AI Used For Metal Detectors

[Gold Ryder]

51 minutes ago, Aureous said:

The term you used was 'Meanwhile they remain prime for a hostile takeover' which anyone would interpret as a corporate move to buy the company. But yeah, we all feel your frustration. We want new tech in a 'timely manner' (11 years for the Z7K) but aren't seeing it. 

However, Nokta are playing a game of 'catch up' and Algo are a new company. Garrett have created a good detector (Axiom) that would have been very competitive if released 10 years ago. FTP are living in lala land and at the same time, Minelab have designs, concepts and plans reaching many years into the future. I don't see any other company able to achieve what we desire. Nokta is going to be the closest I think. Even if they produce a detector as good as the Z7K in performance, at half the weight and with twice the coils....and at half the price....it will be the closest competitor Minelab have ever faced. The E1500 is only beating Minelab in the SDC mid-range section of the market. 

In the meantime.....we wait. 

I keep hearing that. But I yawn in the meantime and in the meanwhile we get held hostage by Minelabs bottomline. The times are changing. Look at the mods. Its cremating the 6000 in the right hands, including mine and as goid or better that the 7000 in others.  Tech led by AI is gonna hit lightspeed faster than we all realize. Mark my words. Its not gonna take much for a new heavy player to step in and grab big market share.  Minelab better have a serious ace up their sleeve and they better do it soon. Id wager you a big fat nugget on it.  

[Aureous]

@Gold Ryder I couldn't agree more. AI will indeed play a significant role in the near future with detectors. The 'heavy players' are already well into it. I don't wager any 'big fat nuggets' on a sure thing 😉 

[phrunt]

I don't think so, small market, smallish businesses making the detectors, not giant budgets and a small market of customers, doesn't bode well for massive investment required to reinvent the wheel.  

Some other big player would have to be interested in doing it, that's very unlikely.

Minelab are hunting for an engineer at the moment, and always seem to have jobs available in engineering but anything to do with AI is never a part of the requirements if at all from the jobs on offer I've seen.  I would think they'd need much fresh blood to take on an AI approach.

A current job on offer mentions machine learning or similar which can in a way be classified as AI but not a big part of it and not likely the plan they have to be using proper AI for detectors, too many variables and not enough absolutes.

Title:  Signal Processing Engineer / Technology Physicist

Location:  Australia, Adelaide, AU

Minelab is a technology company that creates world-leading metal detectors to help humanitarian organisations clear landmines, gold prospectors find life-changing gold nuggets, and historians to recover lost relics. We have a culture of innovation which allows us to be a leader in bringing new technology to customers across the globe. We make everything ourselves; creative design and detailed engineering are the lifeblood of developing new products at Minelab.

Excitement in the Engineering team comes from converting raw ideas into products that exceed customer’s expectations including rapid concept prototyping, architecting, detailed design, product integration, and the transition to manufacturing.

Join Minelab, part of the Codan group, an ASX listed technology company, and build the next generation of world-leading products.

What will you be doing?

Engineering, design, and innovation are the core of Minelab’s sustained global success. 

The successful applicant will design sophisticated signal processing algorithms for Minelab’s metal detector products. You will be working with high-fidelity sensors to detect and classify faint signals in challenging environments.

You will be improving our feature engineering, classifiers, and sensor calibration algorithms, identifying novel approaches and sensor concepts that can drive real value for our customers. You will contribute to all phases of the development lifecycle and engage in a broad range of efforts required to ensure that algorithms are operating on high-quality signals - this can include getting new system prototypes up and running and performing data analysis to validate system performance.

What makes a successful Signal Processing Engineer / Technology Physicist?

To be successful in your application for this engaging opportunity you will require the following attributes:

Bachelor’s Degree or higher in electronic engineering, physics, computer science or similar technical field

PhD in relevant field is highly desirable

Proficiency with numerical computing (e.g., Python scientific stack)

Experience with statistical signal processing, digital signal processing, sensor processing algorithms, machine learning or related areas

Strong grasp of statistics and linear algebra

Experience with C/C++ on embedded devices is desirable

Understanding of electromagnetism and electronics is desirable

[Aureous]

26 minutes ago, phrunt said:

I don't think so, small market, smallish businesses making the detectors, not giant budgets and a small market of customers, doesn't bode well for massive investment required to reinvent the wheel.  

Some other big player would have to be interested in doing it, that's very unlikely.

Minelab are already doing just such work. ALL new detectors cost million$ to develop. ML spent over AU$20M on the 6000 alone. So dont imagine for a second that the budget wont stretch to AI tech. I'm sure both Nokta and Garrett are also busy with what/how AI can be integrated. They would be 100% totally nuts if they weren't. A lot of work is done via contractors and implemented in-house once a working bench tester is completed. Plexus have a large AI division doing experimental work for dozens of different companies. Just coz it aint happening in ML headquarters (at first glance), dont mean it aint happening elsewhere.

[phrunt]

A good read on the history of AI. The History of AI: A Timeline of Artificial Intelligence, it's just become a mainstream marketing buzzword lately with everyday electronics finding ways to incorporate it to drive sales, and it's been designed to meet those needs with the generative AI everyone's now messing around with and a few years ago nobody mentioned it in regards to detectors until it hit the gadget world.

Maybe we'll be saying "Hey Minelab, ground balance my detector" and the detector reply, Ok, I have ground balanced your detector, you can now look for gold" in a few years' time as the gimmick side of it, and maybe they'll use it to benefit the detector in some way, discrimination maybe? improving ground handling? I don't know, but I just don't see it happening.

 

[VicR]

I agree that AI has a way to go. I zoomed in on Google Earth on a patch of  ground at Tibooburra, New South Wales that had a very heavy covering of white quartz which is quite common for the area and ask AI to comment on the picture - AI replied there was hail on the ground.

[Aureous]

26 minutes ago, phrunt said:

A good read on the history of AI. The History of AI: A Timeline of Artificial Intelligence, it's just become a mainstream marketing buzzword lately with everyday electronics finding ways to incorporate it to drive sales, and it's been designed to meet those needs with the generative AI everyone's now messing around with and a few years ago nobody mentioned it in regards to detectors until it hit the gadget world.

Maybe we'll be saying "Hey Minelab, ground balance my detector" and the detector reply, Ok, I have ground balanced your detector, you can now look for gold" in a few years' time as the gimmick side of it, and maybe they'll use it to benefit the detector in some way, discrimination maybe? improving ground handling? I don't know, but I just don't see it happening.

 

I see what you're saying Simon, but this is only a skin-thin layer of what machine learning is theoretically capable of doing. Elec engineers are saying that AI can become a mega processor of information to determine gateways for signal processing, energy processing and much more. In terms of a metal detector, the AI chip could potentially auto-manage all your settings, adjusted 100% to the prevailing ground conditions and the coil. The overall return signal could be analyzed, tiny minute metallic signals plucked from the overall return signal, ground noise eliminated completely and power to the coil increased among a multitude of other potentials. With all of this processing power being done automatically, the detector could be boosted to ridiculous levels in theory. There is no gimmick needed, the AI brain simply works in the background and you wont hear a peep from it! This is the AI potential I am thinking about.... not a computer voice telling me my GB is finished. Its a whole new world of 'maybe's'

[jasong]

 

10 hours ago, phrunt said:

Minelab are hunting for an engineer at the moment, and always seem to have jobs available in engineering but anything to do with AI is never a part of the requirements if at all from the jobs on offer I've seen.  

-------------------------------------

Experience with statistical signal processing, digital signal processing, sensor processing algorithms, machine learning or related areas

It's in the job posting you quoted. A lot of that signal processing stuff can be related too. This stuff is probably specialized AI though, not chatbots. The linear algebra requirement is probably another indicator. 

Separately, and in general - right now the better usage of the LLM's which most people think of as AI today is training local models for research on a home computer with a good GPU. I've had some early success with prospecting-related adaptation lately now that multimodal local models are getting powerful enough to be useful. I will show people and let them use it when I get enough time to finish things. 

It requires a lot of work fine tuning a model specific for the purpose though, not just using an off the shelf AI and asking it general question. I've spent probably 25 days generating my own custom training datasets, and that's just a drop in the bucket. 

[IBMe]

> Python scientific stack

Very interesting. If they were hot on AI, it probably would be mentioned here. AI takes a guess at what is known. If you’re pushing the edge, you’re probably ahead of AI.

https://sagol-python-for-neuroscientists.github.io/textbook-public/classes/class_4/class_4.html