Artificial Neural Network - ANN and detectors

[Dave J.]

The operative word is "if". Nobody is going to bell the cat, because anyone who understands the problem understands that belling that particular cat has no return on investment. The cat will eat the mouse alive.

But suppose I'm wrong, and someone does bell that cat? They've got a metal detector they've invested ten years of their life in, the underlying technology is obsolete, and it's been trained "in a particular ground location" that the trainer cleaned out as a necessary part of the training process. Completely useless even to the guy who built the thing. Thus does the cat eat the mouse even if the mouse bells the cat.

[bbsailor]

A few years ago, 01-20-2013 I made a post describing how microprocessor controlled metal detectors could improve detector performance. See this link: http://www.geotech1.com/forums/showt...light=bbsailor, post#36. If you add ANN to this, it is easy to see how with a recorded history of targets found in a particular ground location with actual depth located, ground moisture and temperature recorded, this could provide a more optimum selection of coil size, coil type, sweep speed, TX pulse width, delay, and detection depth for a class of targets with a known range of time constants.

If standards could evolve, then people could share their optimum settings for target type and size, soil conditions, optimum coil selection size and TX/RX parameter settings.

Joseph J. Rogowski

[Dave J.]

You're missing the point, moodz. replaced the word "used" with "trained". Who will bell the cat? http://mythfolklore.net/aesopica/milowinter/6.htm

Most of us either own dogs or know people who do. There's a saying that in order to train a dog, first you have to be smarter than the dog; and this is why dog training almost always fails. .......Same principle with neural networks.

[moodz]

A neural network could be used to detect a signal in noise. note that your ears are connected to a neural network.

[Koala]

Not quite sure what we are trying to achieve.

A modern coin or a 2000 year old coin it makes no difference I will dig them both.

Bits of copper, brass, lead or silver. Again makes no difference. Dig them all.

Anything that doesn't give a signal I can't give any feedback.

[ford]

We need experimental data to train the neural network
Experimental data much help
Even fuzzy logic
For example, with fuzzy logic
Can be roughly said
The properties of the metal discovered are close to what kind of metals
It may not be possible to determine the metal type by 100%
But the proximity to the kind of metals can be determined

[ford]

WASTE OF TIME !

This is a great goal
Lack of knowledge in this field
Leads to more research
And in these studies
Good results will be achieved
In this context, I will try

[Orbit]

WASTE OF TIME !

[Koala]

A typical detecting day would involve digging between 50 and 100 targets. Rarely are two items the same let alone the same depth and orientation.

If two targets are the same often the signal is different due to iron near by.

[Dave J.]

Eric's right. Fixed industrial applications usually afford a sensing environment where a neural network can be trained. Not saying that a neural network is necessarily the best way to operate the thing, but it's one of the ways and may be the best in some situations.

That's why I narrowed the scope of my post to "in the case of hand-held metal detectors for detecting unknown buried objects". That one really is a showstopper for neural networks.

[Ferric Toes]

Hi Ford,

See PM re an industrial application.

Eric.

[Dave J.]

It doesn't matter how good the neural network is or how sophisticated the processing of the incoming signals is, if you don't have a method for training the neural network you don't have anything. In the case of hand-held metal detectors for detecting unknown buried objects, nobody is ever going to go to the effort to train the neural network. If you think someone will go to that effort, you don't understand the problem.

The idea isn't new. I was kicking it around in the 1980's. The technology has come a long way but the basic problem is still that same one.

[ford]

This book have good information for NN and metal detectors:

(Next Generation of Metal Detectors)

http://s000.tinyupload.com/index.php...96665145846739


we read:
.... It is found that different materials,
have distinct amplitude and phase values. The
amplitude and phase values are also shown
to depend on size and frequency, which leads
to the conclusion, that the frequency can be
chosen product optimal. The results of the signal
analysis, are then used for developing methods
for metal detection and classification....

[sled]

Still don't get the feed back.

Would you have to manually input the depth, size, weight, shape, orientation, and material of each recovered item. Otherwise how would it learn.

Yes you need labelled training data for supervised learning, and with a metal detector on the field this is very tedious work and you need a lot of samples. For a proof of concept I'd start with something simple like whether it's possible to reliably classify an object as silver dollar under lab conditions.

[Koala]

Still don't get the feed back.

Would you have to manually input the depth, size, weight, shape, orientation, and material of each recovered item. Otherwise how would it learn.