Merken

Open Source Software For Land Cover Mapping From Remote Sensing Data

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
search company based in Belgium and ash talking about the sort of where the money to buy more thank you very much and take you for being so patients by can fulfil the expectation that we must be very high by now waiting for a long time so I'm going to talk about open-source sulphur for land government being based on most sensing data this is the outline of what talk and of give the background all also offered for remote sensing and that will jump into a case study which are used to show how this self where was used and abused the methods and some results before include the 1st
backgrounds for English growth promotes as the application that has been previously reserved for the good of commercial packages as we know for example and the imaging commission some also mullahs and likely there has been account of for the over sauce starting of caught with a well known and utility are tools with pie-eyed and although researchers because this is and in the research employment where I'm from many people are using still the commotion packages because the not so either capable or interested but I don't have the time to to use the to to really implemented the algorithms that are needed for example of the machine learning techniques and for that there are some some other packages like below across the Geisel a suit and there is a bit more recently there is a very nice package of the contest space agency in Brown Sofia PPL books and the next speakable vocal about or but that it is quite a large package with an 80 idea velopment Environment command line in the face and and a go with the package I'm talking about now in India's speeches is the peak a tools eye start to work on that on all my might each the White a musical and eye decided to to use it as an open sauce its it's much more humble than that of the 2 books and it's only a command line that want to show what was possible with the case the
study it so it's a big if you can call that of the actively every year to die Tripoli just as remote sensing society is issuing contest and got this is all this about data fusion and mostly have to classify automatically an image and this year it thousand 13 there was a high respectfully which a quiet over used an audience in urban area at together so it had to be taken few the same with a digital surface model a quiet delight sense and those of you are not familiar with high prospectus images of those are of a special kind of all of a digital camera census that are acquiring data in India in the large number of spectral bands so instead of having like we are Logic the camera with only the bands to individual a spectrum it has like for example the case I because he said the has won an 244 very Nehro spectral battle of information to the with a special resolution was to win of up to a half Amida but pixel and that was also the case with the digital service model the digital service mullet gives you hides information so in implementation with a high respectability love information not only on the spectacle side but also on the high so that was a quiet with a minor Lazenby's based sensor and was only 1 band of information which is high and it has been a quiet in the same space resolution and images were already gorgeous that was already helpful the contest
consistent in an automatic classification of 15 classes which is he over here all going back into more detailing the classes in a
minute when al talk about the challenges that the challenges of cost Dostum Challenges otherwise it would be accompanist and 1st of all there was the day diffusion you had to find the best way to combine the 2 datasets delighted information of high hype and his Bickley's which in order to open up to more classification so that the best desiccation results 1 of the challenges was the similarity of the classes some of the cost of very clear near to each other but there was also a cloud shadow cost only 1 part of the match which will turn out to be quite a difficult thing and and and insulted actually saw what already said that now that it had to be automatic so it had to be in automatic last occasion no official interpretation involved and the extra challenger would here is was interested in how can be sold only using only of over sauce tools so
a bit more on the back of the class is seeded classes legends over here and see the 3 classes for across the across the healthy cross stress cross and synthetic across so this and that the grass was for Forest for still ahead in Australia had been Mubarak which was where people run of quite often into this this was distressed Grasso and not the Nobel across the state and could grow to help the across and from the visualize it's very difficult to distinguish tools likely we had high prospective data which is very capable of dealing with those civil differences in class and then 1 of the challenges he joined the not to be much more difficult than the similar classes of land cover which was that some of those closest to a mix of land cover and then use that covers typically well related to promote sensing but because 2 different types on the ground that the problem is when you have to land use that the club's can be very Simba are getting the same actually from the picture from from from the from the material itself that is just how we used as an example residentially commercial area for remote sensing application sections saying the same with rose and highway their build of the same material but that is just how you use it in the club's deferred and then find the foreign apocalypse there were 2 types of 1 was a free parking and the other was to be taken part this is
very difficult to see here for full for that but there was huge Cloud shadows said a shadow clout was hanging over here and it costs a shadow all in this area which is very difficult for another my best occasion that act because actually want to classify order computer classified this building its is another building commercially area here it has to be classified as a bust but from the from a spectrum is still the different because of the culture OK then about
the methods it's it's about classes to provide classification which means which is a machine learning method of which means that you have a number of labelled pixels which were provided by the committee of the organising committee and it enables us order computer to learn how to classify and unseen data so there is a subset which is provided training said and that the approach was using open-source tools only heavily based on Jilin O'Shea are and on the speaking tools which is built on top of that for a remote sensing applications and and machine learning and a would like to invite you to if you are interested you can run the and methods will prisons now and you can download the data you can download the trading said that and also the tools and I'd and generated everything in a script and tried to explain it in week I page so I interested you can just run through it and then do it for yourself As mentioned
the tools are there have been released on June Jeetil 3 you can download them their subdued in Savannah is based gigolo are for some it might be a occurs but for me it's a blessing it's only command line driven just as a G Delve tools are currently only a Linux it's in the interest actuarial on the to do list to to reason for Windows but it's not it's not there the
new so all the talk Ro'i's seeded data that was provided by the organising committee is a minor data is Acoustiguide spectral data and there was a strange simple and a chaining sample was just playing sea is the filed with for each role of 19 number belonging to the latitude where you can be fined pixel and the and the class of long so and based on that that was a train set on the 1st programme to this was a good version of the act testified to affect a five and a used ogee are likely for that by converted in infected filed and over late that factory although the
hype spectrally in order to extract the values that were below those trading sample and when a subtract I'd are extracted them by creating created a new filed with us at tributes the actual condemned the value of the pixels and so that it became a eye obtained vector file containing all information and even subdued actual training for the last 5 to give you an idea
of how was high spectral a signal looks like I'd issued example here of the of the prospect of the degree to which the prospect of the health across agreeing distressed growth in Orange and and synthetic grouse here at what this year this is a spectrum of the electromagnetic spectrum from the usual Fox all the way up to the new infrared and what you see here quite interesting so this for example for the health across the middle of line here is the average spectrum and and I'll social good at a distant and aviation for different pixels of the particular Klaus what you see here this is official parts of being in a group green tea and red if your only see what is he with your eyes is very difficult to distinguish because all the players are all over what you see with a high prospectus answer is that will be all over you have all these bonds here and locally a that is quite easy to distinguish those closest year so again may be on the line they would easy to see their but for example the synthetic aggressive although last across it very difficult for for us to see the difference the cheesy meeting this year's difference in the in the New River the same for
for the bill and that was even more difficult because as if he were talking about the land use different of the houses that were built ordered constructive for the housing for the foot residential area care much like the the buildings that are used for commercial area see spacing commercially as you tradition and there is a lot of overlap with a residential area even worse for
growth as a mention you have roads and highway de exactly the same as it is complete over their foot for the Trading pixels in in those closest again the
popping up total over firms there is another set by 1
going to which the killed in higher spectral that 1 of the problems in hybrid spectral a imagery is that you have a lovely time and high dimensionality any machine learning this is known as the curse of the dimensionality if your damages to high with respect to the number of training pixels you have its at its it to a new post problems and you will not be sold so it's there is a feature selection in Belgrade by your legal take on the all the best features the seabed from a number of 16 high spectral features the classification was not approve of then once which we got the select features from the high spectrum and it was time to use the data with delight and in this case we just did a future future fusion where we go detonated said 16 spectral bands with a height information treated as 1 in which an extracted that information as input Florida classified now
firms you see that it's already better for example injuries and with the chilling commercial area the total overlap almost here there are still some over this is the hiding from which was saying classes residential and commercial is see that most of the majority of the House of houses in the residential area are until 20 25 leaders where his high where the commercial area starts from from 20 leaders for their just behind you can use any nation where classified till combined combining the 2 weeks later you hope that the cost the distinguished so then the actual of
plastic Asians you use both imagery would be to make it a trading sample by extracting by overly the vectors although symmetry trading classified and then you see the classified also with is actually during a you obtained the find it there was 1 final
step which is a mock of friend who field killed during which also part of the of the filtering processes in the ending the tools that this is a filter that uses contextual spatial information in order to get rid of them this all the better effective typically have been these across based are pixel based castigation so there is a lot of small pixellation effects and with it a mock of and appeal field to filter you are able to climb some of the clubs to be in order to get rid of those easy when ice which back in 4th is that some of these effect is being removed so typically we more interesting and more which is a results and and that's what is so filled with the finding of
some some results the obtained overall acuity was 83 per cent which is not a bad result if you see that would they were 15 classes and there was also quite some similarity also the land use it is quite a difficult challenging problems or whether if you look at the winning solutions that was so much better classified was 94 per cent of accurate accurate if you see a looking more details where actually the differences are in this area most that listing on all that it it's quite Simla where differences are mostly is in this area remember this we had this cloud shadow so why did not sold the Cloud shadow part of all this is the rubbish says the of some what the over here some some wrote that should be roads and this is because it's just old stock stock spectrum and what the winning solution did he was it was some kind of this was due to deal with this and the car shadow and it would take me to far too to try to solve with a genetic over sauce tools so there is no real genetic tool that can do now either in India the commercial sulphur packages so if you see how they sold its very up look very dedicated how they sold and it is just not possible with the playing open sauce solution that is ready for use if we go for example to the 10th place all of the data will only 10 basis published so you see here is already much closer because it is cheaper than the 10th place they did not sell this book by their using here that had been made of the same problems are ahead in my classification there and we see that it is very close and also that the data is also much more closest the to eighty-six which explain that that most of the difference is that if you want to win it was the codes to compute so
we've seen in this was this was a contest which is their every year from by easily it's quite a challenge England covered much more easy glance cover up programs that was high spectral data in full time mentioned that is a cause of the invention of the Test 100 of 40 spectral that had to deal with those there was and also that diffusion that the deal with the light of day the combined with a high stick to later there was a cloud shadow an Irish showed at shows that it can be sold of with a readily available over sauce tools and the result was not among the top 10 with closed its of all which is only command line driven no clicks everything which for me is the really nice because you can put in a single script and executed in in less than a minute everything is is used on sidewalk to change or 2 weeks on some things you can just that this is the file without any clicking bold and if you are interested you can do the same and and just download available files the sort of figure for your
pension back along the for questions were of caution for Last during or woman for the last British after the of some of the questions which is why it is of the last by a side and the endless report in the question clashes and you look at the view that the freedom of speech or of of the with your body shape of the estate I figured it it was a world he published a book about all the other to the good question 1st of all for the classifiers for the moment at implemented another 1 with annual that are division the network did not use it here actually because the support victory she which here it is that is well known for dealing well with the high dimensionality in itself is often used in the attack aspectual a remote sensing and this is the reason why the city here I'm not sure if it would change that much by shoot choosing another classified my experience the that bodies is that it's much more a how you deal with it with a trading Data how you to feature selection tried to put in some other features this is mostly if you with talking about good classified as a red and and and and rather forest and an old and real estate the equally good Gosforth might be is that the difference is falling into the after surging another area like resembled trying to do with the culture that would be the 1st thing to do because of the 2nd body is also quite interesting is that there are apart from the spectral and the and the and the high information that of all the information spatial context used it with a random a modified field their with it was used in the in some sense I've played but with some picture parameters especially trying to deal with the roads and the and highways trying to deal with that there is more which meeting highways and in the rules which are more Nehro the conclusion was that it didn't make that much difference or had to spend more time in the hills of possible ZLI the label to iba's Amadeo able to do a 1 or 2 percentage increase but then for me it was more for the case that it was more interesting to have a quite simple solution which was easy default and and good just data long on 1 package and the world will fall off the as the I'm sure that there are some songs from some clever solutions you can use for that the following yesterday's for ahead of the tragic that use to with is similar to that of the of the girls are so that the new rules sorry and and and in the region yesterday when yesterday use 1 you are not always easy to different things depend shopping tries to create new 1 in which a white being pushed for typically if you have a 2 0 panchromatic image without spektr information on a high resolution you try to put in some spectral integration with and that 2 had the best of both worlds here it was the say it was a different technique you had the spectacle bands and you have the of the high information and I'd just created 1 image concatenated and several bands having 1 extra bandits a as a layer for the height and used for the classified find enough creating new image for shopping which is it might be is mostly useful for the usual intimidation and not so much for a prosecution not Perugia for the full most exact unknown 20 minutes away by the treatment she is pretty like a victory only 2 seconds off the best part
Physikalischer Effekt
Resultante
Beobachtungsstudie
Offene Menge
Software
Erwartungswert
Font
Open Source
Überlagerung <Mathematik>
Markov-Kette
Quick-Sort
Computeranimation
Bitmap-Graphik
Formation <Mathematik>
Bit
Datenfusion
Implementierung
Zahlenbereich
Sprachsynthese
Kartesische Koordinaten
Punktspektrum
Mathematische Logik
Raum-Zeit
Computeranimation
Metropolitan area network
Spezialrechner
Virtuelle Maschine
Informationsmodellierung
Digitalsignal
Algorithmus
Gruppe <Mathematik>
Bildgebendes Verfahren
Bildauflösung
Physikalischer Effekt
Beobachtungsstudie
Suite <Programmpaket>
Pixel
Gruppe <Mathematik>
Datenmodell
Softwarewerkzeug
Bildauflösung
Übergang
Programmierumgebung
Software
Dienst <Informatik>
GRASS <Programm>
Suite <Programmpaket>
Flächeninhalt
Digitalisierer
Information
Datenfusion
Programmierumgebung
Data Mining
Resultante
Offene Menge
Interpretierer
Matching <Graphentheorie>
Klasse <Mathematik>
Ähnlichkeitsgeometrie
Überlagerung <Mathematik>
Extrempunkt
Computeranimation
Metropolitan area network
Mereologie
Abschattung
Information
Ordnung <Mathematik>
Datenfusion
Streuungsdiagramm
Data Mining
Informationssystem
Subtraktion
Bit
Klasse <Mathematik>
Kartesische Koordinaten
Baumechanik
Computer
Punktspektrum
Computeranimation
Überlagerung <Mathematik>
Metropolitan area network
Typentheorie
Datentyp
Mixed Reality
Abschattung
Gammafunktion
Wald <Graphentheorie>
Gebäude <Mathematik>
Ähnlichkeitsgeometrie
Überlagerung <Mathematik>
Software
Flächeninhalt
Mereologie
GRASS <Programm>
Garbentheorie
Ordnung <Mathematik>
Normalspannung
Streuungsdiagramm
Normalspannung
Data Mining
Aggregatzustand
Versicherungsmathematiker
Schnittstelle
Offene Menge
Wellenpaket
Pixel
Open Source
Klasse <Mathematik>
Zahlenbereich
Mailing-Liste
Sprachsynthese
Kartesische Koordinaten
Computer
Computeranimation
Homepage
Gefangenendilemma
Teilmenge
Spezialrechner
Metropolitan area network
Virtuelle Maschine
Bildschirmfenster
Skript <Programm>
Ordnung <Mathematik>
Algorithmische Lerntheorie
Programmiergerät
Elektronische Publikation
Pixel
Wellenpaket
Versionsverwaltung
Zahlenbereich
Vektorraum
Elektronische Publikation
Punktspektrum
Computeranimation
Metropolitan area network
Verkettung <Informatik>
ASCII
Funktion <Mathematik>
Stichprobenumfang
Faktor <Algebra>
Information
Ordnung <Mathematik>
Gammafunktion
Subtraktion
Pixel
Green-Funktion
Singularität <Mathematik>
Gebäude <Mathematik>
Güte der Anpassung
Gruppenkeim
Extrempunkt
Punktspektrum
Metropolitan area network
Minimalgrad
Flächeninhalt
Verbandstheorie
Mittelwert
Mereologie
Spektrum <Mathematik>
Gerade
Metropolitan area network
Pixel
Morphing
Vorlesung/Konferenz
Offene Menge
Pixel
Wellenpaket
Zahlenbereich
Ein-Ausgabe
Punktspektrum
Virtuelle Maschine
Metropolitan area network
Programmfehler
Flächeninhalt
Trennschärfe <Statistik>
Gruppe <Mathematik>
Information
Hybridrechner
Gammafunktion
Software Development Kit
Soundverarbeitung
Prozess <Physik>
Pixel
Vektorraum
Speicherbereichsnetzwerk
Computeranimation
Metropolitan area network
Software
Datenfeld
Symmetrie
Stichprobenumfang
Mereologie
Information
Ordnung <Mathematik>
Markov-Kette
Personal Area Network
Data Mining
Gammafunktion
Resultante
Offene Menge
Subtraktion
Klasse <Mathematik>
Punktspektrum
Computeranimation
Kappa-Koeffizient
Digitalsignal
Abschattung
Skript <Programm>
Skript <Programm>
Optimierung
Figurierte Zahl
Softwaretest
Managementinformationssystem
Physikalischer Effekt
Datenmodell
Mailing-Liste
Ähnlichkeitsgeometrie
Elektronische Publikation
Quick-Sort
Menge
Einfache Genauigkeit
Software
Flächeninhalt
Offene Menge
Basisvektor
Mereologie
Codierung
Datenfusion
Personal Area Network
Streuungsdiagramm
Subtraktion
Momentenproblem
Sprachsynthese
Division
Computeranimation
Gruppe <Mathematik>
Trennschärfe <Statistik>
Bildgebendes Verfahren
Schwarzsches Lemma
Bildauflösung
Parametersystem
Shape <Informatik>
Sichtenkonzept
Wald <Graphentheorie>
Schießverfahren
Datennetz
Zwei
Schlussregel
Kontextbezogenes System
Hill-Differentialgleichung
Software
Datenfeld
Verbandstheorie
Flächeninhalt
Mereologie
Information
Verkehrsinformation

Metadaten

Formale Metadaten

Titel Open Source Software For Land Cover Mapping From Remote Sensing Data
Serientitel FOSS4G Nottingham 2013
Autor Kempeneers, Pieter (VITO)
Lizenz CC-Namensnennung - keine kommerzielle Nutzung - Weitergabe unter gleichen Bedingungen 3.0 Unported:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen und nicht-kommerziellen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen und das Werk bzw. diesen Inhalt auch in veränderter Form nur unter den Bedingungen dieser Lizenz weitergeben
DOI 10.5446/15566
Herausgeber FOSS4G
Erscheinungsjahr 2013
Sprache Englisch
Produktionsort Nottingham

Technische Metadaten

Dauer 25:08

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract Open source software is well established for basic raster and vector data processing, with the Geospatial Data Abstraction Library (GDAL) as one of the most well known tools. Its utilities and application programming interface (API) have become a common standard for data format conversion, reprojection, spatial and spectral subsetting. With its command line interface utilities, GDAL is better suited for the automatic processing of very large amounts of data and for repetitive processing tasks than most of its commercial counterparts. Though GDAL provides an excellent API on which more advanced image processing tasks can be built, not all users have the time or programming skills to get involved such development. In particular within the remote sensing user community, there is a large interest in machine learning techniques applied to remote sensing data.

Ähnliche Filme

Loading...