Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Drone-Based Photogrammetry and AI for Automated Arch Survey - Ep 121
214 Plays5 years ago
Dr. Hector Orengo joins us from Spain to talk about a recent paper where his team discusses using photogrammetry and AI to automate archaeological survey. It’s an interesting approach with promising results.
App of the Day
- Webby: Do Not Pay
Contact
- Chris Webster
- Twitter: @archeowebby
- Email: chris@archaeologypodcastnetwork.com
- Paul Zimmerman
- Twitter: @lugal
- Email: paul@lugal.com
Affiliates
Recommended
Transcript
Introduction and Sponsorship
00:00:00
Speaker
We're excited to announce that our very own podcasting platform, Zencaster, has become a new sponsor to the show. Check out the podcast discount link in our show notes and stay tuned for why we love using Zen for the podcast. You're listening to the Archaeology Podcast Network.
Guest Introduction: Hector Orango
00:00:19
Speaker
Hello and welcome to the Archaeotech podcast, episode 121. I'm your host, Chris Webster. Today, I talked to Hector Orango about archaeological drone survey. Let's get to it.
00:00:31
Speaker
Hector Orango has held postdoctoral positions at the CNRS in France and the universities of Nottingham, Sheffield, and Cambridge in the UK. And I'm going to butcher this here, by the way, but he is currently the Ramón y Cajal researcher at the Catalan Institute of Classical Archaeology in Spain, where he co-directs the Landscape Archaeology Research Group and an honorary research associate at the McDonald Institute for Archaeological Research of the University of Cambridge.
00:00:57
Speaker
He is a landscape archeologist specialized in archeomorphological analysis. That's a new word for me. Remote sensing, geospatial analysis, and the development of computational approaches to archeological questions. All right. Welcome to the show, everybody. Paul could not join me today, but hopefully he'll be on here sometime in the future. But for now, we have a great interview coming your way, as I mentioned at the beginning. So Hector, welcome to the show. Thank you. It's a pleasure to be here.
00:01:26
Speaker
Great.
Technological Advances in Archaeological Surveys
00:01:27
Speaker
So, so the topic of the show, of course, is I'll just read the title that I have here, combination of drone based photogrammetry and artificial intelligence for automated archeological survey. So.
00:01:37
Speaker
How did you decide to get into this and where did it come from? I'm a landscape archaeologist. That's my main... I mean, I ended up doing lots of different stuff, but my main interest is landscape archaeology and the such. I had to do quite a lot of harvesting during the last 10 years. And it became very critical when I started working in Greece, despite my... I had quite a lot of experience harvesting in Spain and in the UK and in France.
00:02:07
Speaker
But when I started siren in Greece, man, that was a whole different scale. I mean, we were siren in Thessaly some very early Neolithic sites like 8500 BP, which is okay. This would be Mesolithic in Spain. And the quantity of ceramics, I mean,
00:02:29
Speaker
to do a single line of servicing with a total counting of sets, it was around seven hours and the line was around 70, a bit less than 70 meters. So, you know, it was a torture and Tesla also is very, very hot during the summer, extremely cold during the winter.
00:02:50
Speaker
Yeah, I mean, we had health issues there. I mean, actually, we were using this cloud system to record our Potsai count. And, you know, the iPads and the mobiles were just shooting off. I mean, it was more than 40 degrees and they just couldn't stand, couldn't work with this hit.
00:03:10
Speaker
So that was, yes, yes. We thought that we had to devise a system to do this at least faster and, you know, at the very least, I mean, at this moment, we were not thinking so much of quality, but, you know, just to be able to do it in an reasonable amount of time, because, I mean, just to do the relatively small area we were working in Thessaly, it would have taken us many years on this rhythm.
00:03:39
Speaker
Yeah.
Challenges in the Field
00:03:40
Speaker
So I, I thought of this probably five years ago when I was working on a project in Southern California and it was on a Navy base. And I'll tell you what, man, I had a crew of eight people and they were employees of mine. And we were working on this Navy base that's used to test, um, to test ordinance, you know, like bombs and, and different things. Yeah. Right. It's been used for that for 70 years.
00:04:03
Speaker
And so we had unexploded ordnance technicians with us for each crew. We had two crews of four and each one had an unexploded ordnance technician with them because if we saw anything that was metal or we didn't know what it was, we were supposed to point it out and then they check it out to make sure it's not going to explode.
00:04:19
Speaker
And also there were rattlesnakes everywhere. And I just kept thinking, man, we surveyed 30,000 acres, but only about 1200 of it actually had sites that we recorded. So we spent 28,000 acres just waiting for someone to explode and get hurt. And I'm like, man, if we, and there's almost no vegetation too. So if we could have just done a systematic drone survey
00:04:43
Speaker
and then gone out and tested the hits. Maybe look at the video or photogrammetry and say, okay, there's possible sites here. Let's go out and ground truth those. Yeah. Yeah. Yeah. That's the point. Yeah. Exactly. So what's the vegetation like and what kind of challenges did you have with coming up with the drone-based survey in the areas that you're referring to? Because I know vegetation and terrain and stuff like that is going to always be an issue.
00:05:08
Speaker
Yeah, I mean, in fact, the paper went out in December, but we wrote it, I think, yes, in past February. So during this period of time since we made the initial submission and
00:05:26
Speaker
and the moment it was finally published, we have been working in these specific issues. I mean, the initial methodology was very much designed for landscapes like Thessaly. They're extremely flat, cultivation, they are under cultivation, so there are periods of time in which there's no vegetation and the visibility is very high.
00:05:47
Speaker
But it was evident that most people do not work in such excellent conditions. So our last developments actually include the programming of the drone to follow a specific height about ground. So when you are working in fact now, we are working in this in several areas with
00:06:14
Speaker
topographical conditions that are not that easy with the slopes and vine jars and fruit trees and some types of cultivation. And we have managed to develop a methodology that makes the drone to fly between the trees or between the rows of whatever is on cultivation and follow the slope. And OK, it's it seems to be working pretty well.
Methodologies and Tools
00:06:40
Speaker
I mean, that's probably our next paper.
00:06:43
Speaker
So, yes, that was a problem at the beginning, but now we have managed to solve it. Yeah. So, yes, basically what we do is just to make a preliminary flight, you know, a much higher altitude at around, I don't know, 60, 70 meters, something that, you know, is going to give us a good output in terms of time. With a single slide, you can cover quite an area and then you can get with photogrammetry, you can get a really high resolution digital theorem model.
00:07:12
Speaker
So, you have a model of the terrain, which you use to program. And at the same time, you have an auto-image photograph of the area. So, you can actually program the drone following the natural slope and the vegetation. And you can filter the vegetation, of course. I mean, the drone is not the image, it's just the photographs.
00:07:35
Speaker
So you cannot use a method to filter the vegetation such as you would use with LiDAR that actually can penetrate vegetation. But you can actually select colors like green leaves and a range of colors that you can simply eliminate and create a topography without these trees. So you can keep a constant height above ground when you program the drone flight.
00:08:02
Speaker
OK, so it sounds like you're looking for mostly features and structures and things like that, right? Things that you can not artifacts? Well, the initial methodology, of course, was directed towards a bot search. OK. But we have, yes, because it is what it was taking more time for us. I mean, with the structures usually, you can just go really fast until you find a structure and then you map it and this will take more time.
00:08:29
Speaker
But there are not that many preserve structures. In Thessaly, in fact, you could find a mound, but not much more than this. So yeah, that was the initial objective. But now we are, of course, we have this project in Aptira, in Trace, also in Greece, in which we have lots of tombs. And one of the easiest way to locate them is the distribution of stombs.
00:08:58
Speaker
So yeah, now we are starting to use this also for the distribution of stones, also lithics. And we are trying to move now towards the differentiation of specific portrait types.
00:09:13
Speaker
All right, cool. And what kind of, before we get too much further, I think this will help frame exactly what you guys are doing and how you're doing it when we start talking about limitations and things. But what kind of drone equipment were you using for this? This was a couple of years ago when you were building the data for the paper, correct? Yeah. Yeah. So what are you using for that?
00:09:34
Speaker
The initial idea of the paper, we wanted to create something that was practical, that most people could use without much resources. Because many of the people who do surveys, sometimes they don't have much money to spend.
00:09:54
Speaker
So we just use a commercial drone, a Phantom DJI Phantom 4. So yeah, it's extremely cheap because we wanted to do it with the cheapest possible. We wanted the method to be very, very easy to reproduce. So yeah, we use just this commercial drone that is relatively cheap. I'm not sure, but I think without extras, it's around $1,500 or $600. It's quite cheap.
00:10:23
Speaker
So, yeah, that was the reason and it worked pretty well, actually. Now, of course, we are developing our own drone for Sharvey. But, yes, initially with a commercial drone, with a good camera, you know, the Phantom 4, it's
00:10:40
Speaker
It's pretty stable, it's very secure, it's easy to program, and the camera has a resolution of 21 megapixels, which is excellent. Yeah, that's great. So yeah, it's a very good choice to start with this.
00:10:56
Speaker
Yeah, I've had, uh, I've had thoughts on that. I like to get your opinion on this as well for smaller scale drone surveys, something like a Phantom or another sort of quadcopter would be okay. But the limitations you always run into, of course, battery life and, uh, and things like that. Cause they're only going to run for about 15 minutes. Right. So yeah.
00:11:15
Speaker
One of the things, if you're doing large-scale survey and you have a really high-speed
Comparing Survey Techniques
00:11:19
Speaker
camera on board that can take pictures at a quicker rate but still at a high resolution, fixed-wing drones seem like they're better versus quadcopters, better for a larger survey, large-scale survey. Have you guys thought about developing fixed-wing for the larger surveys?
00:11:35
Speaker
Yeah, yeah. The problem with fixed wings is that I doubt we will be able to find a camera that can get enough resolution to record portrait search that, okay, take into account the ortho images we were generating for this survey were around where
00:11:57
Speaker
I don't know, one millimeter and a half, one millimeter dot one. So every pixel was... So imagine, you know, a fixed wing, it's actually using less battery because it's going fast. The wings allows, you know, less consumption of energy. So the trick is that it has to go at a certain minimum speed to be able to fly. And this speed is, we believe, is quite above
00:12:26
Speaker
what we would be able to. Of course, if you are not interested in pottery sets, but you want to record a distribution of stones, because you are savaging, let's say, a city, or, you know, ancient city, so, you know, yes, a fixed wing would be ideal, and they are fantastic models right now.
00:12:46
Speaker
Yeah, absolutely. I saw one, I think last year from a company out here that's, I mean, it's not cheap. It's like $20,000. It's not that expensive. They do have, I think, some fixed wing options, but like I said, they're priced out for most people in this field.
00:13:10
Speaker
Yeah. So, so how did, you know, after you guys did your, your drone surveys initially, how did those results compare with traditional pedestrian survey as far as accuracy goes? If you compare them with the standard, uh, traditional survey, that is you put lines every five or 10 meters.
00:13:27
Speaker
And you just want to do the line and record as much as possible, but you are not going to recover everything. It does so much better. It's much faster. We are talking here in terms of time per person. If you have 30 people, you are going to be faster than a drone.
00:13:49
Speaker
I mean, at least taking into account the whole of the workflow. But per person, that is for this kind of survey, you only need one person and you can program the flights at home. So you basically need someone that would take the drone to the field and just when it finishes, it changes the batteries.
00:14:10
Speaker
So it's so much faster and it's much more efficient. It's able to locate a significant percentage of boxes, a much higher percentage of boxes than a five meters, 10 meter line survey. However, if we also compare in our initial test, we also compare with our total collection survey. You divide the area in
00:14:42
Speaker
and then you put the person as many hours as necessary to collect everything. And okay, in this case, more pod research were found by manual survey. But of course, it was many, many, many hours of pod search collection. You have to, I mean, if somebody wants to implement this method, has to think that the more pod search, the higher the density,
00:15:06
Speaker
with a grid.
00:15:11
Speaker
the more efficient the method is because you are going to take the same pictures. If you have a density of one or 0.5 boxes per square meter, then you have 200 boxes for square meter. So the more boxes, I mean, it's ideal for areas in which the sites are characterized by high concentrations or ancient towns or this kind of
00:15:39
Speaker
where areas where you have a high density. The lower the density, the less efficient the method is. Okay. Yeah, that makes total sense. But I would like to point out something too, that something based on what you just said a little while ago was if you have say 20 people out doing a survey, it's going to be quicker than a drone. But then looking at all of the factors involved, and I do cultural resource management archaeology here in North America, and
00:16:06
Speaker
I'll tell you what, if I had 20 people out there, I've got other concerns. First off, I have to pay 20 people versus one person on a drone. And second, all those 20 people could get hurt. They could sprain an ankle, they could break a leg, they could, you know, something like that. Whereas if I have one drone operator out there and maybe a second person just for safety, that's what much more efficient, even if it takes a little bit longer. So speaking of speed,
00:16:29
Speaker
Yeah, I want to ask you about, so once you do the drone survey and you produce the photogrammetry, the photogrammetric images, and you've got this amazing resolution, how do you analyze the data on top of that? Do you create the 10 to 15 meter transects and keep that sample size the same, or is it a 100% analysis of the model? Yeah, it's a 100%
Data Analysis and AI
00:16:50
Speaker
analysis. We analyze the whole of the order image.
00:16:53
Speaker
And because, well, you can do it, it's not going to take more time, perhaps a little bit more time, but it would be more difficult to actually divide the image. Oh, sure. We just process the whole image. I mean, at the moment, we are doing this with Google Earth Engine, which is a free platform because we wanted to keep the workflow relative.
00:17:19
Speaker
Okay, we could have also used a photogrammetry, a free open source even photogrammetry program for this because I mean, to generate an auto images does not require all the hardcore computing you would require for that. Sure.
00:17:34
Speaker
through 3D reconstruction. So, you know, with Google Earth Engine, we could just implement our algorithms. Let's say the process we follow for the detection of individual pod search. And this is something that anybody can just register, copy, paste the algorithms and just upload their images and just press run. Wow, that's awesome. Yeah. All right.
00:18:00
Speaker
Yeah, I have to say nobody has, I mean, I know several people that have tried and, you know, they more or less manage. So it's not that difficult. I mean, most people get really scared around code, but we prepare the code in a very simple way. So and the explanations are good. So, yeah.
00:18:21
Speaker
There you go. I think I think it can be done. Nice. Nice. All right. We're going to take our first break. I have a bunch more questions for you and we'll do that on the other side of this break. Back in a second. Chris Webster here for the archaeology podcast network. We strive for high quality interviews and content so you can find information on any topic in archaeology from around the world. One way we do that is by recording interviews with our hosts and guests located in many parts of the world all at once. We do that through the use of Zencaster. That's Z E N C A S T R.
00:18:50
Speaker
Zencaster allows us to record high-quality audio with no stress on the guest. Just send them a link to click on and that's it. Zencaster does the rest. They even do automatic transcriptions. Check out the link in the show notes for 30% off your first three months or go to zencastr.com and use the code ARCHEOTECH. That's A-R-C-H-A-E-O-T-E-C-H.
00:19:12
Speaker
Looking to expand your knowledge of x-rays and imaging in the archaeology field? Then check out An Introduction to Paleoradiography, a short online course offering professional training for archaeologists and affiliated disciplines. Created by archaeologist, radiographer, and lecturer James Elliott, the content of this course is based upon his research and teaching experience in higher education. It is approved by the Chartered Institute for Archaeologists as four hours of training. That's in the UK for those of you that don't know.
00:19:37
Speaker
So, don't miss out on this exciting opportunity for professional and personal development. For more information on pricing and core structure, visit paleoimaging.com. That's P-A-L-E-O, imaging.com. And look for the link in the show notes to this episode.
00:19:54
Speaker
Welcome back to the Archeotech podcast. I'm Chris Webster, your host today. And we are talking to Hector about drone survey, which is a topic that is near and dear to my heart. So in some of the questions that you posed when we set up this interview, you mentioned artificial intelligence. I'm assuming you're using artificial intelligence to, to, I mean, obviously it is a method of artificial intelligence just to produce the model, but
00:20:20
Speaker
And you've mentioned an algorithm using to find the potsherds and things like that. I think is that what you are using artificial intelligence for anything else or do I have that wrong? Is that what you're using to find that?
00:20:31
Speaker
Yes, we are using machine learning algorithms, which is a subfield of artificial intelligence. And the first algorithm, at least the algorithm that is published, is a kind of random forest. It's a type of machine learning algorithm. The point of this is basically that you don't need to program.
00:20:58
Speaker
a code to locate bot cells, that is, you don't need to program the whole process that is saying, for example, okay, detect everything that has this combination of pixels with this shape. This would be impossible because it's very variable.
00:21:15
Speaker
But with machine learning, what we can do is just show samples, have some kind of training data, which in our case would be images of pod cells that we detect and we tell the algorithms that's what we are interested in, so the algorithms try to find.
00:21:32
Speaker
Things like that, okay? And then you can evaluate these results and say, okay, you were wrong here, you were right there. So the algorithm thing proves. That's what we call iterations. So, I mean, in this case, with the second iteration, we had a very good algorithm.
00:21:47
Speaker
But yeah, we did a third one just to try to fine-tune it. So yes, this is the kind of process. Now we are moving forward in another subfield of machine learning, which is called depth learning. We are using convolutional neural networks to improve the results of the detection.
00:22:09
Speaker
Okay, this is still not published, it's what we have been doing during the last month, and it really has improved quite a lot the detection rate. Also, it has improved quite a lot the way in which it detects pottery sets. I mean, the shape is much more accurate.
00:22:32
Speaker
And also, with this new method, we can even extract measures on how sharp are the edges of the pottery cells. That could be very useful, for example, to measure erosion.
00:22:43
Speaker
I think the best part of this method is that finally, with this, we can move forward from all the data we have from surveys were based on densities. I mean, densities that could be based on lines or squares or areas or fields.
00:23:06
Speaker
plot. Now we move to the actual atomic component of survey, which is the actual artifact, or the bot-serve, or the lithic fragment. And this opens a whole new field for archaeological survey. I mean, it's kind of silly because we always have this. It's not a consequence of the method itself. It's how we can now record the bot-serve or whatever the artifacts we are interested in.
00:23:37
Speaker
We have to rethink the way in which we analyze our survey data, for example. Now we can have measures of how big every individual posture is, how it is oriented, if it's eroded, I mean, if the shape is more roundy or it keeps sharp edges.
00:23:57
Speaker
And you can combine this data to extract measures of erosion, redeposition, what is the center, even which areas of sites are in danger because they are closer to the surface, because obviously you will have bigger potholes with sharper edges.
00:24:20
Speaker
So this is something that we were not very conscious when we started doing this, but looking at the amount of that as the tens of thousands of boxes and how they distribute. I mean, this is not just one dot, it's a shape. And this is giving incredible amount of information. And with this method, with dev learning,
00:24:47
Speaker
is that we have boosted quite a lot the detection capability of the algorithms, but also how well it extracts the shape of every individual posture. Yeah, it's amazing to me that you can
00:25:01
Speaker
use machine learning to teach your algorithm about pot shirts because I mean, I've seen pot shirts all over this country, especially on the East coast of the United States. And I'll tell you what, they come in all shapes and sizes. You know what I mean? I mean, they definitely, they definitely look different than say the surrounding rocks because they're usually more jagged. But then if you have say a whole bottom of a vessel or something like that, like a base of a vessel that can be just a big rounded shape. I mean, how,
00:25:29
Speaker
how good is it at finding these things? Or does it even need to be that good? Because if it just finds some pot shirts and then there's others that can't identify, at least you know where to go, right? So then you can go out there and check that out.
00:25:41
Speaker
Yes, yes, yes. It doesn't need to be that good. I mean, one of the innovations we actually introduced in this was, well, now this is a bit outdated with dead learning, but for the first algorithm thing, which is just a machine learning, a random forest algorithm, which is more simple.
00:26:00
Speaker
One of the things we realized was making the pots stand out is the texture. I mean, it's not the physical texture. It's not the topography of the surface. It's the different in pots. I mean, when you go down to this resolution of about one millimeter per pixel, every pot is going to have at least 20 pixels.
00:26:24
Speaker
So the difference in color between pixels is very little for a pot. When you move to the ground, which is close to it, there's lots of rugosity. The pixels have different values.
00:26:41
Speaker
And this is a very good way to detect boxes, apart from the color, of course. Now with depth learning, of course, we have depth learning is taking into account also the variability between objects and pixels, but also the edges.
00:26:58
Speaker
which is why it is so good at drawing, at redrawing the contour of the port cell. So it is not just the color, it's the texture. It's this edge that marks a topographic difference with the surrounding area.
AI Limitations and Future Aspirations
00:27:15
Speaker
It's that these edges tend to be much more regular than anything you would find naturally, a stone.
00:27:21
Speaker
even pots that are relatively round, the edge is going to be much straighter than any stone. So we have problems of course because it would not differentiate between ancient and modern building materials, I mean in many cases.
00:27:36
Speaker
You know, but this is relatively easy because by the way, when you have a concentration, you just check out what it is and you realize very soon that it's not as important. But yeah, it's curious, it's not just the color, it's both certain, most types of material culture we are interested in have a series of physical characteristics
00:28:02
Speaker
that differentiate them quite a lot from their surrounding environment. And this is what we are using to extract them.
00:28:08
Speaker
Okay. Now in that area, it sounds like pot sherds are the dominant type of artifact to identify a feature or a site or something where you want to see. But what are some of the other limitations of this? I imagine one of the limitations would be you've created this to identify pot sherds and it could be identifying other stuff, but it's also not identifying
00:28:33
Speaker
other types of artifacts. Could you teach this same algorithm to search for other artifacts as well? Or would you just run different algorithms over the same data set to find different things?
00:28:45
Speaker
Yeah, that's the process you follow. Because in this case, the algorithm we use is not a classification. That is, it's not telling you, this is this, this is pottery, this is flint or metal, and this is soil. Because there would be too many classes, too many things to classify.
00:29:11
Speaker
Okay, so what we use is a probabilistic approach, that is pottery or not pottery. So in this way, it can work so much better and you also have a percentage of success that you can use to filter out things that are not potters.
00:29:26
Speaker
And so this kind of worked well. So we have also been trying to do this for Linux in Thessaly also, with some of the images we took earlier on. And it works pretty well. But of course, every element is a different algorithm. It doesn't make much of a difference, because the only thing you do is you just copy, paste the same algorithm, and you just change the training data.
00:29:54
Speaker
Oh, yeah. OK, that makes total sense. And yeah, and it works well. You just then you combine the results. I mean, of course, they have different success percentage, but that's something you can easily calculate and compensate for it. But there will be things you will never find. OK, you know, I mean, that's one of the problems of artificial intelligence. I mean, it's I think this is the core of your question. I mean, for example,
00:30:23
Speaker
with this system you will lose anything that you don't expect to find and this is this is a big problem at the bottom I mean for example we have sometimes Neolithic figurines and those are some of them are impressive or beautiful and you know but they have very different shapes they are not always the same figurine and they are broken so
00:30:45
Speaker
You can only detect them by the color, but they are made from local soil. So if you don't create a specific algorithm to locate these neolithic figurines, you will not find them. Or if you have a piece of imported material, lithic that is imported from somewhere that differs in the shape and color than what you are expecting to find there, and you have not trained the algorithm to recognize this, you will lose it.
00:31:13
Speaker
Imagine a piece of written record, clay tablet, it's a risk. I don't think you should just do an automatic survey and forget about visiting the sites. It's just a way to be more conscious and to be more efficient, but you need to have an academic expert, you need to go to the field, you will just save lots of time. Hopefully, if the conditions are good.
00:31:41
Speaker
Well, let me ask you this because I've thought about that as well about the things that
00:31:46
Speaker
you're not programming it to find you're losing those things. And that's what the common complaint is I've had when I've talked to people about doing drone survey, they're like, nothing beats somebody's eyes on the ground and being able to interpret something new, right? Something that something that's either new or, or we just didn't expect to see out there. So then we couldn't program to find that being said, it sounds like you were saying earlier in the first segment that the model is very accurate. I mean, down to, down to a very tight resolution. So,
00:32:14
Speaker
Let's take a Neolithic figurine as an example. If you were looking at the model with your own eyes, not using an algorithm, you're just looking at the image, would you see that Neolithic figurine if you were assuming this? Okay, so using that, this is one thing I've thought about. You've seen
00:32:29
Speaker
We've seen different crowdsourcing methods out there like Galaxy Zoo that helps to identify galaxies. SETI at home was an early one that just used your computer. And then the National Geographic one, the Global Explorer one that does the find signs of looting around the world. What about taking these data? So you guys as archaeologists, you run the algorithms, you go find the things that you're expecting to find, and then you go ground truth those. But then you take the images, strip the locational data so the average person can't know where it's at.
00:32:58
Speaker
And then basically just have people scrub through these images looking for things that don't look natural and then identifying those hits. I've thought about crowdsourcing stuff like that. Is that something you guys have considered? Yeah, yeah. We actually developed a system like that for destruction of features from maps at some point.
00:33:17
Speaker
But yeah, well, there were some copyright issues, so we didn't go ahead with it. But yeah, we thought of it. We thought of it, the problem is having much longer time flights. You would end up with a really huge image, okay? That would take quite a lot to analyze.
00:33:38
Speaker
I mean, yes, you can do that, but you can have like 20 people or 30 people going through a relatively small field for many, many hours. The scale of this is like checking a country.
00:33:56
Speaker
No, because every pixel, it's a millimeter, where most people are used to satellite images, like I don't know, gold resolution, like Sentinel-2, 2 meters, 10 meters, you know.
00:34:12
Speaker
Okay, this is a millimeter, so you actually, yes, that's something you could do if you set up a very good system and you have lots of volunteers that are going to be taking really boring stuff. Many of these crowd-solving methods are really exciting because people are finding things they can identify. I'm not sure if this will, but of course, it's an excellent approach. We were thinking about it at some point,
00:34:42
Speaker
But we had our doubts of how many volunteers would be able to sacrifice their life to the altar of digital archaeology. Hey, I'll tell you what, you put up a website and you put up images and tell people this, there's people with amazing amount of free time on their hands.
00:34:59
Speaker
That will go through. Just image by image. That will definitely work out dry. Nice, nice. So thinking about the archaeological field as a whole, how easy would this be to teach other people who may not be as tech savvy as you and I to be able to understand and do this?
00:35:18
Speaker
Yeah, that's something we wanted to solve at some point. In fact, we asked for a project now for some money to get a real tech expert to actually develop kind of web page in which people do not need to see the code, but they can draw their own
00:35:41
Speaker
the visible photo search or Linux or metal or coins or whatever in the round or the photos and the system will run in the background. You have this front end which is pretty easy. It's like Google Earth or something like that. You can just draw the things you are interested and in the background, the algorithm will simply multiply this data and so images of things the algorithm has been able to find say,
00:36:10
Speaker
Is this what you want? Yes or no? So develop a more intuitive way to do that. I don't know if we will get the funding to actually do it. And the big problem is to find a supercomputing service that would actually allow
00:36:30
Speaker
allowed to create this web front page for people to do it for free. Because the amount of computing necessary for this kind of methodology is huge.
00:36:47
Speaker
So, yeah, I mean, that's one of the things now we were trying to develop our drone that would fly for one hour or more. I mean, a kind of a hybrid drone. But yeah, one of the problems will be like, you know, if you have a drone flying for one hour with a high resolution camera, the amount of gigabytes you are going to end up with, I mean, taking continuous pictures,
00:37:15
Speaker
It's going to be very difficult to process, so we are starting to move to cloud processing, not just the machine learning part of the process, but also the photogrammetry part of the process. This is the bottleneck right now.
00:37:32
Speaker
Yeah. All right. Well, aside from the funding you're looking for for those things, what future developments are we looking for? Money is no object. What kind of developments would you like to see in this field and what kind of things would you like to research given those reduced constraints?
00:37:47
Speaker
Well, I mean, there's amazing hardware and tools out there. I mean, there are magnetometers that you can mount on a drone. It's just incredible. I think thermography has a good in the future play and
00:38:10
Speaker
a really big role and also multispectral imaging. But of course, then you will need to fly the drone much lower because the higher you go in the electromagnetic spectrum, the less resolution usually. So yeah, you have very low resolution compared to the visible light for thermal, a bit better for shortwave infrared. But it's still, you know, you will need to fly the drone really low to get
00:38:40
Speaker
okay perhaps you could get instead of one millimeter per pixel you could get one centimeter and you know and try to correlate the data but that's something i would like to try definitely the shortwave infrared and the thermal because of course i mean with thermal is
00:38:59
Speaker
They have a very different thermal inertia than the stones or the ground. While the ground is relatively cold, you can have a stone really hot and the pot can be in the middle. So depending on the hour you take the picture, you will have different signals, but this would be...
00:39:17
Speaker
this will be really good also we are we are now we are trying to develop in the man is still an issue but if it was not we are we would like to create this drone an hybrid drone that is using some kind of petrol or gas to fly for longer periods of time.
00:39:35
Speaker
with a smaller radar or larger altimeter to keep the constant height to the ground with a differential RTK GPS system. So we know the exact position at every moment with a really good camera. I mean, now there are some cameras that can be mounted on a drone with 40 megapixels or something like that, which means you can fly it much higher.
00:40:05
Speaker
You know and take more more area with a single picture or even increase the resolution because okay with you when you increase the resolution you can you can develop digital terrain models of the ground which would imagine the kind of to have a 3d of the topography of the ground would be a
00:40:27
Speaker
I mean, many of these methods, I'm sure they can be applied to other fields. And you feel that that is interesting on whatever lies in the ground.
00:40:37
Speaker
So yeah, I think that would be really interesting. I also would like to try the drone mount, the magnetometer, because I would like, I think with this, we can obtain very complimentary kind of information. Sure. Okay. Also look a bit in the, you know, what's below what is visible. So yeah.
00:41:01
Speaker
I mean, we are living a very interesting moment. Technology is moving so fast. I mean, drones. I mean, the first one I used, I think it was back in 2014 or so. You know, in just a few years, in four years, I mean, the quality is just incredible. It has changed so much. Now you can do things that were completely impossible four years, five years ago.
00:41:28
Speaker
Yeah, I actually picked up the new DJI Mavic Mini a couple months ago when it first came out just to have something a little more portable. Yeah, for certain experiences. And man, this thing is tiny, but it runs for a solid 30 minutes, can handle, you know, 20 to 20 to 25 mile per hour wind speeds and still maintain stability on that camera. And sure, the camera's got some limitations versus a higher end camera, but you know, right tool for the right job.
00:41:51
Speaker
And this thing basically fits in your pocket and I've got three batteries with it. And so I can run an hour and a half of video and it's just, it's phenomenal. I can't, I can imagine. I also have a, yeah, I have an Inspire, a DJI Inspire Pro 1 as well.
00:42:09
Speaker
I mean, that thing was drone. Yeah. I mean, that thing was $5,000 and this little Mavic mini four years later is taking just as good of image as that, as the drone from five years ago. Right. Yeah. No, I mean, one of the big limitations though with drone is the legislation is constantly changing now. I mean,
00:42:33
Speaker
In many of the countries in which we work, you need to actually write a project in order to solve it. Looking at the future, that's going to be one of the big limitations of this methodology. It's not the kind of flight that anybody should be worried about, because at the moment, the thing is flying three meters high. It's in areas that, by definition, you don't have people around.
00:43:02
Speaker
you know, very few people. Yeah, yeah, yeah, that is going to be a problem. I mean, with with the mabbits that they are supposed to be designed to be more usable in more conditions, because they are extremely light. And at least in Spain, the the they still need to go through all kinds of permits, somebody with a with a special permit and some special training needs to take the flight. So
00:43:33
Speaker
It's pretty complex. Yeah. Yeah. All right, Hector. Well, that's all the time we have. Thank you so much for coming on and I really look forward to, uh, future things. If you guys get that funding and you do some other research and you put some other stuff out there, please come back on the show and, uh, and tell us all about it.
00:43:49
Speaker
Well, actually we are using the same method, but using satellite imagery to find sites. So yeah, perhaps this will be another interview. We'll bring you right back on and talk about that because it's where everything is going. And I think doing this sort of, for lack of a better word, remote archaeology is going to be more cost affordable because archaeology projects, whether you're doing academic archaeology or
00:44:20
Speaker
contract, you know, um, uh, in this country called just professional archeology.
The Future of Drones in Archaeology
00:44:24
Speaker
There's no, there's never budget for it. There's never money. So the way, if we can reduce, if we can increase what we can see by reducing the number of people, because people are the most expensive aspect of it and yet still get the same, if not better.
00:44:38
Speaker
Yeah. I have to say that, you know, when we tried to publish this paper, there were lots of complaints from reviewers and colleagues that read the paper. You know, I mean, you know, because, OK, it's expensive to have a large team, but usually in academic archaeology, you have students and some of them ironically complained about the poorest students. They would not get the training. But for God's sake, I mean, which kind of training it's to learn how to pick up
00:45:05
Speaker
pieces of poetry. This is not designed to take the job out of anybody. I think it's just a way to use much better the time in a more efficient way. You're going to need lots of people still. And the training is much more transferable. It's not just learn how to find the pieces of poetry in the ground that you only need a few hours or a couple of weeks to do it really, really well.
00:45:33
Speaker
So yeah, yeah, yeah, yeah. It's people. It's the big issue. Yeah. Yeah. I completely agree. We're taking people from the menial task that you can basically train almost anyone to do of finding things and then allowing them to focus on the science of the analysis and the research and just.
00:45:50
Speaker
let the drones do the busy work and then we come back and do the science, right? I mean, I think that's, you're totally right. Um, we don't need to focus so much on that, but you know, this country, most of us that are, that do CRM archeology, uh, or call it contrite archeology. I mean, that's what most of our time is spent doing. We don't do any science, right? We spend most of our time just walking in the desert or digging shovel tests and finding things. And then other people, a very small section of people actually get to do the analysis and stuff like that. And I'd like to open that up to more people.
00:46:20
Speaker
Yeah. I know. I know. All right. Well, we've got to go, but thank you for coming on, like I said, and maybe we'll have you back on again soon to talk about the satellite stuff or some other work that you're doing. So thanks a lot, Hector. No, no. Thank you very much, Chris. It was a really a pleasure to be here tonight.
00:46:37
Speaker
You may have heard my pitch from membership. It's a great idea and really helps out. However, you can also support us by picking up a fun t-shirt, sticker, or something from a large selection of items from our tea public store. Head over to arkpodnet.com slash shop for a link. That's arkpodnet.com slash shop to pick up some fun swag and support the show. Hey everybody, Chris Webster here back for the app of the day segment for Archeo Tech 121. I want to tell you about an app just in case it gets crazy popular and they start charging more for it.
00:47:07
Speaker
I actually haven't been able to log into it on my phone, but I've looked at it online. I think they were having a problem because they were just announced on the news, but it's called Do Not Pay. And if you look up Do Not Pay All One Word on the App Store, you'll find it. It's on the Apple App Store and the Google Play Store.
00:47:22
Speaker
They call it the world's first robot lawyer, and basically on their homepage here at DoNotPay.com, it says you can fight corporations, beat bureaucracy, find hidden money, sue anyone, and automatically cancel free trials. What some of that means is they started because the guy who developed it lives in San Francisco, and parking tickets are
00:47:41
Speaker
a huge thing there, but basically you can just punch in your parking ticket details and it automatically submits to the right agency, basically fighting the ticket. And they found that in 50% or more of cases, if you fight the ticket, you'll get it either reduced or removed.
00:47:56
Speaker
And so this just does that automatically for you. Some of the things you can do is they said it'll even help you in like small claims court. It'll produce a, you put in all the details and it'll produce a script and it'll say, say this, if they say this, say this, if they say this, and it's pretty good. It's been tested many, many times.
00:48:14
Speaker
One of the other things you can do is they talk about free trials and they said what companies count on with free trials is that you're going to forget you did the free trial. So you sign up, you always have to sign up with a credit card and then at the end of 30 days or whatever the trial period is, they just start charging your card, right? Well, these guys
00:48:31
Speaker
produce basically a credit card number that's not actually a credit card number. It's legit. And when the service tries to ping it, it comes back as a legal credit card, but it's not actually a credit card and it goes nowhere. It's tied to no bank, no money, no anything. And if you just use it for the free trial at the end of the free trial, it'll cancel that credit card and we'll cancel the free trial.
00:48:55
Speaker
Sounds pretty cool. I actually don't know how they make any money. I think they probably do different things like taking percentages and things like that, but you can cancel subscriptions with this. Again, there's all kinds of different things.
00:49:10
Speaker
I think we're just scratching the surface on this one, but I wanted to tell you guys about it as a quick little app of the day for all my frugal listeners out there to the archaeology podcast network, because I know if you're an archaeologist, you're probably having, you're always having troubles with money because we simply don't get paid enough. So this could be one way to help you save a little money, manage your subscriptions and not get overcharged for things. So check that out. Do not pay.com.
00:49:35
Speaker
Well, that's all we have for this time around. I'm hoping Paul can join in again shortly. He's had some, uh, he's had some things going on at home, so he hasn't been able to join us. He hasn't gone anywhere. Uh, he will be back at some point in the future, but until then we've got a lot more interviews scheduled between, uh, we're recording this at the end of January between now and April and a lot of exciting things coming up. So stay tuned for that. And then please.
00:49:59
Speaker
share this episode wherever you heard it. If you're listening to this in a podcast player, typically you can just find the share icon right there, send it out to your socials. That helps people see it, gets more information out there to the world and lets people know about the stuff that we're talking about. So thanks again for listening and we'll see you next time.
00:50:21
Speaker
Thanks for listening to the Archaeotech Podcast. Links to items mentioned on the show are in the show notes at www.archpodnet.com slash archaeotech. Contact us at chris at archaeologypodcastnetwork.com and paul at lugall.com. Support the show by becoming a member at archpodnet.com slash members. The music is a song called Off Road and is licensed free from Apple. Thanks for listening.
00:50:47
Speaker
This show is produced and recorded by the Archaeology Podcast Network, Chris Webster and Tristan Boyle in Reno, Nevada at the Reno Collective. This has been a presentation of the Archaeology Podcast Network. Visit us on the web for show notes and other podcasts at www.archpodnet.com. Contact us at chrisatarchaeologypodcastnetwork.com.
00:51:08
Speaker
Thanks again for listening to this episode and for supporting the Archaeology Podcast Network. If you want these shows to keep going, consider becoming a member for just $7.99 US dollars a month. That's cheaper than a venti quad eggnog latte. Go to archpodnet.com slash members for more info.