SPA Project

Project Title   Support by Pre-classification to Specific Applications
Project Acronym   SPA
Contractor(s)   MEEO (Italy)


Project    Context
   How it will work



In the framework of Image Information Mining techniques applied to satellite imagery, many approaches are being analyzed and implemented in order extract the maximum level of information from multispectral images and from the availability of large databases of different sensors at different time.

The four keywords for modern image processing systems are:

  Image Automatic (fast and objective approaches)
  Image Multitemporal (analysis of large image databases with multitemporal techniques)
  Image Multi-sensor (to join together advantages coming from different satellite platforms)
  Image Distributed architectures (making use of communication infrastructure and leave processing system at the provider's premises)

The Support by Pre-classification to Specific Applications (SPA) project aims at simultaneously addressing all these themes, making use of fully objective preliminary image processing tools (like the MEEO SOIL MAPPER®), implementing new processing systems and providing support to other countries on using ESA Environments.

The SPA Activities are organized on three main sub-projects, each one with a specific aim:

  Image AVNIR-2 Cloud Detection System (AV2CLS): implementation of a full automatic image processing chain for cloud detection on very high resolution images;
  Image Support to PECS projects, to provide support to entering ESA countries on the use of ESA Environments based on distributed architecture (SSE and KEO);
  Image (A)ATSR Land Classification system (ALCS): implementation of a fully automatic land cover change / land use analysis system based on a large database of (A)ATSR data scalable to other satellite sensors.



The objectives of the SPA project can be summarized as follows:

AV2CLS Sub-project: implementation of a fully automatic image processing chain for cloud detection
AV2CLS main objective is the implementation of an innovative cloud detection system for Visible-to-Near Infrared (VIS-NIR) sensors and the integration of the developed software into KEO.

Main innovation of the SPA-AV2CLS project resides on the application of successive steps in order to allow, at the end of the process, detecting clouds using VIS-NIR satellite images with high spatial resolution (ALOS-AVNIR-2 data with 10 m resolution).
The system will be integrated in two operative modes:

  Image KEO Environment:
- As a series of KEO CLI modules for AVNIR-2 radiometric calibration, spectral pre -classification with SOIL MAPPER® and cloud detection;
- As a KEO FEP module that chains each CLI module for the processing of user-provided AVNIR-2 Level 1B images in CEOS format;
  Image As a real time image processing module to systematically generate cloud cover information from Level 0 data (to be confirmed).

PECS Support Sub-project: Support to the FLOREO and Safe City GIS PECS Project
In the framework of the SPA Project, specific activities have been proposed to support two PECS Projects:

  Image FLOod Risk Earth Observation monitoring (FLOREO), Czech Republic;
  Image Safe City GIS, Poland.

The PECS project support is focused on:

  Image providing the FLOREO and Safe City GIS Project teams with training and support on the use of the ESA Service Support Environment (SSE) and the Knowledge-centre Earth Observation (KEO) system;
  Image allowing the FLOREO and Safe City GIS systems using the SOIL MAPPER® and PM MAPPER® MEEO products free of charge for the duration o the project.

Moreover, specific activities will be defined with the two teams in case common interests are identified.

ALCS Sub-project: Automatic (A)ATSR Land Classification System
In the framework of the SPA Project, a further advanced system for automatic information extraction from a long-term satellite image database will be implemented.

ALCS will allow the end user accessing in real time to a 12-year database of (A)ATSR data with global coverage to identify specific land use types (e.g. agricultural fields, urban areas, …) as well as defining new features to identify specific phenomena of which the user knows the evolution (in terms of duration and associated land cover classes evolution).

The innovation introduced by ALCS resides on:

  Image the type of system to be integrated (automatic land cover change/ land use maps extraction);
  Image The flexibility of the system to integrate different missions with different spatial resolutions;
  Image The implementation architecture, that allows real time operations over a large image database.

The final system will be implemented with two main interfaces:

  Image Application interface;
  Image Expert user interface.



AV2CLS architecture
AV2CLS will be implemented though a fully chainable architecture made of single processing tools that are provided separately or chained and wrapped by a KEO FEP module.

This approach was required by the spectral characteristics of the AVNIR-2 sensor: since AVNIR-2 only allows imaging the earth from visible to near infrared frequencies, clouds cannot be univocally distinguished from snow and bright areas only by spectral information.

Then the selected approach (spectral pre-classification and second level stratified processing) was selected and is being implemented.


The provided processing modules are:

  Image Input data interface (for ALOS-AVNIR-2 Level 0 or Level 1 data);
  Image Radiometric calibration module (to convert input data to Top of Atmosphere Reflectance maps);
  Image Spectral pre-classification module (to classify the calibrated images by means of the SOIL MAPPER® software);
  Image AVNIR-2 cloud detection system, made of two sub-modules:
- The pre-classification map segmentation module (to identify only those segments from the classification map that are candidate to be cloudy areas);
- Rule-based could detection system to evaluate, for each segment, whether or not it represents a cloud.

Output cloud masks are provided in two formats:

  Image A GeoTIFF raster image with same dimension and projection of the original image (binary mask);
  Image A vector file in shape format.

Support to PECS projects FLOREO and Safe City GIS
The architecture of the support to the FLOREO and Safe City GIS project is related to the architecture of each PECS project system. Both systems will be realized with a distributed architecture.

Details on the systems architecture, integration of each project system with SSE and MEEO products will be detailed during the project.

ALCS architecture
The ALCS system architecture is based on the existing database implemented during the CARD project that contains the 12-year of classified images processed with SOIL MAPPER®.


In order to exploit at the maximum extent the semantic meaning of the CARD database, multitemporal information mining techniques must be implemented and applied. Since it is not possible to apply a multitemporal pixel-based analysis technique on data contained on the CARD database, a new system that allows remapping all CARD data onto a common grid and applying to the resulting database multitemporal techniques is implemented on the ALCS sub-project.

ALCS architecture is made of the following modules:

  1. Remapping module
    The remapping module aims at remapping the CARD images onto an earth fixed grid, to allow the application of pixel-based multitemporal analysis. The selected earth-fixed grid is made of tiles of ¼ of degree by ¼ of degree; each tile is made of 64x64 pixels; each pixel has a spatial resolution of about 450m. The remapping module generates, for the entire 12-year database, about 300 million tiles made of classified pixels with semantic meaning that will be managed by the tiles database manager (see next paragraph).


  2. Tiles database manager
    The tiles database manager (DBM) scope is to manage the 300 millions tiles to allow fast queries based on space, time and semantic. Almost real time (less than two seconds) three-dimensional queries can be realized through the optimization of the DBM structure and the use of specific hardware platforms.
  3. Multitemporal rule definition module
    The availability of 12 years of classification maps worldwide distributed with a revisiting time of days permits to develop and apply fully automatic multi temporal applications to detect and extract specific phenomena. A series of evolution models can be defined on a pixel basis, modeling the time evolution of the pixel in the feature (classes) domain.

    The multitemporal rule definition module permits to define specific evolution models based on the behavior of land cover classes with respect to their temporal evolution.

    Two modalities are foreseen to define multitemporal evolution models:

      Image by writing a text file based on a defined schema that contains all needed information for the model definition;
      Image by means of an advanced interface that allows loading a series of tiles and allows defining the new model though a graphic tool.
  4. Multitemporal rule application module
    The role of the multitemporal rule application module is to apply the defined evolution models over a set of tiles resulting from a two or three dimensional query performed by the DBM.
    The Multitemporal rule application module is the core of the ALCS system since it allows joining the output of the ancillary sub-systems (remapping, DBM, Ruleset Definition) extracting the result of the elaboration.
  5. System Interfaces
    Two system interfaces are defined, dedicated to two different type of users:

    Application User Interface: permits to a standard user to access to the system and apply existing evolution modes to the entire database.

    Expert User interface: besides the Application user Interface functionalities, allows defining and storing new evolution models through:

      Image a text interface, that allows the user to compile a xml file containing the designed evolution model based on a pre-defined template
      Image an advanced graphic tool for the definition of new evolution models that allows connecting to the Tiles BD, downloading a defined time series of Tiles, designing a new evolution model and recoding it into the ruleset DB.



How it will work

AV2CLS will be integrated as:

  Image KEO CLI modules to allow using the single processing modules for:
- AVNIR-2 Level 1 radiometric calibration;
- Classification with SOIL MAPPER® of a radiometrically calibrated AVNIR-2 image;
- Cloud detection from a spectrally pre-classified AVNIR-2 image.
  Image KEO FEP module to allow extracting from a AVNIR-2 Level 1 image the cloud mask; the FEP module chains the single CLI modules to implement the full process.
  Image Automatic image processing chain module for the systematic processing of AVNIR-2 Level 0 scenes.

Support to PECS Projects
The support to the FLOREO and Safe City GIS PECS projects will allow:

  Image Integrating SSE services coming from the PECS Projects;
  Image Implementing processing tools based on customised systems (i.e. SOIL MAPPER®-based processing modules on the FLOREO system).

The ALCS system will be integrated on customised hardware and software system. Technical requirements will be defined during the project execution.


Contributors to this page: andreadv .

Page last modified on Wednesday 22 of December 2010 15:00:22 CET by andreadv.