Skip to content

GitLab

Last edited by Victor Poughon
Page history

OTB Users Day 2018

Thanks to our sponsors!!!

logo_Cnes irstea

General Information

logoIncrust OSGeo_project

Purpose

This page is about organizing an Orfeo ToolBox users meeting in 2018. Previous events:

Dates and Times

The 4th Orfeo ToolBox users’ days, which will be held in Montpellier on the 19th of October 2018, following the Theia conference.

Location

In Agropolis International

Access/Practical information (in French): https://theia2018.sciencesconf.org/resource/acces

Schedule

The morning session will be dedicated to short talks from users and contributors, as well as discussions. The afternoon will be divided into several training sessions for different levels of use. This day is open to everyone, including those that have never used OTB before.

08:30 - 09:00 : Registration desk opens

09:00 - 09:30 : Welcome (Amphitheatre Louis Malassis)

09:30 - 10:30 : Talks Part 1 (Amphitheatre Louis Malassis)

Image segmentation partitions the image into patches that are internally consistent in terms of image properties. This is an important step in those (most common) cases in which landscape units of interest are configured by many non-uniform pixels and thus single-pixel properties are not sufficient to characterize a unit type. Segmentation becomes increasingly important as remotely sensed imagery becomes of increasing spatial resolution. In this presentation, after a brief introduction, I will focus on practical aspects of segmentation with OTB such as the impact of varying the different parameters involved in Mean-Shift based segmentation.

Monitoring vegetation dynamics in nature reserves requires accurate mapping of habitats. In August 2017, we used a drone to acquire LAS images and high-resolution (2cm)orthophotos on islands of nature reserve of St-Mesmin along the Loire river. At the same time, we mapped habitats based on vegetation sampling every 15 m and the phytosociological nomenclature published by the Botanical Conservatory of the Paris Basin. The OTB workflow allowed us to map 9 vegetation types but with an overall kappa of 0.60 only. If the results were satisfactory for the different riparian forest types (min F-score = 0.70) , the various tools tested under OTB failed to correctly classify open areas as reed-bed (F-score = 0.31) and grassland (F-score = 0.18). Two hypotheses can explain this problem: the low accuracy of the field GPS data and some subjectivity in the interpretation of the phytosociological nomenclature.

Included in the Guiana Shield ecosystem which is one of the largest blocks of intact tropical forest worldwide, French Guiana play a critical role in mitigating climate change, preserving biodiversity and regulating water of the Amazon basin. Under low pressure in the past, degradation of this fragile ecosystem is growing, especially driven by gold mining activities or illegal agricultural activities. In this context and for a project provided by the french forest national office (ONF), in order to deliver a near real tile deforestation monitoring process, we develop a set of tools allowing to monitor deforestation using Sentinel-1 data. The main process of the approach are (i) Calibration and Orthorectification over user study area (shape file), (ii) Temporal Adaptive speckle filtering, (iii) Adjust time series level and smooth it, (iv) Deforestation analysis. Deforestation analysis is based on applying a threshold of the differences between a reference date (usually the oldest date) to the current date. If this change is significant we expect a deforestation process and give the current date as deforested label in the deforestation map. This open source tool box is mainly based on Orfeo ToolBox for Sentinel-1 pre-processing and is based on specific python script. All the toolbox is integrated in QGIS and available in a QGIS RemoteSensing toolbox for windows or using the following for Ubuntu users https://gitlab.com/clardeux/FOSS4G-fr-2018-Atelier-Radar.

10:30 - 11:00 : Coffee break

Group photo at 10:30 in front of Agropolis building

11:00 - 12:20 : Talks Part 2 (Amphitheatre Louis Malassis)

The ZOO-Project is an open source software which provides a Web Processing Service (WPS) compliant and developer-friendly framework to easily create and chain Open Geospatial Consortium (OGC) Webservices. It supports both 1.0.0 and 2.0.0 versions of the OGC specification. It can handle execution of WPS Services implemented in various programming languages and can also take advantages of various processing engine such as OrfeoToolBox (OTB). Using this ZOO-Project support gives users the capacity to use OTB applications remotely from any WPS client. The presentation will quickly introduce the specific capabilities of the ZOO-Project such as automatic publication of result using OGC Web Services through MapServer and how this can used when using OTB applications as WPS services. Also the current development status of the High Performance Computing (HPC) support in the ZOO-Project will be presented. With this support, one will be able to not only to execute OTB applications that runs locally to the WPS Server but also invoke OTB applications that will run on HPC server.

For 5 years now, CEREMA's satellite team is engaged in designing a large scale land cover map by supervised classification coupled with specific database processing. In order to follow public policy in territorial planning, these maps over several acquisition dates are compared to detect land change and monitor the landscape. Because of uncertainty of each measurement, the comparison introduces a lot of over-detection. These differences have to be filtered to identify relevant ground change. Since OTB's latest version offers a function to produce confidence map linked to classification, it has been tested to characterize difference areas. This experimentation shows how this map, representing classification algorithm capability, cannot directly be used as a quality layer and must be upgraded to become suitable for thematic users.

Time series of remote sensing images can be used to produce homogeneous and up-to-date landcover maps. We used Orfeo ToolBox to test whether MODIS time series can be classified using a detailed nomenclature alike national forest inventories nomenclatures. To identify best practices for classifying time series, three algorithms were compared: Maximum Likelihood, Support Vector Machine and Random Forest. For each algorithm, we optimized training, temporal compositing and selection of input features. Our results showed a clear improvement in classification accuracy when spectral bands were used instead of vegetation indices alone. Temporal compositing had a major impact when the whole phenological cycle was used for three consecutive years. Random Forest produced the best classification.

The differential SAR interferometry (DInSAR) technique relies on the processing of two SAR images of the same portion of the Earth's surface taken at different time. The aim is to analyze potential events (earthquake, destruction, ...) by highlighting differences between SAR images. DInSAR involves a set of tools such as creation of deformation grids , coregistration or building of interferograms. An Orfeo Toolbox remote module (DiapOTB) contains all necessary steps and allows to launch a complete DInSAR chain. The DiapOTB module were used with Sentinel-1 data with satisfactory results.

GeoStorm is a geospatial platform developed by CS SI since several years to handle various types of data from satellite, GIS database, IoT network or social network and performs analytics on them. It was instantiated on the Earth Observation Innovative Platform Testbed Poland (EO IPT Poland). CS SI has try to integrate in this platform a set of open source processing from Orfeo ToolBox, SNAP Toolbox and Sentinel-2 For Agriculture and Theia expertise centers. We will present in this presentation the feedback of this experience.

12:20 - 13:30 : Lunch

10 minutes walk to the restaurant ("Aboyeur").

map_restau

Offered by sponsors

13:30 - 17:00: Workshops (coffee break 14:45 - 15:00 in the main hall)

We planned to organize 4 sessions in parallel. Important note: please read carefully the instructions for the workshop.

Session 1: Install party and introduction to the ORFEO ToolBox

  • Level: Beginners

  • Workshop Contents: Learn how to install to install and configure OTB on your system. Learn how to access to OTB applications and perform remote sensing image processing at scale through a ergonomic interface. Learn more about Monteverdi and how to access to OTB processing in Quantum GIS.

  • Workshop Requirements: The workshop is composed of a suite of guided fill-in the blank exercises, and does not require any technical background, but basic knowledge in remote sensing will help. Trainees must bring their own laptop.

  • Organizer: Victor Poughon (CNES) and Yannick Tanguy (CNES)

  • Room: Badiane

  • Planning:

    • 13h30-14h : introduction to OTB
    • 14h-14h45 : installation and configuration demonstration of OTB applications with some exercises (steganography)
    • 14h45-15h : coffee break
    • 15h15-16h30 : pre-processing : from optical calibration to orthorectification
    • 16h30-17h : questions, help to configure QGIS, etc.

Session 2: Writing an Orfeo ToolBox processing chain in Python

  • Level: Intermediate

  • Workshop Contents: How to configure your environment to use Python PI for Orfeo ToolBox applications? How to create, set parameters, and run Orfeo ToolBox applications from within Python? How to chain several applications to form an in-memory pipeline? How to connect Orfeo ToolBox applications with NumPy? All those questions (and many more!) will be answered during this workshop. We will develop a processing chain for water surface detection from a Sentinel-2 time series over the area of the Laguna de la Nina in Peru. The concepts that will be learned during the workshop can easily be extended to the full set of Orfeo ToolBox applications, which have a common API. The workshop is therefore a good introduction to start writing similar processing chains.

  • Workshop Requirements: The workshop is composed of a suite of guided fill-in the blank exercises, and does not require a heavy technical background, but basic knowledge in Python and in remote sensing will help. Trainees must bring their own laptop with an up-to-date OTB installation.

  • Organizer: David Youssefi (CNES) and Manuel Grizonnet (CNES)

  • Room: Bambou

Session 3: Deep learning with OTB

  • Level:Intermediate +

  • Workshop Contents: Introduction to a new remote module called otbtf enabling the use of deep learning techniques with real world geospatial data. This remote module uses the high performance numerical computation library TensorFlow to bring the deep learning magic into OTB. How to train and apply deep nets to Remote Sensing images? We will see how to do it on various THEIA images.

  • Workshop Requirements: Basics in image/signal processing is required. Oracle Virtualbox must be installed and working on your computer. We will work on a Ubuntu virtual machine that includes the whole already-installed stuff. Familiar with command line is a prerequisite. A bit of knowledge of Python would be better, but not mandatory. Minimum hardware configuration: 16Gb RAM with at least 4x CPU but 32Gb would be great as the VM will be hosted on your computer. For this session, feel free to come with your own data (images, terrain truth, etc.)!

  • Organizer: Rémi Cresson (IRSTEA) and Julien Michel (CNES)

  • Room: Argane

Session 4: Questions & answers (15:00 - 17:00)

Register at at registration desk to meet with a developer and ask for help, or discuss feature request.

  • Level:

  • Contents: We'll brainstorm on technical topics, with the objective of writing some feature requests to help to build the future of the OTB roadmap.

  • Hot topics (TBD):

  • Organizer: Julien Michel (CNES)

WORKSHOP INSTRUCTIONS

Here are some information you shall read to be ready for our different workshops.

  • You have to bring your own laptop !
  • You have to download a number of files before the workshop, please read very carefully the following details for each session (see below). All documents are available in English or French.
  • We remind you that all sessions start at 13h30

Sessions 1 and 2:

Session 3

Additional instructions for all sessions:

For Session 1 -- Initiation to OTB (Install Party) (room Badiane)

  • You can install QGIS on your laptop and we will show you how to configure it to use OTB
  • We encourage you to install and configure OTB, but we will be there to help you, if needed

For Session 2 -- Python workshop : water surfaces analysis with OTB (room Bambou)

  • You have to install OTB on your laptop
  • You have to check your Python installation and OTB configuration (see the install guide) so you can use OTB in your Python (2.7 or 3.x environment)

For session 3 -- Deep Learning with OTB (room Argane)

  • Don't forget the power supply for your laptop

If you have any problem, you can contact us directly by mail.

Social events

The organization team will meet Thursday at "les frères Berthom" at 18:30 and probably go for dinner in a nearby place. Feel free to write us if you want to join us.

Guidelines

Guidelines for participants

Here are a set of questions that can help you prepare your talk :

  • In what context are you using OTB, what part of OTB and what for ?
  • What other software do you use along with OTB ? Do they have to interact with each others ?
  • What is working and what is not ?
  • What performances do you get (time, quality) ?
  • What is missing ?
  • Did you achieve everything you planned to ?

Guidelines for all speakers

  • Lets make the talks short. Each topic should be presented in 10 min (and 5 minutes questions)
  • Slides and materials should be in English to ease publication. If possible the presentation should be in English.

Guidelines for the tutorial sessions

  • Instructions for the tutorial will be send by email by the organizers

Solutions to workshop exercises