled]

ANN for target discrimination is a very difficult task, generating a good training and test set alone is probably worth a thesis on its own.
Although it would be interesting to try something along the line of Recurrent Neural Networks, LSTM, Time Delayed Neural Networks or Dynamic Time Warping.

Perhaps it's easier the other way around and try to model the ground signal using a neural network, also easier to get some test data

And keep in mind that most detectors only deliver phase and amplitude, so a very sparse feature vector, where a NN might not yield much improvement over conventional signal processing. Certainly interesting to see how it can be done with a NN.

Cool would be to correct for signal variations due to the search heads orientation usinng a gyro and accelerometer or using the neural network to control the transmitter's amplitude and phase, i.e some kind of reinforcement learning.

[Qiaozhi]

It might be a technically interesting exercise to build a metal detector with an artificial neural network, but practically I doubt it will be any better than the existing methods of discrimination.
Also, you would need to somehow train the network to classify various targets, and air tests would give different results compared to items buried in the ground matrix. One possibility would be for the detector to record the amplitude and phase for a particular target, then (after digging it up) the user could assign the target to a predefined class. However, in practice, the target IDs can have an enormous overlap, and the results are also dependent on type of material, depth, ground conditions, size, shape and orientation. At the end of the day, target IDs are only a guide.

Regarding the idea that various material types have a specific frequency is pure pseudo-scientific nonsense. It is true (for example) that 50kHz is good for finding small flakes of gold, 15kHz is good for finding smaller objects, 1.75kHz can provide salt rejection, and a range of 10kHz to 20kHz offers a good compromise. But this does not indicate that the frequency will assist in classifying particular targets.

As I said earlier ... an interesting technical exercise, but the results may be disappointing.

[ford]

Ok, Qiaozhi

Thanks

[Qiaozhi]

it's true,
In frist, i must obtain information about target,

Sample this information:

The most common frequencies used are 5KHz for gold and 8.7KHz for silver; other elements have different frequencies.

http://www.geotech1.com/cgi-bin/pages/common/index.pl?page=lrl&file=/projects/mfd1/index.dat



[ATTACH]40087[/ATTACH]

Your link points to the Basic MFD Project by Carl Moreland. However, we no longer discuss LRLs or dowsing in the Geotech forums. You need to go to www.longrangelocators.com to discuss these pseudo-scientific devices.

I presume you did not read Carl's sarcastic comment at the end of the article: "If no obvious target is found when you reach 200 feet, then the target was most likely subatomic gold particles - you cannot see it and most chemical analyses will not detect it either, but the MFD will."

Basic Rules of the Forums

[ford]

VLF principle is the best
type to discriminate metal, it
sends out a constant magnetic
field, and receive it in another coil.
When the received field is
changed in the phase there must be
a metal close. The amplitude of
the signal tells how deep / big the
object is.

[ford]

it's true,
In frist, i must obtain information about target,

Sample this information:

The most common frequencies used are 5KHz for gold and 8.7KHz for silver; other elements have different frequencies.

http://www.geotech1.com/cgi-bin/pages/common/index.pl?page=lrl&file=/projects/mfd1/index.dat

[Gwil]

Hi ford
　
You will have to think about exactly what basic information the detector can obtain about the target, before it is processed by the NN. There is the amplitude and phase of the target signal at the transmit frequency or frequencies, and possibly the way these change as the search head moves. I think that's everything?

Good luck,

Gwil

[Skippy]

This concept will have some niche applications, maybe autonomous UXO-detection equipment. But for the hobbyist, or the gold prospector, it's likely to be simpler to use the neural network between ones ears, and put design effort into presenting info to the ears in an intelligible manner.

[ford]

Metal Detector Based on Genetic Algorithm:

[ford]

I will try to do this!
my professor at the university maybe help.
He is Top scientist in 2004 by the institute ISI.
Ph.D. in Electrical Control from the University of the USA.

I would like use members experience for this idea.

Regards

[Koala]

Not sure how it would help for relic hunting
A small signal near the surface or a large signal deep. Until you dig it out the detector has no way of knowing the difference.
There are infinite size shape depth.
The best discriminator is your spade.

[Orbit]

Something like EEG

[ford]

Good


But this (THESIS2.pdf) refers to the detection of metal contaminants in a mass production industrial environment.


Here we can develop the idea of a new building for public use.


I simulated NN in the MATLAB software for classifying forms.


If we can define many of the properties of each metal, NN can classify them.


For example:


We use a lot of reference signals and use it to send it to the environment and we use a feedback signal for analysis.
Each metal has a specific frequency.


Tell me if you think this is vain.


regards

[Qiaozhi]

Hi all

I have one idea !


For discrimination and detect we can use Artificial Neural Network - ANN.


NN can classification targets and patterns.

Neural networks have been used on a variety of tasks, including computer vision, speech recognition, machine translation, social network filtering and ...

https://en.wikipedia.org/wiki/Artificial_neural_network

Regards

This is not a new idea.

[ford]

Hi all

I have one idea !


For discrimination and detect we can use Artificial Neural Network - ANN.


NN can classification targets and patterns.

Neural networks have been used on a variety of tasks, including computer vision, speech recognition, machine translation, social network filtering and ...

https://en.wikipedia.org/wiki/Artificial_neural_network

Regards