311
Deep learning CV and Earth Observation data. What satellites can see? Monitoring of Environment, Land management; Urban planning and Mapping, Mapflow.ai etc. Сontact us at hello@geoalert.io or @godnik0
Thanks to the collaboration with Andrey Pirogov (gisgeo.org) we started our new regular campaign ("#Mapflow weekly update" / https://lnkd.in/dbS8Ex35), based on the information from our social media and blog for the past week.
Subscribe to our #newsletter to stay tuned for the product updates and applications of #Machinelearning in #Geomatics.
https://mapflow.ai/subscribe_newsletter
The #benchmarking project is going on.
We are testing buildings model against the manually labeled aerial images in #Africa, Zimbabwe. #Openaerialmap
Ukrainian Cities in the dark after Russian strikes on power infrastructure - how it can be seen from Space. (NPP nighttime imagery) ⚫️
The illuminated area of the capital city has reduced approx. two times comparing 2021 and 2022. #fuck_war
What difference it makes after tuning the model for the specific territory.
By "tuning" we mean the additional training of the model using the relatively small number of labels. In this case we tuned the model for Saudi Arabia 🇸🇦. #buildings #saudi #segmentation
Lawn watering in Tashkent (photo by Vlad Ignatiev, COO)
#random #life
Geoalert September Newsletter 🍁
#Know-How? Pan imagery
Panchromatic satellite image looks greyscale as it uses a single band for a greater spatial resolution. Pan-sharpening is a data fusion technique when the pan image is combined with a low-resolution multispectral image to achieve a higher resolution while preserving all spectral attributes.
The segmentation neural network is basically indifferent to pan-image if it's trained in an appropriate way of augmentation. E.g. the building footprints extraction model works good enough on panchromatic images as it used to be on RGB.
Sometimes you just don't have the actual RGB image. E.g. in #Maxar satellites, they have the WorldView-1 that carries only a panchromatic image sensor.
However, when it comes to more sophisticated and not-so-universal algorithms as a classification by building types - it's better not to use panchromatic instead of RGB as the model is trained on the color images and is used to more features.
Online payments in #Mapflow is back! 🎉🥳
Thanks to integration with #Payze we restored the option to top up the user balance in credits.
Here you go to check the updated user profile in case you are running out of credits for Mapflow.
BTW If you are interested in premium satellite images or extended support you are welcome to subscribe to one of our commercial plans.
#Mapflow - #QGIS plugin got the updated version (v 1.6.4) - now it shows the data validator messages as well as the Web app.
We encourage you to update if you use QGIS working with our services.
The correct answer to the last #GeoQuiz (and we got the only one approximately correct to our email) is:
This is some territory in mainland China 🇨🇳 as you may check by coordinates on the image.
And all the maps in China (accordingly to the Chinese laws) are transformed to some secret local coordinate system. 😎
The difference between Mapbox and Google is because only Google transforms their maps to Chinese CRS.
The interesting fact is also that Hong Kong doesn't apply these rules.
Why is this shift?
Mapbox Satellite vs Google.
Send us your answer to hello@geoalert.io
We will post solution the next day and top up your Mapflow account if your answer is right. 😁 🧑🎓 #GeoQuiz
Mixed photos from #Geoalert #team summer vacations.
Summer is over, the adventure never ends 🤟
Read our first longread 🙈 vision about building the #SaaS product in the market of Earth Observation analytics.
Some thoughts to highlight:
> If you build some SaaS product to automate the analysis you have to be cautious about your market positioning as long as the Data providers rule the game
> SaaS highly depends on the Data as a service model and both contribute significantly to the market development by increasing the data consumption and engaging more analysts to start using EO data
> Remember video and music subscriptions on demand? You can stream movies but you cannot download’em…
> What proves them to be SaaS companies — one can start playing around with the product as long as he has an account registered
> Look, when Mark from Descartes Labs is regretting the way they wouldn’t but could be the “modern AI consulting” company — he is talking about large customized enterprise solutions, isn’t he?
> In our case, it means that Mapflow has two revenue streams: Subscriptions and Tailored Projects (services)
> Developing this market through the services is a lengthy road, so the innovators should beware of the long-term commitment to create an impact and turn the ideas into commercial reality. Building a SaaS / B2B product looks like a viable and reasonable strategy here
> As they used to say in Openstreetmap — “The map is never complete”. Not only because of the blanks or incompleteness of the map features. There are changes that go all over the World. And having them timely detected and interpreted in the way of location intelligence is the Map no one can complete but everyone in the community can contribute to
You can read the full article here
Hotspot anomalies (#fires) and the smoke plumes in the area of Enerhodar where the Zaporizhzhia NPP is (AUG 23). The nuclear plant is disconnected from power grid after nearby fires. ⚠️
Читать полностью…
The #Mapflow data validator has been updated - it checks the area of the processing against the availability of imagery source. If your area doesn't intersect with the image source or the image resolution doesn't correspond with the model's requirements - you won't be able to run the processing using this model. 👌
Читать полностью…
Cool stuff by #Kontur. They are developing the new model for estimation of losses caused by natural disasters.
"The highest accuracy score was achieved for estimating tornadoes’ monetary loss due to the greater number of events and their narrower spatial spread."
https://www.kontur.io/portfolio/disaster-loss/
#news #partners
We've updated the #pricing page of #Mapflow. Hopefully it's becoming more clear. (However the integrated billing system for all our application is still under development).
Pay attention that we added special terms for Academic license. Feel free to ask your questions we are ready to help.
One more picture from the Space and from our team.
This week as you may know started Diwali - one of the most beautiful and bright World fests. For us it also represents the victory of light over darkness, the life - over death and war if you wish.
Sometimes it's getting darker in the other parts of the World, but the light will overcome.
❗️Service downtime announcement
We are planning some maintenance on our servers, so all our services, including Mapflow.ai, Mapflow API and QGIS plugin will be unavailable on Saturday, October 22, for the 2-hour period in 15:00 - 19:00 UTC.
In the meantime, we are preparing some major releases that we will announce next week, stay tuned! 😉
City patterns extracted by user with #Mapflow #Roads 😎
Читать полностью…
https://www.space.com/crew-5-anna-kikina-first-russian-on-spacex-mission
Space is international and it shows us the way we can live and cooperate on Earth 🌎
Multistorey building patterns in Domodedovo, Russia. #Mapflow #Buildings_with_height
Читать полностью…
We've come across a picture of aerial image embroidery (source) and got curious: what would the neural network see here?
After masking and generating the georeference (at spatial resolution of 2.6 m per pixel) we could process it with the #Forest model at Mapflow.ai. Not bad, what do you think?
Maybe #AI still can't admire the beauty, but it certainly approves that this is a very realistic piece of art! ❤️
We started working on the #benchmarking project the goal of which is to test #Mapflow models on different sites and patterns across the Globe.
The benchmark locations are based on the statistics of Mapflow usage - the more frequently processed areas.
We will share updates soon.
We've implemented the billing (processing) history into the User Profile section. It's the first iteration enabling you to search across your processings by the period of time. Next, it will be extended with the functionality to select the dates and download the billing report for those who are subscribed to the commercial plans.
#Mapflow #Updates
We are looking for #QA / #Support engineer 🔥 - check our openings - to join our team. Perfect if you have some #GIS background as our work is related to maps and data.
#vacancies #job
We took it forward to test #Mapflow on the imagery samples from Chinese satellites provided by #HEAD Aerospace Group. Although there is some difference in the smaller details visible in the images if compared to the highest resolution (#EarthScanner: 0.5 PAN, 2 MS vs #WorldView-3: 0.3 PAN, 1.25 MS) - the #buildings_detection algorithm works perfectly well.
It outlines building rooftops and predicts building heights by automatic estimation of the length of the building shadow. We will share more details on further experiments with this new prominent source of imagery.
#Kontur map shows the huge areas affected by #floods in India. The map also helps to draw the attention to the blanks and gaps in the #Openstreetmap coverage against the affected areas.
Читать полностью…
We are getting more requests about buying credits for #Mapflow. Our previous online payment system has become unavailable for international payments, that's why we look for and move to another one and it takes a while.
In the meantime Mapflow users can buy credits paying by invoice or reach out to us and we will find the way.
For those who are ready to switch to one of subscription plans we have them more to come, including the use of commercial satellite data by the leading data providers.
Check how it works with #Maxar - https://youtu.be/FlV7WLfFDrY
Southern France #wildfires Jul 18, #NASA thermal anomalies map
Читать полностью…
We have released the #Mapflow data validator. Now users can see the detailed description in case of the failed processing ⛔️.
You can find the whole list of error codes / descriptions in our documentation.