I'm a big fan of yours and am saddened to see you advertising for a firm that offers, at best, very dubious investments covered in shiny marketing. I wonder how much it took
@@bananaman2495 You don't want to buy art for someone else to enjoy? I really wish creators would do a bit more research on some of the shady advertisers on UA-cam. If one reads non-sponsored sources about this advertiser, it quickly becomes apparent the advertiser is very shady.
You are a very smart person and I’m sure it was very hard decision, but please for silicon sake just create a $50 or 100$ tiers, but stop advertising this crap.🙏
Nice ad transition! though I will have to concur with the others regarding the sponsor. Don't sell yourself out! Your videos age well and I hope to continue watching them in the years to come.
Jon, I want to echo the other people who commented on the same thing. Please, please make multiple expensive subscription tiers (ie, like PBS Space time does) and keep your channel ad free. I think you will be able to generate more revenue long term. I am donating to put my money on the line. You research very interesting topics and there is a non-negligent subset of viewers who will support you for a higher amount! Please!
I'm constantly amazed by how 'well rounded' the knowledge of the host of this channel is. Every subject he does is something I find fascinating. How someone can make a smart video on satellite imagery AND a subject as polar opposite as the nuances of Australia's gas woes is beyond my own capabilities. Most Australians don't even completely understand our gas market, so to have someone with a North American accent explain it to me is wild.
He didn’t mention that the USA is using this tech to map the war in Ukraine against Russia, and this and AI strategies are big reasons Russia is struggling
I am a an engineer working in this field and I want to make my compliments to the author. very accurate and clear explanation of the current state of the art technology in remote sensing and earth observation!
What they did for Microsoft Flight Sim in 2020 was quite cool. Where previous simulators would populate the world with a pool of generic building model, the new MSFS produces more accurate individual buildings using AI. They had a company scan all tiles of the Bing Maps with an AI whose job was to find all kinds of structures, trace their basic floorplan, judge height and colors. Which in terms allows the game to have in a ideal scenario, each individual building placed accurately within it's plot on top of the imagery. The height and odd architectural features obviously don't always translate well but it's an incredible step up. The color recognition further enhances accuracy by identifying the correct roof type and replacing clouds with a generic terrain texture. And also placing down trees and forest more accurately than the normal forest layer on OSM
Although they did pay a company to analyze Bing Maps data, they only did that for bigger and more popular cities such as Paris. Most locations pull data from OpenStreetMap where data about buildings is crowdsourced. Buildings that have typos in their crowdsourced data will often create bugged buildings in-game.
aren't most of the buildings just generated with normal autogen, like FSX or X-Plane? Also the high detail cities like New York, Toronto, Tokyo etc still use manual photogrammetry scanning
You’ve tapped into a universe of value. SAR, Hyperspectral, and gravitational sensors on low cost satellite swarms means a host of applications that will change risk, finance, insurance, agriculture, logistics and more. It’s a trillion dollar opportunity that few see, let alone know how to apply. AI makes it accessible and actionable because on satellite processing lowers the cost of tasked data sets. Glad you are showcasing a topic that excites me.
Hey, not sure if you care about my opinion at all, if you do: I really loved your channel so far. And doing some ads is fine too. But please make sure you clearly state where an ad/sponsoring segment starts next time (you made very clear that it was sponsored when you ended the segment, but i was so shocked that i almost closed the video before that). Your integrity is one of the highest goods you have and doing such smooth transitions from the topic of your video to the sponsor is one of the easiest way to lose the trust we have in you.
I believe what u meant to say was “thank u for bestowing upon me so much insight and knowledge, but seeing as how I’m such a sensitive and anal t wat, would u kindly consider delineating ur ads more clearly?”. I fixed it for u. Ur welcome
I have worked in this industry for almost 15 years, working on projects and contracts for many of the products discussed in this video. While geospatial machine learning/AI products are getting better by the day, many of them are not quite ready for prime time. While I don't disagree with the conclusion of the video, I would say that the biggest challenges ahead for these firms is identifying buyers for these alpha or beta state products. Either way, fantastic video! One of my favorite channels on youtube.
the problem is at the end of the day we don't actually have AI. the current machine learning is in reality multidimension optimization instead of single dimension optimization in the past.
@@wei270 Well said Wei, I totally agree with you. Multidimensional optimization is great, and has a lot of uses, but it's definitely not AI. I hope there will be a shift in the usage of the term "AI" in the future, but I'm doubtful.
The companies I've worked for that have succeeded with AI based products normally use machine learning as a filter for obvious cases and then pass on anything with ambiguity onto a team of humans that make some set of judgements. Generally this is a very useful business model since it means you can solve problems that weren't economically feasible previously before the filtering step.
I am amazed by the depth and breath of this channel, as a remote sensing person. This went beyond my expectation for a layman explanation of remote sensing, very clear very cogent just like your other videos. Kudos and very well done!
This isn't entirely related, but it's a fun story: In the late 50s/early 60s, the US determined that an increasing number of Soviet Union soldiers must be stationed in Cuba. They did this by noting that spy plane photographs showed a lot of public parks were being marked out as soccer fields. Cubans play baseball; Russians play soccer.
As a researcher on this field i can say that AI is the new "hot stuff" in image processing and analysis in remote sensing, but as the technology progress, so does the work/headaches. There are so many variables to be considered that, in my view, interpreting results with quality and distinguishing between good and garbage was, is and will be an "art". I think AI helps in the heavy lifting, but the final conclusions and corrections have to be given by a specialist. Nowadays there are so many options at the click of the button, that many mistakes are easily made, which of course i also made, and took time and study to correct. I don´t mean that it should not be used and improved, but we should avoid relying solely on that, just because it worked before Anyway, amazing video, keep up the good work.
@@LearnItalian24-mz7hg Of course, ask is the way to learning. I would recommend to build the knowledge to the ground up, first the basics of remote sensing, GIS and its techniques, how it works, pros and cons. A good chunk of people goes straight to AI/Neural networks/deep learning things and complicate things when a good old technique or some clever workarounds solve it just fine. Then in parallel get to know coding like python, IDE and google earth engine to name a few, those things will help automate big and bigger workflows, with time, getting to know AI tools will feel a natural extension. When you get in this stage, get some course on udemy/coursera or good book about this subject to guide you. Also, remote sensing is an art in some ways, accumulate experience, do analysis by hand to get the feel for adjustments and technique uses, the quality of your results depends on the quality of your know how. Best of luck to you.
@@LearnItalian24-mz7hg I would suggest you to learn well the basics, the concepts and techniques, there are already tons of techniques that solves most of the problems, in parallel, learn to code, python is a great start, then you can use what you learnt in a bigger scale, automate workflows and so on. With this know-how you will know better on when and how to apply AI to those analysis. Some things in remote sensing are like art and with experience you will know how to distinguish and characterize it. Best of luck to you.
One of the "big" tricks in identification is based on spectral resolution. You can pick out a given material based on its reflected or emitted spectrum. This applies to everything from chlorophyll to asphalt. With the right filters, you can pick anything you want out. The hard part is knowing what to look for. As a for instance, with stored full-spectrum image data, you could probably do a study of roadbuilding by looking for the emission lines from the long-chain hydrocarbons used to make asphalt.
Yeah, I didn't give this video a Thumbs Up because of that. I want to know when I'm hearing an ad. In this case, it took much too long to notice he was off-topic.
At minute 13.42 It's a fortified city of Palmanova, North East of Italy, founded in XV century . I live 20 km from that city and i have done my military service there in 1984. Thank you for the imageñ
YES. Thank you, this topic is bleeding under my nails since months. I recently joined my local Openstreetmap community and I really feel that disruption is around the corner. ML won't see street names, but streets curbs with 10cm accuracy. Places like India are still missing mapped roads in many places. Good satellite images can see the street conditions nicely, and daily. We map based on that, and ML will make it better than us. We have gigantic amounts of training data by applying the existing map and satellite data to train the model. I think it's why Microsoft and Meta bound together to create Overture maps, while putting OSM into stress but also opportunities. Look at Tesla FSD beta, it maps roads live with high accuracy on the fly while driving. Give this a 10x or 100x progress and we have disruption.
big data and satellite imagery can have a significant impact in various fields such as agriculture, urban planning, natural resources management, and disaster response. For example, in agriculture, satellite imagery can be used to monitor crop growth, identify areas that need irrigation, and predict crop yields. In urban planning, satellite imagery can be used to track urban growth and identify areas in need of infrastructure development. In natural resources management, satellite imagery can be used to monitor deforestation, illegal fishing, and other activities that may have a negative impact on the environment. Finally, in disaster response, satellite imagery can be used to identify areas affected by natural disasters, assess the extent of damage, and aid in the distribution of relief supplies. Overall, big data and satellite imagery provide valuable insights that can help organizations make more informed decisions and improve the efficiency of their operations.
All of the Asianometry UA-cam vids are excellent. Absolutely the best, but the internal advertising degrades the quality and cred. Keep up the great work.
I can't Believe you segued into a Masterworks advert without signalling it in any way. Shame on you for not clearly distinguisihing between your content and advertising.
Been watching your channel since some of your first episodes. I have recommended you to many people as your videos are some of the most informative I've ever watched. Keep up the good work!
14:00 For the farmland detection we have already using infrared satellite images. They show clearly man-made objects, different crops, forests and water.
Everyone here should be knowledgeable enough to know what an ad is. I have never trusted an ad on face value... Do you write complaints to anyone running a detergent ad? Or a tire ad?
This Masterwork add is highly questionable, 1,7 trillion sounds much but is just the GDP of the Netherlands and Belgium also an insanely small market which seems inherently risky. While also it seems to be at an macro economic scale to be just a bad time to invest into alternative investments as there already is getting a lot investment into them so it kinda seems like telling people to buy crypto when it is high. From how i see it this are extreme high risk new finical products and seems rather just something for sophisticated investor that understand they are taking a massive risk but it seems rather unethical to push this onto average people as an save investment. From what i think i know i don´t think any expert will tell you to buy more then 10-20% as the high risk portion of the portfolio which could burn totally.
one czech company has a cubesat that has machine learning algorithm bthat decides which images are good and interesting and only sends them back to earth, they cannot send everything cause of the limited bandwith.
Some of the coolest satellite images I've seen this year were the Tonga volcano and the shockwaves transiting the globe. No mention of how lidar seems dissolve vegetation? While not satellite images a lot of pre-Columbian ruins were identified in Central America recently. I can see how this technology will revolutionize battlefield surveillance and things like nuclear compliance.
the utilization of large-scale data sets and remote sensing technology through satellite imagery can have a significant impact on various industries and disciplines. The ability to process and analyze vast amounts of data, in combination with the high-resolution images captured by satellites, facilitates the optimization of decision-making processes and operational efficiencies in fields such as agriculture, urban planning, natural resources management, and emergency response. For instance, in the realm of agriculture, the implementation of satellite imagery enables the monitoring of crop growth, identification of regions requiring irrigation, and forecasting of crop yields. Furthermore, in urban planning, the utilization of satellite imagery permits the assessment of urban expansion and identification of areas in need of infrastructure development. Additionally, in natural resources management, the utilization of satellite imagery facilitates the detection of deforestation, illegal fishing, and other activities that may have a detrimental impact on the environment. Lastly, in emergency response, satellite imagery can be employed to determine the areas affected by natural disasters, evaluate the extent of damage, and aid in the distribution of aid. In summary, big data and satellite imagery provide valuable information that can enhance the intelligence of organizational decision-making and enhance operational efficiencies.
The thing with "Super Resolution" and all other linear algebraic techniques is although they will on average match reality that is only because reality is regular. A.I. produces what is expected to be their by making a guess that is on average right. But what it does that by taking the "average", it makes a prediction that everything will look the same as the training set. Training sets are limited and biased is known and probably unknown ways. When reality departs from the the training set A.I. still assumes the training set is truth. A.i. will predict the training set. When something outside the correlation between training set members occurs A.I. flounders. A.I. doesn't derive principles, it predicts the usual. Because what is usual, usually happens A.I. is accurate. When the unusual happens A.I. usually fails, but those failures are unusual and who cares about unusual events?
I'm sorry but that sponsor was shoehorned in a really uncomfortable way in your video, i had to dislike the video despite finding the rest of the content really insightfull.
I personally like the idea of using this tech for land classification, that is, what type of vegetation or farmland a given area has: is it mostly eucalyptus or pine trees; corn or wheat; desert or grassland? I remember using a mobile app showing land identification/classification a few years ago (and whose name I've forgotten), but it still had some blatant errors in my area...
For the purposes of image recognition, 'super resolution' is 'object detection', they're equivalent. To understand this, think of what happens if your ML algo hallucinates pixels so that pixels that are of a garage are upscaled to look like a truck.
"Can you guess what this?" Given the size and shape of the two areas that seem to be theme parks, and the large parking lots to the direct west and south east, as well as the large seemingly convention centre south, Disneyland
This is a massive industry with massive implications. Especially for militaries. China in particular is investing big time in this field to acquire the ability to track US mil assets as much as they can and as close to real time as they can. This synergizes with their massive ballistic missile fleet to give them the ability to strike US assets thousand + miles away. While the US has developed and will develop a constellation of satellites to track (in real time) missiles such as ballistic and hypersonic missiles. This track will cue other missiles to shoot down these threats. Its a back and forth race between the two
Super-resolution is false resolution. It is essentially AI cartooning, not enhancement. The cartoonist draws a higher-resolution "artist's conception" of a photograph. The result is an artificial hypothesis, an illusory parlor trick, not a photograph. That's what smartphones produce today, where the 1 or 2 MP diffraction-limited micro lenses somehow result in the 10 or 20 MP image files. We are then round-tripped to believe the lie that these cartoons are photographs because they are plausible, better resolved, and prettier than the truth. It's a lie we want to believe. AI flattery. A photograph is a scalar intensity map of a light field's surface. Images produced by human or artificial intelligence are not photographs, they are drawings of imaginary models.
You should have included that pixelated portrait of obama that the upscaling technique turned into a generic white man as a demonstration for the limitations of the single frame super resolution algorithms.
Great video as usual! Regarding Super-Resolution (around 7:30), perhaps GAN (Generative Adversarial Networks) is another one that worth mentioning. GFPGAN is an example of this, and since it is trained on human faces, it is capable of doing an exceptional job with the super-resolution of facial images. Same thing can be done with GAN for other types of objects or scenes. Regarding Satellite Image Analysis (around 10:44), beyond the four types mentioned there is another: Scene Understanding. This is about making a general description of a scene, such as "there was a tornado that passed this area" or "this is a farm land that has suffered severe bombardment with many wrecked vehicles," etc. For example, the Salesforce BLIP is a system capable of doing this type of analysis. This is quite different from using deep learning to classify images into fixed number of categories, since it is capable of recognizing a very large number of objects and scenes, and then provide a textual description. Other than using this for satellite image analysis, this also has important applications for many areas, such as many types of autonomous vehicles/robots that operates in the open world.
How is "scene understanding" done? Do you have to train your deep learning model on a repository of images that have been appropriately text-captioned?
@@manofsan Yes. BLIP (or OpenAI's CLIP) is pre-trained on a large dataset of image-caption pairs in order to understand the relationship between text and images.
@@kaihuchen5468 - what a coincidence, I was trying CLIP the other day, and I've just now tried it again because you've mentioned it. I found that it generates some useful text with extra unrelated garbage words added on. Perhaps that's because an input image can generate broader text associations than what's useful. Interesting tool, though.
Yet another prospect over the horizon is to train ChatGPT on the text derived from applying Scene Understanding to satellite images. Here I will let ChatGPT do all the explanation by itself: My question to ChatGPT: If we apply scene understanding to satellite images to create textual descriptions, then train ChatGPT on these text, what kind of advancement can be achieve in the management of satellite images? Reply from ChatGPT: Scene understanding, which is the ability of a computer to interpret and understand the contents of an image, can be applied to satellite images to generate textual descriptions of the image content. Once the scene understanding model is trained, it can be used to generate textual descriptions of new satellite images, which can then be used to train ChatGPT. By doing this, ChatGPT would learn to generate text that describes the content of satellite images in a coherent and contextually appropriate way. This process of scene understanding and text generation can bring several advancements in the management of satellite images: 1. Automated image annotation: By generating textual descriptions of satellite images, the images can be automatically annotated and organized, making it easier to find and retrieve specific images based on their content. 2. Image search: The textual descriptions generated by the scene understanding model can be used to perform text-based image search, allowing users to find satellite images based on their content. 3. Image captioning: ChatGPT trained on satellite image descriptions can be used to generate captions for satellite images, which can be used to improve accessibility and understanding of the images for people with visual impairments. 4. Image analysis: ChatGPT can be used to analyze satellite images, by identifying patterns and features, it can be used for different applications such as land-use monitoring, natural resource management, and disaster management. Please note that all of the above are based on the assumption that the scene understanding model is well trained and accurate. Also, the quality of the textual descriptions generated by the scene understanding model will affect the quality of the text generated by ChatGPT.
@@manofsan The internet-scope data that CLIP trained on is quite noisy. That's why Salesforce's BLIP put in some effort to deal with noisy data in order to produce better textual description.
4:10 Laser comms will further increase the amount if you include high quality video footage that currently cant be transferred because of bandwith limitations.
I really like your stuff, and am 100% behind getting sponsored. I am just saying, I am absolutely certain that Masterworks is going to be revealed as a scam within 18 months.
i personally experienced this ad read as somewhat inappropriate. the sentences leading up to the sponsor reveal did not make it clear that they were actually part of the ad read and as a viewer i was kept in the belief that the arguments given were part of establishing a base to support the hypothesis of the video, which i experienced as quite misleading. ad reads are fine, but please make it clear to the viewer that they are watching one :)
Hi John, I love the videos you produce. But to younger audiences out there, remember that masterworks is a sleight of hand, it is like one of the overpriced items you are persuaded to buy but serves no utility or purpose. Please be informed, billionaires don't invest in art, they just park their excess/ black income to avoid taxes and tax authorities' / government's oversight. That being said, never invest your/your parents' hard earned income unless you understand the asset and its market from a purely finance perspective. To me masterworks is just another FTX in making.
I am a little surprised there was no mention of satellite optics, but I guess that's not at all related to semiconductors, and it's, surprisingly, often done by hand. Huygens Optics channel has a neat series on it.
Good video! I love the AI generated images too! What was the words used to make the image at 0:04 ? The planet earth with a black starlit background with a satellite in orbit?
I thought he’d mention searching airplane crash locations and missing ships and boats. That would do wonders for finding people fast. The cameras could be satellite, boat or airplane based.
Very Intresting what development has happened to this term with the help of ai. It makes me remember back to the year 92 when a CD Rom set of satellite images from Germany was sold to watch it on a Windows 3 PC.
ChatGPT exists, and therefore has more causal powers irl, but within the works of fiction that those other AIs exist, they are more powerful that ChatGPT is irl. This seems obvious?
8:20 That's Disneyland. The large structure in the upper left is Starwars: Rise of the Resistance and that's the 5 freeway by the upper right corner. What did I win?
Does anyone know the best keywords to use when searching for these AI models for working with satellite imagery? For example if you want to count the number of cars in a satellite image, you need a model that is specifically trained for object detection from aerial images but when I search for "satellite image object detection" I feel I'm only finding a small number of results, whereas I expect there is plenty of research/discussion that's going on that I'm just not finding.
i got the short end of the genetic stick and have multiple chronic genetic diseases. I am hoping AI processing can help to extrapolate the numerous simulations of medical treatments. (sorry about the way the sentence is written, i can't think of how to write this properly) (also apologies if it's off topic... i see the "unlimited possibilities" and just wanted to say something.)
Skip the waitlist and invest in blue-chip art for the very first time by signing up for Masterworks 🖼️: masterworks.art/asianometry
I'm a big fan of yours and am saddened to see you advertising for a firm that offers, at best, very dubious investments covered in shiny marketing. I wonder how much it took
@@bananaman2495 You don't want to buy art for someone else to enjoy?
I really wish creators would do a bit more research on some of the shady advertisers on UA-cam. If one reads non-sponsored sources about this advertiser, it quickly becomes apparent the advertiser is very shady.
You are a very smart person and I’m sure it was very hard decision, but please for silicon sake just create a $50 or 100$ tiers, but stop advertising this crap.🙏
I hope you are getting a good price for your dignity.
Nice ad transition! though I will have to concur with the others regarding the sponsor. Don't sell yourself out! Your videos age well and I hope to continue watching them in the years to come.
Jon, I want to echo the other people who commented on the same thing. Please, please make multiple expensive subscription tiers (ie, like PBS Space time does) and keep your channel ad free. I think you will be able to generate more revenue long term. I am donating to put my money on the line. You research very interesting topics and there is a non-negligent subset of viewers who will support you for a higher amount! Please!
What's that going to get him
I'm constantly amazed by how 'well rounded' the knowledge of the host of this channel is. Every subject he does is something I find fascinating. How someone can make a smart video on satellite imagery AND a subject as polar opposite as the nuances of Australia's gas woes is beyond my own capabilities. Most Australians don't even completely understand our gas market, so to have someone with a North American accent explain it to me is wild.
He didn’t mention that the USA is using this tech to map the war in Ukraine against Russia, and this and AI strategies are big reasons Russia is struggling
@@StoutProper he did, right at the end.
Generality is the mask of superficiality.
@@andreasproteus1465 This comment sounded smarter in your head than it actually is.
Too true! It’s unfortunate that he has such a sketchy company sponsor the video though.
I am a
an engineer working in this field and I want to make my compliments to the author. very accurate and clear explanation of the current state of the art technology in remote sensing and earth observation!
I don’t know how he can be so prolific (and accurate) on so many disparate topics, one of my favorite UA-cam channels
What they did for Microsoft Flight Sim in 2020 was quite cool. Where previous simulators would populate the world with a pool of generic building model, the new MSFS produces more accurate individual buildings using AI. They had a company scan all tiles of the Bing Maps with an AI whose job was to find all kinds of structures, trace their basic floorplan, judge height and colors. Which in terms allows the game to have in a ideal scenario, each individual building placed accurately within it's plot on top of the imagery. The height and odd architectural features obviously don't always translate well but it's an incredible step up. The color recognition further enhances accuracy by identifying the correct roof type and replacing clouds with a generic terrain texture. And also placing down trees and forest more accurately than the normal forest layer on OSM
Although they did pay a company to analyze Bing Maps data, they only did that for bigger and more popular cities such as Paris. Most locations pull data from OpenStreetMap where data about buildings is crowdsourced. Buildings that have typos in their crowdsourced data will often create bugged buildings in-game.
then you add DLSS on top of that, insert the meme "yo, I heard you like image reconstruction, so..."
Gpu power, memory volume and speed, and data transfer rates had to improve to allow that to be applied.
aren't most of the buildings just generated with normal autogen, like FSX or X-Plane? Also the high detail cities like New York, Toronto, Tokyo etc still use manual photogrammetry scanning
Interesting! The advances in AI and satellite imagery make this type of technology so accessible to everyone. 🤩
You’ve tapped into a universe of value. SAR, Hyperspectral, and gravitational sensors on low cost satellite swarms means a host of applications that will change risk, finance, insurance, agriculture, logistics and more. It’s a trillion dollar opportunity that few see, let alone know how to apply. AI makes it accessible and actionable because on satellite processing lowers the cost of tasked data sets. Glad you are showcasing a topic that excites me.
Hey, not sure if you care about my opinion at all, if you do: I really loved your channel so far. And doing some ads is fine too. But please make sure you clearly state where an ad/sponsoring segment starts next time (you made very clear that it was sponsored when you ended the segment, but i was so shocked that i almost closed the video before that). Your integrity is one of the highest goods you have and doing such smooth transitions from the topic of your video to the sponsor is one of the easiest way to lose the trust we have in you.
I believe what u meant to say was “thank u for bestowing upon me so much insight and knowledge, but seeing as how I’m such a sensitive and anal t wat, would u kindly consider delineating ur ads more clearly?”. I fixed it for u. Ur welcome
Disneyland at 8:20
Palmanova at 13:41
Ogata at 13:50
I have worked in this industry for almost 15 years, working on projects and contracts for many of the products discussed in this video. While geospatial machine learning/AI products are getting better by the day, many of them are not quite ready for prime time. While I don't disagree with the conclusion of the video, I would say that the biggest challenges ahead for these firms is identifying buyers for these alpha or beta state products. Either way, fantastic video! One of my favorite channels on youtube.
the problem is at the end of the day we don't actually have AI. the current machine learning is in reality multidimension optimization instead of single dimension optimization in the past.
@@wei270 Well said Wei, I totally agree with you. Multidimensional optimization is great, and has a lot of uses, but it's definitely not AI. I hope there will be a shift in the usage of the term "AI" in the future, but I'm doubtful.
@@wei270 computer thinks for me. good enough to be called AI lol
@@DimbleWally It's just a layman's term at this point, it doesn't change the profound impact of the tech.
The companies I've worked for that have succeeded with AI based products normally use machine learning as a filter for obvious cases and then pass on anything with ambiguity onto a team of humans that make some set of judgements. Generally this is a very useful business model since it means you can solve problems that weren't economically feasible previously before the filtering step.
I am amazed by the depth and breath of this channel, as a remote sensing person. This went beyond my expectation for a layman explanation of remote sensing, very clear very cogent just like your other videos. Kudos and very well done!
isn"t masterworks a scam ? I had great respect for this channel, wtf.
Yes, they fudge their numbers by for example not including art that did not resell in their profit calculations.
This isn't entirely related, but it's a fun story: In the late 50s/early 60s, the US determined that an increasing number of Soviet Union soldiers must be stationed in Cuba. They did this by noting that spy plane photographs showed a lot of public parks were being marked out as soccer fields. Cubans play baseball; Russians play soccer.
I don’t know how you come up with such great and complex content on such a consistent schedule. I really admire your work!
As a researcher on this field i can say that AI is the new "hot stuff" in image processing and analysis in remote sensing, but as the technology progress, so does the work/headaches. There are so many variables to be considered that, in my view, interpreting results with quality and distinguishing between good and garbage was, is and will be an "art".
I think AI helps in the heavy lifting, but the final conclusions and corrections have to be given by a specialist. Nowadays there are so many options at the click of the button, that many mistakes are easily made, which of course i also made, and took time and study to correct. I don´t mean that it should not be used and improved, but we should avoid relying solely on that, just because it worked before
Anyway, amazing video, keep up the good work.
Can I ask you for some suggestions and recommendations for new crazy students in this field? Thank you.
@@LearnItalian24-mz7hg Of course, ask is the way to learning.
I would recommend to build the knowledge to the ground up, first the basics of remote sensing, GIS and its techniques, how it works, pros and cons. A good chunk of people goes straight to AI/Neural networks/deep learning things and complicate things when a good old technique or some clever workarounds solve it just fine.
Then in parallel get to know coding like python, IDE and google earth engine to name a few, those things will help automate big and bigger workflows, with time, getting to know AI tools will feel a natural extension.
When you get in this stage, get some course on udemy/coursera or good book about this subject to guide you.
Also, remote sensing is an art in some ways, accumulate experience, do analysis by hand to get the feel for adjustments and technique uses, the quality of your results depends on the quality of your know how.
Best of luck to you.
@@LearnItalian24-mz7hg I would suggest you to learn well the basics, the concepts and techniques, there are already tons of techniques that solves most of the problems, in parallel, learn to code, python is a great start, then you can use what you learnt in a bigger scale, automate workflows and so on. With this know-how you will know better on when and how to apply AI to those analysis. Some things in remote sensing are like art and with experience you will know how to distinguish and characterize it.
Best of luck to you.
Tahnk you man@@0tiochico
As a map geek, this was one of your best videos, Jon! You put so much information in ~20 video
The use of cell phones by Nokia to help generate road paths and keep them current. Hence autonomous vehicles.
@@brodriguez11000 - doesn't the Waze app do that too?
One of the "big" tricks in identification is based on spectral resolution. You can pick out a given material based on its reflected or emitted spectrum. This applies to everything from chlorophyll to asphalt. With the right filters, you can pick anything you want out. The hard part is knowing what to look for. As a for instance, with stored full-spectrum image data, you could probably do a study of roadbuilding by looking for the emission lines from the long-chain hydrocarbons used to make asphalt.
Noooooooooooooo! Not a masterworks ad ;-;
Well get that bag anyway jon, you deserve it!
Nobody deserves to profit from a pyramid scheme.
@@lekhakaananta5864 Except the Egyptians.
The sponsor transition was devious.
Yeah, I didn't give this video a Thumbs Up because of that. I want to know when I'm hearing an ad. In this case, it took much too long to notice he was off-topic.
@@GregConquest Add is not good for creativity, interrupts your thinking.
At minute 13.42 It's a fortified city of Palmanova, North East of Italy, founded in XV century . I live 20 km from that city and i have done my military service there in 1984. Thank you for the imageñ
would you recommend visit that city? it has a cool street layout, but from a quick street view walk, it actually did not look that interesting.
im amazed how much this MAsterworks managed to get in everyones channel
YES. Thank you, this topic is bleeding under my nails since months. I recently joined my local Openstreetmap community and I really feel that disruption is around the corner. ML won't see street names, but streets curbs with 10cm accuracy. Places like India are still missing mapped roads in many places. Good satellite images can see the street conditions nicely, and daily. We map based on that, and ML will make it better than us. We have gigantic amounts of training data by applying the existing map and satellite data to train the model. I think it's why Microsoft and Meta bound together to create Overture maps, while putting OSM into stress but also opportunities. Look at Tesla FSD beta, it maps roads live with high accuracy on the fly while driving. Give this a 10x or 100x progress and we have disruption.
What a great presentation and well organized digest of a subject I knew nothing about.Well done!
Asianometry video about AI in spaaaaaace! On Friday! let’s gooooooo!
big data and satellite imagery can have a significant impact in various fields such as agriculture, urban planning, natural resources management, and disaster response.
For example, in agriculture, satellite imagery can be used to monitor crop growth, identify areas that need irrigation, and predict crop yields. In urban planning, satellite imagery can be used to track urban growth and identify areas in need of infrastructure development. In natural resources management, satellite imagery can be used to monitor deforestation, illegal fishing, and other activities that may have a negative impact on the environment. Finally, in disaster response, satellite imagery can be used to identify areas affected by natural disasters, assess the extent of damage, and aid in the distribution of relief supplies.
Overall, big data and satellite imagery provide valuable insights that can help organizations make more informed decisions and improve the efficiency of their operations.
All of the Asianometry UA-cam vids are excellent. Absolutely the best, but the internal advertising degrades the quality and cred. Keep up the great work.
"but the internal advertising degrades the quality and cred.". So, you're gonna pay for his food and rent?
You never revealed what the low-resolution image was.
I guessed Disneyland.
Just completed a remote sensing minor, you nailed this topic!
Great script, ChatGPT could not have done better. 😀
I love your videos! Keep up the great work and covering all of these intriguing topics
Always a pleasure to listen to👍🏻👍🏻祝你新年快乐，万事如意🎉🎉🎉🎉🙏🏻🙏🏻
I like your channel a lot but I really really really really wish you didn't shill Masterworks or anything like that.
I can't Believe you segued into a Masterworks advert without signalling it in any way. Shame on you for not clearly distinguisihing between your content and advertising.
Disappointing how you segued that sponsor ad into your tech story. Count me out, over-and-out.
Been watching your channel since some of your first episodes. I have recommended you to many people as your videos are some of the most informative I've ever watched. Keep up the good work!
"Traditional classifiers miss a lot of sarcasm".
Wait.. oh, he said sargassum.
But It's still true that language AI often doesn't get sarcasm.
It would've been better if the open source project's link was added in the description.
Isn't Palantir doing this already?
Yes
14:00
For the farmland detection we have already using infrared satellite images. They show clearly man-made objects, different crops, forests and water.
Fantastic video! You get better every time. Much love from Victoria 🇨🇦
0:56 what a smooth and sleepy transition from the title topic into an ad. not terribly cool.
The only possibility that's missing in all this is privacy.
Nobody can hear you screaming about privacy at 80,000 feet 😭
ceilings
@@tradfi950 no
Satellite imagery coupled to drone coverage, and with these new in-home automated monitors like Alexa... no place to hide and be alone.
@@0neIntangible exactly.
There is additional data gathered from sensors that are not radioelectric. Gravimetric or magnetic.
Disappointed by the sneaky advertisement for crappy art.
Everyone here should be knowledgeable enough to know what an ad is.
I have never trusted an ad on face value... Do you write complaints to anyone running a detergent ad? Or a tire ad?
Yeah it was not on par with this channels usual quality
If you want to sell crappy you gotta be sneaky.
Great work. Thanks for sharing. Learned so much. Thanks for steering away from hyperbole!
dude you are so good at this
This Masterwork add is highly questionable, 1,7 trillion sounds much but is just the GDP of the Netherlands and Belgium also an insanely small market which seems inherently risky. While also it seems to be at an macro economic scale to be just a bad time to invest into alternative investments as there already is getting a lot investment into them so it kinda seems like telling people to buy crypto when it is high.
From how i see it this are extreme high risk new finical products and seems rather just something for sophisticated investor that understand they are taking a massive risk but it seems rather unethical to push this onto average people as an save investment. From what i think i know i don´t think any expert will tell you to buy more then 10-20% as the high risk portion of the portfolio which could burn totally.
Thanks for your work, enjoyable to watch as always
one czech company has a cubesat that has machine learning algorithm bthat decides which images are good and interesting and only sends them back to earth, they cannot send everything cause of the limited bandwith.
You are very good at this!
Some of the coolest satellite images I've seen this year were the Tonga volcano and the shockwaves transiting the globe.
No mention of how lidar seems dissolve vegetation? While not satellite images a lot of pre-Columbian ruins were identified in Central America recently.
I can see how this technology will revolutionize battlefield surveillance and things like nuclear compliance.
the utilization of large-scale data sets and remote sensing technology through satellite imagery can have a significant impact on various industries and disciplines. The ability to process and analyze vast amounts of data, in combination with the high-resolution images captured by satellites, facilitates the optimization of decision-making processes and operational efficiencies in fields such as agriculture, urban planning, natural resources management, and emergency response. For instance, in the realm of agriculture, the implementation of satellite imagery enables the monitoring of crop growth, identification of regions requiring irrigation, and forecasting of crop yields. Furthermore, in urban planning, the utilization of satellite imagery permits the assessment of urban expansion and identification of areas in need of infrastructure development. Additionally, in natural resources management, the utilization of satellite imagery facilitates the detection of deforestation, illegal fishing, and other activities that may have a detrimental impact on the environment. Lastly, in emergency response, satellite imagery can be employed to determine the areas affected by natural disasters, evaluate the extent of damage, and aid in the distribution of aid. In summary, big data and satellite imagery provide valuable information that can enhance the intelligence of organizational decision-making and enhance operational efficiencies.
The thing with "Super Resolution" and all other linear algebraic techniques is although they will on average match reality that is only because reality is regular. A.I. produces what is expected to be their by making a guess that is on average right. But what it does that by taking the "average", it makes a prediction that everything will look the same as the training set. Training sets are limited and biased is known and probably unknown ways. When reality departs from the the training set A.I. still assumes the training set is truth. A.i. will predict the training set. When something outside the correlation between training set members occurs A.I. flounders. A.I. doesn't derive principles, it predicts the usual. Because what is usual, usually happens A.I. is accurate. When the unusual happens A.I. usually fails, but those failures are unusual and who cares about unusual events?
I'm sorry but that sponsor was shoehorned in a really uncomfortable way in your video, i had to dislike the video despite finding the rest of the content really insightfull.
I personally like the idea of using this tech for land classification, that is, what type of vegetation or farmland a given area has: is it mostly eucalyptus or pine trees; corn or wheat; desert or grassland?
I remember using a mobile app showing land identification/classification a few years ago (and whose name I've forgotten), but it still had some blatant errors in my area...
very interesting and timely. thank you for all the work and then sharing.
Love your Channel, mate. A welcome diversion!
Pls no more ads👍🏼
Ugh @ masterworks sponsorship, super gimmicky investment scam.
For the purposes of image recognition, 'super resolution' is 'object detection', they're equivalent.
To understand this, think of what happens if your ML algo hallucinates pixels so that pixels that are of a garage are upscaled to look like a truck.
"Can you guess what this?"
Given the size and shape of the two areas that seem to be theme parks, and the large parking lots to the direct west and south east, as well as the large seemingly convention centre south, Disneyland
I would like somemore videos about microfluidics.
Congrats on getting sponsored!
This is a massive industry with massive implications. Especially for militaries. China in particular is investing big time in this field to acquire the ability to track US mil assets as much as they can and as close to real time as they can. This synergizes with their massive ballistic missile fleet to give them the ability to strike US assets thousand + miles away. While the US has developed and will develop a constellation of satellites to track (in real time) missiles such as ballistic and hypersonic missiles. This track will cue other missiles to shoot down these threats. Its a back and forth race between the two
hey i give ya props on that segway my boy
A terrific techy video.
So, what I took from this video: the b.s in that CSI show where they zoom in and in and in, and enhance a grainy video will now actually be possible.
Great video - perhaps a future one on smart grids? (power system stability, smart meters, federated learning, data privacy, cybersecurity)
17:40 where can I get a link to this? Would love read more about walruses!
yayay the MetOp as the thumbnail
+100 social credit from amateur weather satellite reception community
I can assume a lot of classified work is being done with deep learning and hyper spectral imaging.
Super-resolution is false resolution. It is essentially AI cartooning, not enhancement. The cartoonist draws a higher-resolution "artist's conception" of a photograph. The result is an artificial hypothesis, an illusory parlor trick, not a photograph. That's what smartphones produce today, where the 1 or 2 MP diffraction-limited micro lenses somehow result in the 10 or 20 MP image files. We are then round-tripped to believe the lie that these cartoons are photographs because they are plausible, better resolved, and prettier than the truth. It's a lie we want to believe. AI flattery. A photograph is a scalar intensity map of a light field's surface. Images produced by human or artificial intelligence are not photographs, they are drawings of imaginary models.
Agreed. They are interpolations/extrapolations -- essentially a sophisticated "best guess" --- but still a guess.
You should have included that pixelated portrait of obama that the upscaling technique turned into a generic white man as a demonstration for the limitations of the single frame super resolution algorithms.
Great video as usual!
Regarding Super-Resolution (around 7:30), perhaps GAN (Generative Adversarial Networks) is another one that worth mentioning. GFPGAN is an example of this, and since it is trained on human faces, it is capable of doing an exceptional job with the super-resolution of facial images. Same thing can be done with GAN for other types of objects or scenes.
Regarding Satellite Image Analysis (around 10:44), beyond the four types mentioned there is another: Scene Understanding. This is about making a general description of a scene, such as "there was a tornado that passed this area" or "this is a farm land that has suffered severe bombardment with many wrecked vehicles," etc. For example, the Salesforce BLIP is a system capable of doing this type of analysis. This is quite different from using deep learning to classify images into fixed number of categories, since it is capable of recognizing a very large number of objects and scenes, and then provide a textual description. Other than using this for satellite image analysis, this also has important applications for many areas, such as many types of autonomous vehicles/robots that operates in the open world.
How is "scene understanding" done? Do you have to train your deep learning model on a repository of images that have been appropriately text-captioned?
@@manofsan Yes. BLIP (or OpenAI's CLIP) is pre-trained on a large dataset of image-caption pairs in order to understand the relationship between text and images.
@@kaihuchen5468 - what a coincidence, I was trying CLIP the other day, and I've just now tried it again because you've mentioned it. I found that it generates some useful text with extra unrelated garbage words added on. Perhaps that's because an input image can generate broader text associations than what's useful. Interesting tool, though.
Yet another prospect over the horizon is to train ChatGPT on the text derived from applying Scene Understanding to satellite images. Here I will let ChatGPT do all the explanation by itself:
My question to ChatGPT: If we apply scene understanding to satellite images to create textual descriptions, then train ChatGPT on these text, what kind of advancement can be achieve in the management of satellite images?
Reply from ChatGPT: Scene understanding, which is the ability of a computer to interpret and understand the contents of an image, can be applied to satellite images to generate textual descriptions of the image content.
Once the scene understanding model is trained, it can be used to generate textual descriptions of new satellite images, which can then be used to train ChatGPT. By doing this, ChatGPT would learn to generate text that describes the content of satellite images in a coherent and contextually appropriate way.
This process of scene understanding and text generation can bring several advancements in the management of satellite images:
1. Automated image annotation: By generating textual descriptions of satellite images, the images can be automatically annotated and organized, making it easier to find and retrieve specific images based on their content.
2. Image search: The textual descriptions generated by the scene understanding model can be used to perform text-based image search, allowing users to find satellite images based on their content.
3. Image captioning: ChatGPT trained on satellite image descriptions can be used to generate captions for satellite images, which can be used to improve accessibility and understanding of the images for people with visual impairments.
4. Image analysis: ChatGPT can be used to analyze satellite images, by identifying patterns and features, it can be used for different applications such as land-use monitoring, natural resource management, and disaster management.
Please note that all of the above are based on the assumption that the scene understanding model is well trained and accurate. Also, the quality of the textual descriptions generated by the scene understanding model will affect the quality of the text generated by ChatGPT.
@@manofsan The internet-scope data that CLIP trained on is quite noisy. That's why Salesforce's BLIP put in some effort to deal with noisy data in order to produce better textual description.
4:10 Laser comms will further increase the amount if you include high quality video footage that currently cant be transferred because of bandwith limitations.
I really like your stuff, and am 100% behind getting sponsored. I am just saying, I am absolutely certain that Masterworks is going to be revealed as a scam within 18 months.
Wait, ain't this masterworks a scam project by those who own Kamikoto knives and Established Titles?
i personally experienced this ad read as somewhat inappropriate. the sentences leading up to the sponsor reveal did not make it clear that they were actually part of the ad read and as a viewer i was kept in the belief that the arguments given were part of establishing a base to support the hypothesis of the video, which i experienced as quite misleading. ad reads are fine, but please make it clear to the viewer that they are watching one :)
Hi John, I love the videos you produce. But to younger audiences out there, remember that masterworks is a sleight of hand, it is like one of the overpriced items you are persuaded to buy but serves no utility or purpose. Please be informed, billionaires don't invest in art, they just park their excess/ black income to avoid taxes and tax authorities' / government's oversight. That being said, never invest your/your parents' hard earned income unless you understand the asset and its market from a purely finance perspective. To me masterworks is just another FTX in making.
I am a little surprised there was no mention of satellite optics, but I guess that's not at all related to semiconductors, and it's, surprisingly, often done by hand. Huygens Optics channel has a neat series on it.
Good video! I love the AI generated images too! What was the words used to make the image at 0:04 ? The planet earth with a black starlit background with a satellite in orbit?
I thought he’d mention searching airplane crash locations and missing ships and boats. That would do wonders for finding people fast. The cameras could be satellite, boat or airplane based.
Very Intresting what development has happened to this term with the help of ai. It makes me remember back to the year 92 when a CD Rom set of satellite images from Germany was sold to watch it on a Windows 3 PC.
13:51 is Akita, Japan. Used to live there
Thanks for sharing
8:17 Disneyland. Not sure what it says about me that I recognized it instantly
Fun fact, Nazi Germany took the first picture of earth from Space, from a V-2 rocket.
I can't say that's a terribly fun fact. Any win for the Nazis, that was a loss for humanity at large.
HAL-9000 vs. Skynet vs. chatGPT: which one is more powerful?
chatgpt cant launch nukes
I'm a big Skynet fan.
Define powerful 😄
ChatGPT exists, and therefore has more causal powers irl, but within the works of fiction that those other AIs exist, they are more powerful that ChatGPT is irl.
This seems obvious?
ChatGPT hasn't achieved time travel --- yet
Great video!!
art is soul manifest
soul is that by which we desire
desire is a vacuum spirit is a wind
All these images could have come from commercial or military planes. You haven't shown they come from "landsat".
8:20 That's Disneyland. The large structure in the upper left is Starwars: Rise of the Resistance and that's the 5 freeway by the upper right corner. What did I win?
6:41 that was a dalle image right?
This is cool content!
Wish google maps can include closed roads/highways from government works data
So why not apply these models to imagery collected by drone? If you could extract objects list manholes and catch basin these a market for that.
This channels content is insane
Anyone else getting that "ENHANCE !" vibe ?
Please, separate clearly advertisement, don't sneak it into the video.
Does anyone know the best keywords to use when searching for these AI models for working with satellite imagery? For example if you want to count the number of cars in a satellite image, you need a model that is specifically trained for object detection from aerial images but when I search for "satellite image object detection" I feel I'm only finding a small number of results, whereas I expect there is plenty of research/discussion that's going on that I'm just not finding.
Can I ask you for some suggestions and recommendations for new crazy students in this field? Thank you.
i got the short end of the genetic stick and have multiple chronic genetic diseases.
I am hoping AI processing can help to extrapolate the numerous simulations of medical treatments. (sorry about the way the sentence is written, i can't think of how to write this properly)
(also apologies if it's off topic... i see the "unlimited possibilities" and just wanted to say something.)