Jump to content

New Algorithm Reduces Noise


Recommended Posts

Is this the future of detecting?  I think so.

Take a set of data points and process them (just as we do with our brains) and voila!  It doesn't have to be magic when you have good science to make it repeatable.  

Watch out missed nuggets!

Currently, a type of software based on a machine-learning algorithm called deep learning has been shown to be effective at removing the blurriness or noise in images. These algorithms can be visualized as consisting of many interconnected layers or processing steps that take in a low-resolution input image and generate a high-resolution output image.

 

 

https://scienceblog.com/520757/smart-algorithm-cleans-up-images-by-searching-for-clues-buried-in-noise/?utm_source=feedburner&utm_medium=email&utm_campaign=Feed%3A+scienceblogrssfeed+(ScienceBlog.com) 

  • Like 8
  • Oh my! 1
Link to comment
Share on other sites


i agree, high energy light beam quality and beyond, with greater pixel definition(s) and enhanced visuals. the more clues science can give us the better

  • Like 1
Link to comment
Share on other sites

Data points can be in any form.  They can be sound, light, magnetic, etc.  If you already have a sensitive collector of data this algorithm will make it see better.

Now you have to bring these ideas into market perspective.  When would Minelab obsolete all of its detectors in favor of a technology this powerful?  Not any time soon and by that time many of our nugget patches will be nearly barren but that makes it easier to find a few targets rather than a few targets in a lot of trash.  Everyone can imagine ... 

Steve is right.  It takes innovation a while to be incorporated.  Many of us here grew up watching the 'Space Race' and were told that we would benefit from all the money spent.  We have.

Chet has probably seen much more than the rest of us and he can't tell ... even after he has been retired.

  • Like 2
Link to comment
Share on other sites

This is a type of a neural net AI. They can be extremely versatile. My roomate back in college was doing his PhD thesis on them. They can be open source and put onto a chip and set to learn and solve an almost infinite number of problems. Sometimes the most efficient solutions they come up with make no sense to humans we almost certainly wouldn't come up with them ourselves in that way.

To understand just how versatile they are, recently a different type which was named AlphaZero was introduced to chess. Starting from nothing but the basic rules of the game, it did nothing except played games against itself and within 24 hours it was strong enough to beat the best human chess player in the world. In 3 days it figured out new ways to play the game and beat the strongest computer chess program humans had written in all the decades we've had computers. Now there is talk about pointing this very same AI to the basic rules of physics and biology to see what it comes up with.

These AI's can easily be put on a chip that are affordable (like $20), especially smaller and more specific ones. Though the really powerful ones still run on supercomputers.

I've been trying say for well over a decade now that there are tons of places that detectors could improve with modern techology and a company that simply is willing to pursue stuff that is at this point, "old news" in the tech world. That's ignoring completely relatively novel stuff like AI which alone could be a new frontier.

It's hard to really put it in perspective how far behind the tech curve detector companies are compared to really any other segment. They are dinosaurs. Even some of the Minelab stuff.

  • Like 3
  • Thanks 2
Link to comment
Share on other sites

The bottleneck may be the amount of processing power needed.  Steve H. mentions the speed at which that can be minitiarized and made cost effective (calculator analogy) but it does depend upon demand.  Dedicated processors (as opposed to general purpose supercomputers as was used for the Google AlphaZero work Jason mentions) will help for the simplification/miniturization, but can that be simply borrowed from another more in-demand application?

CCD cameras are another example of progress.  They used to be mega-expensive for modest (in today's world) 1 megapixel packages.  Now every cellphone has many thousands of times better resolution and it comes effectively for free(?) as part of the deal.  But again, the demand for digital cameras (the intermediate evolutionary step) was high.  Metal detectors?  Not so much.  That's why it will need to piggy-back off of an in-demand application.  But for sure the potential is there.

  • Like 4
Link to comment
Share on other sites

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

A Raspberry PI 4 at $55 is a full computer, with negligible weight gain, in which you can run an AI plus interface sensors (like a metal detector coil). It has a 64 bit quad core CPU, wireless, USB 3, etc. 

Shows just how cheap powerful electronics can be built today. There are competitors and various other peripheral devices built by essentially garage based, hobby companies for about the same price. A startup detector company could compete if they had good, highly skilled engineers and scientists, and thought outside the box.

AI aside, just looking at the world of stuff you can do with the massive amount of cheap computing power avaiable to even hobbyists today - Minelab appears to be the only company looking at software solutions, based on their patents I've read so far. So again, the rest of the pack will find themselves left in the dust with nothing but a patent wall to contend with when they finally wake up to modern technology in another 10 or 15 years. 

  • Like 2
Link to comment
Share on other sites

Jasong,

This is the stuff I thought might be on the chips in the Equinox and it could be on the 6000 if they wanted it but then what would they do for the next generation detectors?

Maybe Geo Sense is a marketing name for an algorithm!

  • Like 2
Link to comment
Share on other sites

I think the physical detecting principle would have to be changed or enhanced somewhat for this to really change the treasure hunting game.  I don’t think magnetic induction principles alone can achieve huge gains at this point even with AI applied.  A combination of induction balance or PI for conductive metal detection combined with compact/low power/high resolution Ground Penetrating Radar would be killer for deep, larger cross-section targets.  Magnetic induction would still be needed and would probably dominate for detection of small targets and tiny natural gold.  

Also use of AI and Augmented Reality combined with ultra precise ground mapping would also be a useful tool.  Imagine donning a pair of Augmented Reality glasses that could enable you to see where your coil has actually been to ensure complete ground coverage at a site, visual target logging (to ID target density/concentration).  I mean how many targets are not recovered simply because you didn’t get the coil over the target and not due to an inherent limitation of the machine.  The ultimate objective would be finally giving you the “X Ray” vision ability to actually peer into the ground and “see” buried targets in situ and in real time and without a semi-trailer of electronics needed to achieve that goal.

  • Like 4
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...