Add to Watchlist

Machine Learning For Remote Sensing : Orfeo ToolBox Meets OpenCV

215 views

Citation of segment
Embed Code
Purchasing a DVD Cite video

Formal Metadata

Title Machine Learning For Remote Sensing : Orfeo ToolBox Meets OpenCV
Title of Series FOSS4G Nottingham 2013
Author Michel, Julien (CNES (DCT/SI/AP))
Jaen , Arnaud (CS SI)
Grizonnet, Manuel (CNES (DCT/SI/AP))
License CC Attribution - NonCommercial - ShareAlike 3.0 Unported:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal and non-commercial purpose as long as the work is attributed to the author in the manner specified by the author or licensor and the work or content is shared also in adapted form only under the conditions of this license.
DOI 10.5446/15551
Publisher FOSS4G, Open Source Geospatial Foundation (OSGeo)
Release Date 2013
Language English
Production Place Nottingham

Content Metadata

Subject Area Computer Science
Abstract Orfeo ToolBox is an open-source library developed by CNES in the frame of the Orfeo program since 2006, which aimed at preparing institutional and scientific users to the use of the Very High Resolution optical imagery delivered by the Pleiades satellites. It is written in C++ on top of ITK, a medical imagery toolkit, and relies on many other open-source libraries such as GDAL or OSSIM. The OTB aims at providing generic means of pre-processing and information extraction from optical satellites imagery. In this talk, we will focus on recent advances in the machine learning functionality allowing to use the full extent of OpenCV algorithms. Historically, supervised classification of satellite images with OTB mainly relies on libSVM. The Orfeo ToolBox provides tools to train the SVM algorithm from images and raster or vector training areas, to use a trained SVM algorithm to classify satellite images of arbitrary size in a multithreaded way, and to estimate the accuracy of the classification. The SVM algorithm has also been used for other applications such as change detection or object detection. But even if it is one of the most used function of the OTB, the supervised classification function did not offer a single alternative to the SVM algorithm. However, the open-source world offers plenty of implementations of state-of-the-art machine learning algorithms. For instance OpenCV, a computer vision C++ library distributed under the BSD licence, includes a statistical machine learning module that contains no less than height different algorithms (including SVM). We therefore created an API to represent a generic machine learning algorithm. This API can then be specialized to encapsulate a given algorithm implementation. The machine learning algorithm API assumes very few properties for such algorithms. A method has to be specialized to train the algorithm from a samples vector and a set of target labels or values, and another to predict labels or values from a samples vector. Thanks to templating, these methods handle both classification and regression. Two other methods are in charge of saving and loading back the parameters from training. File format for saving is left to the underlying implementation, and the load method is expected to return a success flag. This success flag is used in a factory pattern, designed to be able to seamlessly instantiate the appropriate machine learning algorithm specialization upon file reading. It is therefore not necessary to know which algorithms the trained parameters files refer to. This new set of classes has been embedded into a new OTB application. Its purpose is to train one of the machine learning algorithm from a set of images and GIS file describing training areas, and output the trained parameters file. Another application is in charge of reading back this file and applying the classification algorithm to a given image. With these two tools, it is very easy to train different algorithms against the same dataset, evaluate them with the help of another application which can compute confusion matrix and classification performances measurement so as to choose one or several best algorithm along with their parameters. The resulting classification maps could then be combined into a more robust one using yet another OTB application, using classes majority voting or Dempster-Shafer combination. Our perspectives for using and improving this new API are manyfold. First, we would like to investigate further the use of the regression mode. We also would like to investigate the performances of the new machine learning algorithms for other tasks achievable with OTB, such as object detection for instance. Last, we would like to evolve the API so as to export any confidence or quality indices an algorithm can output regarding its predictions. This would open the way to the implementation of new active learning tools.
Series
Annotations
Transcript
Loading...
it would be quite a relationship with what we still just getting into produced took its about much in building and remote sensing and there will introduce you to a new of the capabilities of the of to book brought by British to open to the game in this field so this is the author and usually Iowa stopped by which the facts and figures and but a decision to do it the other away so stop by the King of much of the for a month and of what is required and the and by winning should use a we we brought to the average 1 that social pigs and those shorter and by the end with facts and figures but so if you want to leave but for the launch she gets the interesting Message case and back to the
UK the US seems to know before stops to mesh learning steps especially as to well as justification we have to make steps we training so that this is something we sold to produce the producer soak and we of Castigations of the training is are used to making some kind of more than from training some and then the justification is applying the smooth to decide because of a specific entities and in remote something we usually justified Dixons so the message from the image of features like we we sold from a generation where within the British but we can also classified segments or cheques and we may also testified that she was in the majors if we were wages and we want to to be a poet so that is to to know about learning about sensing and other be interested to know full now that it is used as a bit libraries and it also adds application begins so something you can use without containing anything from command line off from should be sold off in my took everything we showed that I've done it with the complaint says that is not a single specifically to to and so they are really just introduce you to work that they do should chain at the end of the show you how we wrote a book and and it will so requirements for
machine on the road for a sense of justly winning the mesh long ago isn't so that stops at being until the and 3 in the 8th in New uses used as the and especially implementation and it is where the used in remote sensing reasons that it that if you parameters has produced said it is good in addition to space and we can start in the programmes which began entry to the from the attack again guarantees and the extension to the world and we saw will be so you the domain arisen is to get the public to get extended quite easy to make fun of the fact the and algorithm tried to to find the best the based separation between the the said that some of the best in those said and and but then we have
to to do other things like numbers of using it to be made for instance here and a show of the year and the number of young age and it was the kind time in the Blue Room at night and in fact it is 1 of the future and says zooming against seas and the and the and the on loan and is now a disused beating menus open 6 and the and the distance between 200 and 600 so if you in the present the because they should go is that if we use something for instance based on a clear day and distance then this each of the 3 to dominate the world can you can have from London to avoid this weekend try to bring for the future and to save the same Branch and its or so and then system America stability if this French is something between 0 and 1 also think it is going well and this is the number of vision of the government that you can do just being on the next retching but it is not greatly resilient to play as you can also do the stretching back to the the run or you can go so you with a sense of reduced number position so a and the and other thing is that you have to use the settlement from Fault trading statement that patients are being put to do we have and his team mates that the from from your in Britain and you will find and then update consistently the dynamization throughout the depressed
and then there is the trains that so it is undivided Trinity which they application into a suite of 18 and that things that they they do it because in the end to train images as white and the so that we know are of application training data reuse reuse yesterday defined so we we would pay for the guns in the United team of the training area with the under tributes classes review but you can see his point to the station and what we do from the players is that we some but the ways in which you take that has we may require meant that training some that the word of the year some position within the native really not only as the risks are rather something of these of these cases for for each would so though this is something you only need a teenage and Europe should find with the the training and the take out of the side is comes from the UK
and then becomes the system as German we have a vision of a computer and using tricky applications and we have to you can answer regulations that of an indication that the you should use a different editions said that the training said because it with which he gets the more and more realistic edition of the pestilences and you can use ISER different but I don't fall for Fault body of of the front row them something of the same party used for the fault for the training and the this is application it so it's accepts the justification mapped out as as and but so you can actually get to make the films of the were and in the issue of was processes of this kind of thing you can do it so it is that's about the case and we are put some way you know measures such as the treats decapitation which takes into it and then accurate and what she the precision and occurred and go for it is about from the Germans and then wants you to change your plans for the estimated the Government's is you may want to apply it to was the images of the euro which is so we of the Solidification to do this you can read back demoted you you estimated from the fine art and the justification for the strange but he does so you can processes arrest as large as you want it will just put the West of which the not trusted interesting and so to the fungus edition in and the 2 straight away so that might be a few get against time and the time of the really interesting thing and and we have some was persisting with of regular risation so I'm and a producer of the movie about a month before the regularization the dinner at the board and that the group is the the same throughout move to try to remove isolated just before the classes different from the 1 they both because they are likely to be no easy so Aravis in the approach which is majority voting in the given the boots and the and the justification of the decision vision and Laura organisation picnic may of later and the sort of vision of the sky as application imagine you train different his players from the difference in age is from future said that you have several classification maps and you want to take the best of each 1 of them and the application with an chimpanzees ISA by among the majority voting this time on the case by a because they are you can also use something that is made NEMO clever is the stench of the communication from computer mattresses this would take into account that some classified has already inaccurate discussed and the river accurate for this also agreed take this kind of information so now meeting
opens see so this is really great that we have a lot of things to think about but it is not as accurate as it is a shame ironic to the run for the thing is in the middle of and the and so on but the during this paper truncheons boosted 150 space is a superb could can generate interest in the area and that the case this is not what this is a joke that really really got some some of these kids backs so
athletics right to do with 2 to during was 1st the find minimum meshes with just model that will be able to and the and the mesh along and and we came with his so of looks is that is something that is a breach of the above and put some of that and we can set gets training and its time and we have had some sympatry goes to try to predict the singers and the and also some of the nation to save the trend where to find it looked the front of the class to make a single arranged they are your that says it has insisted that the need to be of the payment for each this so once we find generate there is that it is what we did with the disease and extracts from the decision of the 2 books of serious are various International in end there is a need and we at the weekend but was classes representing on a sugar and England is that can be found in the UK and so you're a singular yank with a sense and you can reach a new deal and the reason so you about like going to the gym the and reality and the and both from early this year and from open city denied thing is that the 1 from open includes Temperamentals musicians Kate we nearest Naples followed that uses the same because is they are decision trees training the other day at the Rose which is also sometimes and committee of decision tree is training and tourism we have great and which was the trees log was to take the occupation of the talks which we ask to of all to wrap around addition code to defend the case and we have no money but has indicated that there and we are and forest became so
let's is a nice when we have got discussed via under a single she is that we can easily trained were set of them and compiled and the city's the reasons we have victory in which he beat testing the case the test on the tourism against later that scale which is datasets available from London is in website so clear up the confusion metrics matches is so it's with the the G perinatal so don't try to say OK a white light is a forest of owning the everywhere and show you you choose this long which was the latest in a run of 4 at the Knot victory it's just a different volatility just to show you that you can the simply run on the accurate and and get on and get the performances of the trauma of the some of this
is to say that now we can really do we can answer to the 1st question we can be as good design and the sugar weekend at least once a peacefully because as a test of machine learning that and it is right that because we know we have to maintain the and is so long and it would be quite a
Case and and then we have something because of mesh learning with factory said that during the training steps you a you actually camp of writes about the model you you trained as a fighter and this is left to the underlying implementation of the law is so when we are only at the end it was quite simply because we are the world as him with so when we ought to read back the that new that it was as all or nothing but no time we've rectified and and we don't know what kind of great which rent and we want to read it back and the other 9 possible with and so and the user should not have to know which which would be the 2 load throughout to serve preliminary and the and the factory but which were dimension that trading inviting witnesses and by the end of the underlying implementation of the EU is so we of the reading tension returns to open success and machine learning with a victory between just try to instant estate which each of the last and between returns the 1st that the successful year due the with so the sway the user and 1 needs to know which of the 2 Define actually so now go briefly to some exemplars
yesterday was decisions code which I can do which opens treatment and that he could actually try something to to to illustrate the stock to buy 45 data that these are the things you do to to make the time series which where a quiet during but for the life so they try to get the same rate is eating time as well as the future of something into to which so these days that the best way to bring to provide for us to be a persisting and but you and I put it on in despotic example of a focus on for instance getting what over the last and best resolution was to find a way address into some programme because it had no training training area of trains some and I'm not very good at the foot of the potential and the next thing what would and and use of the of the things that's fine you about was them so that by the tried to simply used to this and that it has to be a vision for the is still dear to cast is 1 of these made of what tournaments from Princeton opens but my that only about a reserve or what the and through is made of anything at the idea of try to was sometimes some of some of the deals type of classes confided in the end the from this Iago equate large should find from for my array of interest and a selected the number of 100 1 thousand and 500 for each passes for training and the 3 thousand for each of education so there is a
because it looked like the you have to accusations that it so isolated Street the
made use less interesting here we see the real thing the pain of the loss of some shadows and the and the UK
where the motion of cake and here we are still Wiltshire the Welsh shadows and the need to comply information between the application tried to go very has between the IRA you to some of need against the public to results should be sold out there is the training
set for from Poisson's in the need was cast and there is the
vegetation said I'd just need to mention that his body is expected from the same wood double must from from its the case and here are the
results for the full classified idea that she was and when at the end of the year 1 of the most current time steps from the state where the world for the better job in fact about 3 J package to get the 1st is the 1 estimate injured straight training set training so that those who don't is estimated from the renovations the and the 3rd is estimated from the British set but including the see looking so including the is the the increase is the key part of the equation but in the cases because it's reducing the assessing the programme about what you can see also is that there is a slight increase the cap and kept 1 of these for that because the sea is the multi does rather than it did in seems to be modish agreed that the general public can so anyway isolated
isolated but this remaining classes
classifications so want from which dates so this is the end of the year
for the 1st date this is
qualifies for visiting the and the
seeds of this is that as ball for the last day of care and then the
idea of during devoting communication of these and tourism organisation with a reduced 1 and there is the so I get a job of work about 1 to 2 hours of physical for England to fight and this
is the kind of dish so I'd everything and in the UK is just amounting to education business that is not in this case though this is just to give an example of what can be done just the results
and the and the app Irit like to bring the British described very but as something that is coming next with the at data which other I'd resolution data in 0 points out 4 of the 5 metre resolution is the river to the city's changed because because of that it originated today so instead we can do Object-based justification which means we are justified the the region's segmented for from segmentation of the case and is can capture more information like statistics all shape all of it enabled us to get that kind of thing about the application to the from and the exact matchday segmentation very large image so it shouldn't to to of this England facial as a should find so you enter the majority of distinguish was issued by the best or most no size limitations because they images of the produced and we have the Victoria justification application which is basically the same as what officials are shown here but for the Little data who because they put with down to get with the same tourism and the same the same Walker look for a kick in the case that said so it was an
example of this is 1 1 9 of the innings with a reviewed and this is the segmentation we get the situation by over later when the page and it shows 100 and 71 thousand polygons and reclassified Xaus but he goes into a bar in a boom in the tank for the year is just
so this is quite from using and we come in the next 3 days may be and a flight to Boston and its so
tax and figure and the last time but is but this is everything you need to know that I'd just
jumped to this year as a result is still best thought is with websites and if you want to close for the year we ball in the air and
just to take a message to know we are the best them and order messianically under development simulating extensive more for we provides are decide task and providing took place that the reason to upset the teenagers and you can read this tourist industry is the saviour of the application to which in turn can be removed from 1 to the other 2 are changes in the way to talk about the case but we see a talk that and coming next to the direct daughter of the rest of the city dishing to and tourist office implementation of the road rapturous thank you so I
for you the stretches of the stops on the way request to provide for more details questions perhaps you could have heard a lot of changes in Boston
Machine learning
Computer font
Message passing
Computer animation
Personal digital assistant
Average
Decision theory
Authorization
Figurate number
Game theory
Message passing
Field (computer science)
Domain name
Spacetime
Implementation
Algorithm
Support vector machine
Computer-generated imagery
Virtual machine
Parameter (computer programming)
Wave packet
Data model
Programmer (hardware)
Chain
Latent heat
Object (grammar)
Kernel (computing)
Hausdorff dimension
Extension (kinesiology)
Pixel
Message passing
Library (computing)
Multiplication
Machine learning
Addition
Algorithm
Electric generator
Spacetime
Polygon mesh
Decision theory
Patch (Unix)
Open source
Parameter (computer programming)
Bit
Cartesian coordinate system
Template (C++)
Open set
Virtual machine
Message passing
Computer animation
Estimation
Social class
Remote procedure call
Library (computing)
Separation axiom
Extension (kinesiology)
Point (geometry)
Polygon
Suite (music)
Random number
Beat (acoustics)
Multiplication sign
Computer-generated imagery
Dynamical system
Branch (computer science)
Distance
Hand fan
Number
Wave packet
Workstation
Database normalization
Numerical analysis
Vector space
Reduction of order
Normal (geometry)
Message passing
Pixel
Position operator
Linear map
Physical system
Stability theory
Social class
Area
Machine vision
Computer file
Range (statistics)
Instance (computer science)
Cartesian coordinate system
Maxima and minima
Word
Sample (statistics)
Computer animation
Raster graphics
Statement (computer science)
Stability theory
Decision theory
Multiplication sign
Cohen's kappa
Matrix (mathematics)
Data model
Computer network
Process (computing)
Noise
Fuzzy logic
Pixel
Sampling (music)
Process (computing)
Spacetime
Mapping
Computer file
Price index
Thread (computing)
Virtual machine
Flow separation
Forest
Sample (statistics)
Raster graphics
Telecommunication
Self-organization
Quicksort
Row (database)
Random number
Digital filter
Neighbourhood (graph theory)
Support vector machine
Algorithm
Computer-generated imagery
Infinity
Regular graph
Wave packet
Measurement
Chain
Social class
Voting
Message passing
Subtraction
Booting
Information
Computational number theory
Machine vision
Planning
Cartesian coordinate system
Local Group
Voting
Computer animation
Personal digital assistant
Social class
Bayesian network
Matrix (mathematics)
Freezing
Machine vision
Linear regression
Structural load
Model theory
Gradient
Multiplication sign
Decision theory
Scientific modelling
Machine code
Data model
Thomas Bayes
Prediction
Computer network
Forest
Moving average
Website
Library (computing)
Social class
Machine learning
Metropolitan area network
Decision theory
Numbering scheme
Computer file
Parameter (computer programming)
Metric tensor
Forest
Sample (statistics)
Network topology
Website
Random number
MUD
Support vector machine
Set (mathematics)
Artificial neural network
Decision tree learning
Open set
Twitter
Wave packet
Social class
Knot
Software testing
output
Normal (geometry)
Message passing
Default (computer science)
Rule of inference
Scale (map)
Addition
Polygon mesh
Scaling (geometry)
Classical physics
Machine code
Template (C++)
Single-precision floating-point format
Computer animation
Personal digital assistant
Function (mathematics)
Social class
Bayesian network
Mathematical optimization
Matrix (mathematics)
Reading (process)
Implementation
Inheritance (object-oriented programming)
Algorithm
Support vector machine
Model theory
Multiplication sign
Scientific modelling
Virtual machine
Infinity
Machine code
Open set
Wave packet
Data model
Pointer (computer programming)
Machine learning
Computer network
Data mining
Factory (trading post)
Pattern language
Software testing
Fuzzy logic
Message passing
Library (computing)
Polygon mesh
Structural load
Computer file
Physical law
Virtual machine
Forest
Computer animation
Factory (trading post)
Hausdorff dimension
Software testing
Social class
Bayesian network
Reading (process)
Polygon
Product (category theory)
Random number
Decision theory
Multiplication sign
Tournament (medieval)
Time series
Grass (card game)
Hand fan
Wave packet
Number
Image resolution
Programmer (hardware)
Casting (performing arts)
Video game
Bit rate
Data mining
Process (computing)
Website
Message passing
Address space
Social class
Area
Focus (optics)
Machine vision
Water vapor
Linear programming
Instance (computer science)
Machine code
Video game
Maxima and minima
Forest
Message passing
Computer animation
Equation
Social class
Series (mathematics)
Computer animation
Information
Linear programming
Insertion loss
Acoustic shadow
Website
Cartesian coordinate system
Message passing
Resultant
Wave packet
Ocean current
Random number
Multiplication
Algorithm
State of matter
Set (mathematics)
Multiplication sign
Ext functor
Set (mathematics)
Mereology
Table (information)
Wave packet
Estimator
Programmer (hardware)
Process (computing)
Sample (statistics)
Computer animation
Spherical cap
Personal digital assistant
Message passing
Resultant
Metropolitan area network
Random number
Sample (statistics)
Computer animation
Algorithm
Set (mathematics)
Mathematical singularity
Message passing
Table (information)
Supremum
Social class
Random number
Support vector machine
Algorithm
Physicalism
Forest
Voting
Process (computing)
Radial basis function
Computer animation
Personal digital assistant
Radius
Telecommunication
Self-organization
Message passing
Resultant
Metre
Point (geometry)
Web page
Polygon
Statistics
Mobile app
Support vector machine
Computer-generated imagery
Boom (sailing)
Shape (magazine)
Image resolution
MKS system of units
Object (grammar)
Vector space
Statistics
Information
output
Pixel
Message passing
Linear map
Scale (map)
Dialect
Information
Computer file
Polygon
Attribute grammar
Limit (category theory)
Cartesian coordinate system
Shape (magazine)
Computer animation
Raster graphics
Personal digital assistant
Radiometry
Bildsegmentierung
Identical particles
Frame problem
Support vector machine
Software developer
Multiplication sign
Computer-generated imagery
Electronic mailing list
Interface (computing)
Machine code
Emulation
Blog
Software
Statistics
Process (computing)
output
Message passing
Library (computing)
Linear map
Computer font
Software bug
Dataflow
Online help
Image processing
Electronic program guide
Open source
Computer program
Coma Berenices
Wiki
Computer animation
Radiometry
Computing platform
Bildsegmentierung
Figurate number
Resultant
Machine learning
Implementation
Satellite
Algorithm
Real number
Software developer
Computer-generated imagery
Cartesian coordinate system
Message passing
Mathematics
Computer animation
Raster graphics
Personal digital assistant
Order (biology)
Software framework
Moving average
Office suite
Message passing
Task (computing)
Loading...
Feedback

Timings

  510 ms - page object

Version

AV-Portal 3.8.2 (0bb840d79881f4e1b2f2d6f66c37060441d4bb2e)