Pattern Recognition Software For Use In Identifying Fossils?

[smileylizard]

Hi, I'm new to The Fossil Forum. I've noticed surface features on dinosaur bone fossils that exist on species without regard to the scale, mineral makeup, or geophysical location of that item. Striations, bumps, curvatures, texture, etc. I was just wondering if there is a recommended application or graphics viewer which contains an algorithm which could be used to more easily identify and confirm various fossil discoveries? I started looking at DigiKam, but decided to ask here before going down a potentially immersive tangent. even outline of shape would seem beneficial to me at this point...thanks for your time.

[FossilDAWG]

I'm not aware of any efforts in this direction. It seems to me it would be possible to do this, but I personally don't see the point other than as an interesting exercise in programming. You would have to have a set of images to train the program, and another to test it, and all would have to be samples already known to be bone or not bone, and that judgement would have been made by humans. The human brain is marvelously adept at recognizing patterns, sometimes too good (such as when we see faces in clouds for example). What would be the useful purpose of a bone recognition program? To be useful it would have to be able to classify specimens that are so ambiguous an experienced human observer would be unable to decide, and such specimens would likely be so marginal as to be uninteresting.

I hope you will forgive me for sounding skeptical. I'm the kind of person who happily uses technology to do things I can't possibly do on my own, but if I can do it myself I prefer that. For example, I routinely use computers to compare millions of DNA sequences, and to assemble transcriptomes and even genomes, tasks that would take lifetimes of boring work to do "by hand". On the other hand I love to drive and am not at all looking forward to the day of the self-driving car, when human control will be considered too risky and we will all be confined to the passenger seat.

Don

[Doctor Mud]

Sounds like quite a challenge. I can see a use for automated identification if many IDs of similar fossils, such as microfossils were required.

I know that palynologists who count fossil pollen have been developing a technique for automated counting of fossil pollen grains.

This could be useful since hundreds of grains need to be counted to get an accurate impression of vegetation from the past.

This was a university in New Zealand in Palmerston North I believe.

The difficulty with the fossil pollen is that sometimes it is fragmentary, folded, at different angles so in the end they found that their machine was best for counting pollen in honey.

Sounds strange, but you can pay a lot for genuine manuka honey and it has to have a certain % of manuka pollen to be deemed authentic.

Good luck with your technological quest.

[painshill]

There is perhaps still untapped potential in what are known as “expert systems”. They’re designed for computers using the same kinds of algorithms used in traditional botanical and zoological “keys” where you follow a sequence of questions designed to lead you to a species identification: “Does the peripheral squibnocket have three interstitial papillary lobes? If so, go to question 8” and then: “Is the anterior progenital operculum wider than 1cm? If so, go to question 43” and then ultimately there’s a “bingo” moment when your specimen is identified.

The problem with those kinds of keys is that you need to know your way around the subject of what you’re trying to identify and fully understand the terminology of what is being described before you can use them. They're usually text-based with not much visual assistance. They’re also often family-specific, so you may need to know you have a Hymenoptera rather than a Diptera before you can begin. The advantage that an expert system brings is that the questions are “dumbed down” to a vocabulary that anyone can understand and can easily be supplemented by simple explanatory diagrams, example pictures and so on. We use systems like that at the company I work for in a manner that means someone relatively unskilled can perform a task way above their skill-set because the system is providing them with expert knowledge and experience they don’t have themselves.

I was fascinated by an inexpensive toy about the size of a cell-phone I bought as a kid’s gift a while ago. It claimed to guess any object you thought of by asking no more than twenty yes/no questions. The questions were wacky but simple to answer (Is it warm to the touch? Does it make a noise? Is it bigger than a duck? Is it made of wood?) and it got the answer right with remarkable consistency even when I thought I had selected an obscure item. An expert system is effectively a sophisticated version of that.

I recall there was a pilot expert system for fossil identification developed in the UK in the 1980’s (called E-Fossil or something like that) but I don’t know what became of it. I’ve never seen that kind of software subsequently offered for amateur use although British Petroleum developed a system called “Vides” (Visual Identification Expert System). It’s used for identifying microfossils in drill cores (foraminifera, dinoflagellates, radiolarians, ostracods, conodonts etc) down to species level as an aid to dating rocks and determining the depositional environment (and hence the prospect of striking oil). There’s no pattern recognition involved but it “de-skills” the identification process as well as reducing the donkey-work. There are more than 100,000 species of foraminifera alone for example.

Edited October 29, 2014 by painshill

[Auspex]

Who knows where 'machine learning' software may go? This digital infrastructure is currently producing encouraging results in birdsong recognition (still a ways to go, though). For something like fossils, though, any practical application may have to wait for the development of full-blown neural network programming.

[FossilDAWG]

I saw a newspaper article today about how in the US alone 2.4 million jobs are threatened by robots currently under development. One such robot, for example, once fully developed (apparently not far off) will be able to navigate around a house, clean, do laundry, prepare meals, and administer medications, effectively replacing live-in nurses.

It is commonly argued that robots would only replace workers in low-skill, monotonously repetitive or high-risk jobs. However, we've already seen 90% of bank tellers replaced with ATM machines, and it's expected these days that you'll scan your own groceries and bag them yourself. There are few enough jobs for paleontologists of any stripe these days, yet we're talking about replacing micropaleontologists with expert systems that can sort and ID microfossils, and even automatically generate a report about depositional environment and age, no doubt including correlations with other rock units and drilling recommendations.

Airline pilots these days rarely actually fly anything. Rather they are expected to sit in front of the instrument panel, scanning gauges but touching nothing for hours on end. In theory they will step in if an emergency arises the computer can't handle. We've already seen crashes happen because pilots don't remember how to fly and just trust the computer. Humans have not evolved with the ability to sit and watch for days and weeks on end, doing nothing but watching dials and gauges, and still keep total focus. This sounds like some form of medieval torture to me.

The same fate awaits anyone who enjoys driving. Your days are numbered.

I doubt there is anything humans do as a gainful occupation that could not eventually be built into an expert program running a robot. Nothing productive will be left for humans. Our only job will be to consume. Hopefully not each other.

Sometimes I think there may be a lesson to be learned from the QWERTY keyboard. Initially typewriter keyboards were designed so that frequently used letters were grouped together for efficient typing. Too efficient, as it happened. Typists quickly became so fast the metal arms of the old manual typewriters could not jump up, strike the ink ribbon, and return before the next arm/letter jumped up, so they would lock together and jam the machine. The solution was to arrange the letters inefficiently, to slow the typists down. Paradoxically in this case slow is fast, as the pace was now slow enough that the machine did not jam, allowing more words/minute to be typed. It's interesting that we're so invested in the QWERTY keyboard that we still use it for computers, even though computer keyboards can't jam and other configurations would allow more efficient typing.

What would be the lesson? Sometimes efficiency in one thing needs to be sacrificed to achieve better overall efficiency. In an economy, individual businesses may benefit from increased efficiency, so they can produce their product more cheaply and maximize profit. However, there must be a market for the product, and the market is people who not only want but can also afford the product. If one manufacturer can automate it's production facilities it may be able to reduce costs and increase profit. If every occupation is filled by robots, mechanical or computer, with very few people actually employed, markets will disappear along with the economy. Maybe we would be wise to choose some path that appears less efficient from a production cost standpoint, yet works better overall because it preserves a market for the goods we produce.

In the meantime, I find learning to identify fossils fun and challenging. I don't mind help from computers, such as access to papers or keys that help guide the IDing process, under my control. An expert system that takes a photo, analyses it, and spits out a correct ID in a couple of nanseconds would kill the hobby for me.

Don

[tmaier]

I've written a script to automatically reply to threads and insert funny stories that sometimes have a point. I have the bugs mostly worked out, and it has really elevated my lifestyle by freeing me from reading the threads.

the threads

the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads the threads

WATCH OUT! IT'S GOING TO BLOW!

[tmaier]

Actually, I'm an electrical engineer and my specialty is design of automated systems like you are describing. There are software packages that are easy to use for counting objects, once you train the software what the objects look like. It's fairly expensive for an individual, but for an automation line in a mid sized company it often pays for itself over hiring a person or designing an electromechanical gizmo that might wear out. Also handy for laboratories that can afford it. I don't know of any open source versions.

I got into this back in the late 1970's. At that time the US was the world powerhouse of manufacturing, and a lot of it was done by hand. It was realized back then that if we didn't move to automation, the factories would move to countries where the labor was cheaper and the US would crumble, or have to find a new way to make a buck. Well, here we are and most of the manufacturing HAS moved overseas, and we never did fully automate.

[Auspex]

Technology is, and has always been, about efficiency. The meaning of "efficiency" itself is a chimera, shifting form every time a new technology is brought to bear.

H. sapiens used to get plenty of exercise, but labor saving devices made that harder to do and time saving devices somehow made it harder to find the time to exercise. Enter wearable electronics that monitor your heart and respiration rates to 'help' you exercise more efficiently...most of these devices are manufactured roboticly.

[FossilDAWG]

In one of those odd but interesting coincidences, as I was driving home this evening I caught a segment on NPR (similar to the BBC or CBC for those not in the US) that seems relevant. They were talking about the difficulty manufacturers have hiring people with the needed technical skills. One company they discussed in particular manufactures robotic welding machines used to assemble turbines for jet aircraft engines. These robots are the size of a building. The company has about a dozen positions they can't fill because they can't find qualified people. The problem is, twenty years ago the factory employed over 2,500 people, so it was cost-effective for the company to run its own "school" to train people to work with their highly specialized equipment. Now the manufacture of the welding robots is itself so mechanized that the work is done with a work force of only a few dozen people (down from 2,500), and it isn't cost effective for the business to maintain a training program. It's an interesting conundrum: the company says it isn't efficient for them to train workers, and they think high schools/technical colleges/universities should be training people. On the other hand, educational institutions don't find it efficient to run whole programs to train students to use the highly company-specific machines when there are so few jobs for their graduates. I hadn't really thought of that before, but it makes sense: as work forces go from thousands to mere dozens to run a factory, and those few jobs become highly specialized, it becomes both expensive and inefficient to maintain the teaching staff and equipment to train workers.

I'm all for using technology for tasks humans can't do, such as managing huge amounts of data (assembling genomes from millions of DNA sequence reads for example) or manipulating large and heavy pieces in a manufacturing stream. In the past, it seems that as new technologies came on board they brought new jobs, which replaced old jobs. When cars replaced horses there was less work for blacksmiths but many more jobs for auto mechanics. I don't see that happening today, rather it seems automation is allowing a handful of people to oversee the computerized/robotic technology that has displaced thousands of workers. An example from my own line of work: twenty years ago DNA sequencing was done by highly skilled human technicians. When the human genome project started it employed several thousand technicians. Due to the need to generate a huge amount of sequence data quickly, robots were built to replace the human technicians. Today when I need to do a sequencing project, I send my samples to a large sequencing facility. I put the DNA samples into tubes, stick on a label with a bar code I get from the facility, and go online and enter all the necessary information about the samples before shipping them by overnight courier. The next morning when the samples reach the facility, someone opens the envelope and dumps the tubes in a hopper. This is the only time a human touches the samples. The tubes are loaded automatically into holders, the bar code is read, and the tubes are shunted off to the appropriate machine set up to run samples like mine. Usually by noon I'll get an email with sequence data for all my samples. The system is highly efficient, and so quite inexpensive (about $5 a sample), because the facility receives a constant stream of thousands of samples/day from researchers all over the country. Twenty years ago it would have taken four or five days to get the data, and probably 5 or 6 technicians would have worked on the samples. Today it takes less than 24 hours from putting them in the mail to getting results, but the only human to touch them is the very low-skilled minimum wage person who slits open the envelope and dumps the samples into a hopper. No jobs have been generated for those redundant technicians, there is no room for humans in the process any more.

I enjoy trying to identify this echinoid or that trilobite, and would prefer not to have a robot do that for me too.

Don

[tmaier]

Schools should NOT be training young people for specific jobs in industry. That is a really lucrative idea for industry, but a bad one for the workers. Secondary schools need to stick with general skills of reading/writing/rithmatic and if industry wants workers to know their product, then it is their responsibility to teach them. If a student is taught a specific product line, then it locks that student in to working for that one company. You are not teaching the student, you are grooming them for slavery to one master. It locks the student into a future abusive relationship. The worker needs the skills to move around and sell their skill set to many employers, if we are going to use this free market system. A very specific skill set makes the worker no longer free, and the employer will exploit that.

I've designed a lot of things that have wiped out a lot of jobs. When you see it happen first hand, you feel dirtied, but the jobs that were replaced were minimum wage, very repetitive, sometimes dangerous, dead-end jobs. On these types of jobs the person can't make enough money to live life. The people who need to operate the machines and repair them make several times more money, enough to live a decent life. So 12 people lose their jobs and two people get a job at three times minimum wage. Now they can eat, live in a house, and have a family. That wasn't really possible with the minimum wage jobs. Also, the minimum wage jobs were so repetitive and boring that the workers would show up drunk or stoned. The job is a torture to them. The employer treats these minimum wage workers with disrespect because they are easily replaced. When a worker has a valuable skill set then the employer has to behave themselves because abusing the worker could result in having to find and train a new one.

When an employer invests time and money on training an employee, it creates a parity between employer and worker. Without the parity then the relationship devolves to become abusive. Mutual respect in required for a healthy working relationship. Now that the unions have been pretty well crushed, the only path to parity is when the worker has a skill set that is valuable to the employer and that creates a barter system between the two, and that can't be achieved with unskilled, minimum wage labor jobs.

[JohnJ]

Labor relations aside, I think the parameters involved in developing pattern recognition software to identify fossil material to the Family level would be practically impossible to achieve. Think of the variations that occur within a single species of vertebrates.... It might be 'easier' to accomplish with invertebrates, but a project like this would be an enormous undertaking with very limited application, imo.

[tmaier]

Yes, many things that are possible are never actually accomplished because the profits aren't there to make a living. The system would have to be trained and that takes a lot of time and effort, but nobody wants to pay for that.

Having designed a lot of automated systems, I can tell you that the way you start something like this is to observe the people who are doing it manually. What inputs do they use? What is the logic they use (algorithms)? Then you select actuators, sensors, and processing unit to mimic that action. I could see a system that would take the fossil and do a 3D scan of the surfaces, and then run through a database of patterns to resolve a list of possibilities. That may sound like science fiction to some people, but the autonomous systems being developed now are already doing that. It is proven technology.

There are cheap cameras in use in even the lowest of the automation industry that you can plug into your computer, take some images, and then train the camera to find certain objects in the field of view. Then you detach the camera from your computer and it will autonomously detect that object when it sees it. You can even define rules for that object, about how it should be orientated. These things only cost a few hundred dollars. Guess what you can have for 300 million? You can get facial recognition systems that can recognize faces faster and more accurately than a human.

The objects don't have to be perfect for recognition. They can be partially obscured and mutilated, rotated, and color can be ignored. You need to train it to pick up certain diagnostic points, just like a human does. Like I said, you mimic what the human does, and often it can be done much faster and more accurately.

So there is nothing out of reach of present technology, but who will pay to train this?

EDIT:

I'd like to summarize the above explanation by saying, if you can definitively describe how YOU do something, I can design a system to replace you.

The failures come when the task can't be well defined. That explains why "artificial intelligence" never happened. We never could define what humans are doing whilst being "intelligent".

Edited October 30, 2014 by tmaier

[Auspex]

... I'd like to summarize the above explanation by saying, if you can definitively describe how YOU do something, I can design a system to replace you. ...

My job security is rock-solid then; I have hardly any idea what I'm doing.

[painshill]

Don’t know if any of you have participated in “Zooniverse”, which is a collection of web-based “Citizen Science” projects that use the efforts and abilities of volunteers to help researchers deal with the flood of data that confronts them. The main website is here:

https://www.zooniverse.org/

From the comfort of your armchair, you can help scientists with classifications of various kinds, in association with specific projects. Each project has tutorials to familiarise and train you in recognising particular features before you can ultimately sign up to begin submitting real data. Naturally, behind the scenes, there are safeguards to protect against spurious data and to cross-check the performance of particular contributors for alignment. But, beyond that, you don’t really need to know anything (or at least, not very much) about the subject where you are making the assessments. Have a go with some of the tutorials. It’s fun.

For example, you can help the Habitat Camera Mapping System (HabCam) team and the Woods Hole Oceanographic Institution in classifying the ocean environment and its life-forms here:

http://www.seafloorexplorer.org/?utm_source=Zooniverse%20Home&utm_medium=Web&utm_campaign=Homepage%20Catalogue

Or you can help astronomers classify galaxies from images produced by the Sloan Digital Sky Survey (SDSS), the Hubble Space Telescope, and the United Kingdom Infrared Telescope (UKIRT) here:

http://www.galaxyzoo.org/

These kinds of projects are based on simplified expert systems that help overburdened scientists wade through quagmires of data, identify patterns, and ultimately focus their expertise on things that are interesting or unusual. For example the Galaxy Zoo project has been running since 2007 and in the first year alone, some 50 million classifications were contributed by over 150,000 people.

Edited October 31, 2014 by painshill

[fossilized6s]

I really don't see the point to making such a device. The best "recognition software" IMO is the trained eye. Knowledge of what you are looking at is hands down the best way to identify a fossil. This just seems like a lazy way out of something like reading and learning. So in a way we have this technology at our finger tips already, schooling, books, internet, etc. It just may take a bit more time. Plus in the world of fossils, the journey is half of the fun!

[Fossildude19]

I really don't see the point to making such a device. The best "recognition software" IMO is the trained eye. Knowledge of what you are looking at is hands down the best way to identify a fossil. This just seems like a lazy way out of something like reading and learning. So in a way we have this technology at our finger tips already, schooling, books, internet, etc. It just may take a bit more time. Plus in the world of fossils, the journey is half of the fun!

I agree Charlie. Couldn't have said it better myself.

Regards,

[Auspex]

I really don't see the point to making such a device. The best "recognition software" IMO is the trained eye. Knowledge of what you are looking at is hands down the best way to identify a fossil. This just seems like a lazy way out of something like reading and learning. So in a way we have this technology at our finger tips already, schooling, books, internet, etc. It just may take a bit more time. Plus in the world of fossils, the journey is half of the fun!

I fear that any 'point-and-shoot' device for identifying fossils (or anything else, really) would sidestep the most satisfying element of the endeavor: learning something about the organism and it's environment.