KES-B Project
| Project Title | Knowledge Enabled Services | |
| Project Acronym | KES-B | |
| Contractor(s) | GTD, UNINOVA, Starlab |
|
Context |
| Objectives | |
| Architecture | |
| How it Works | |
| Output |
| Context | 
|
The EO data is currently retrieved via attributes with no link to information content. Experts then interpret EO data (images in particular, provided in heterogeneous formats and analysed by different tools) to extract the needed information, through human intensive, complex and hence expensive processes. Finally, the information extracted from EO data must still be combined with information from other domains in order to support users (end-users, service providers, decision makers, ...).
In this context, EO data exploitation is currently limited or even made impossible when large data sets (tera / peta bytes archives) have to be systematically analysed (e.g.: for aspect of human, social, environmental, … relevance). The situation is worsening with the new missions (even constellations), which bring also broader sensor varieties and higher data rates.
Another aspect to be considered is the difference in terminology and culture related to the user interst domains, when compared to the data descriptions provided by EO terminology or the experts in one specific domain. This could prevent the reuse of ouput information by users of other domains, because the image content (its meaning) and the semantic terms to represent it, are many times subjective, depending on user interest and culture.
In order to find solutions to the above problems, it is necessary to:
|
Identify and implement unattended or supervised feature extraction algorithms | |
|
|
Define and implement an environment for easy, scheduled and controlled exploitation of resources (e.g.: data, algorithms, procedures, storage, processors, ...), for example to automate the generation of products | |
|
|
Support users in easily identifying and accessing required information or products by using their own vocabulary, domain knowledge and preferences |
| Objectives |
|
The aim of this project is to find solutions to the above last two items and to implement a prototype to demonstrate them. The demostration will by necessity include as many as possible different feature extraction algorithms, compatibly with project size.
In essence, the project intends to address the issue of bridging the gap between EO products and users, who need them but require assistance in translating their domain problems into information queries compatible with the EO domain descriptions. This issue has been the subject of a number of initiatives and this project explores the possible contribution of knowledge-based technologies.
More in detail, this project addresses:
|
|
Workflow environments | |
|
|
Distributed data and processing environments | |
|
|
Semantic classification and retrieval methods | |
|
|
Machine learning | |
|
|
Knowledge acquisition and sharing within user communities (semantic interactions) | |
|
|
System personalisation | |
|
|
Semantic optimisation of search engines | |
|
|
Processing engines |
The objective of the project is to devlop the prototype of a system aimed at enlarging the use of EO data by improving the data exploration capabilities. It could be installed, configured and used in different contexts, ranging from large archive and processing sites to specialised service providers' environments.
| Architecture |
|
In terms of architecture, KES-B is a web-enabled, distributed, (semi)automatic, user-centred system, which permits to search multi-domain data and information. It supports replicas hosting specific portions of data or information for one or more domains or providing specialised functionalities.
Figure 1: Architectural overview by layers
The above figure summarises and provides an overview of the architecture in 5 layers. The top layer is composed by the client (or collection of clients) to supply the required functionalities. The second layer is the WEB server, providing web-enabled capabilities for the deployed functionalities. The central layer provides the core capabilities of the system, particularly the control structure and the aggregation of tasksolvers. The fourth layer includes the repository servers, while the fifth one is the "blackboard" (or aggregation of blackboards) where different data types are really stored.
Furthermore, figure 1 illustrates three different roles interacting with the system:
|
|
User | |
|
|
Expert | |
|
|
Administrator |
Regarding the implementation, COTS will be used as much as possible and complemented by specific software development to fulfil all the complex features requested to the system.
The KES-B server prototype may be instantiated in many different ways, for example as ‘repository’, ‘quality control’, ‘generic facility’ and ‘service provider’. The following drawing illustrates the referred instances.
Figure 2: Instances of the KES-B prototype
|
|
Repository | |
| Locates and feeds data from archives and Internet. Generates products from data and archives them. | ||
|
|
Quality Control | |
| Assess and reports on quality of products and instruments using dedicated algorithms applied systematically or on sample basis. The data can be obtained from archives, internet, local storage or dedicated telecommunication links. | ||
|
|
Generic Facility | |
| Represents a complex operational environment (e.g.: charged with various types of activities, which acn be mission, data, time, -dependent), where KES-B could permit easier management and maintenanceof the environment, where components need to be linked within procedures, and procedures need to be executed under. Full “knowledge??? of components, procedures and of their interfaces easies these tasks, minimising also the possibility of errors. The data could be obtained from archives, Internet or locally. The output could be published on Internet. | ||
|
|
Service Provider | |
| In this environment, KES-B could allow to define and automate the extraction of information, the generation of specific products, the provision of services. The KES-B prototype will be a Service Provider instance, hosting, as test services: - Winds & Waves - Algae Bloom - Oil Spill - Ship detection |
| How it Works |
|
The three different user types can interact with the system as follows.
The 'user' will be capable to
|
|
Query and browse products: | |
| - Products will be organised within a catalogue according to relevant attributes | ||
| - The user will be able to access the catalogue using spatio-temporal queries and his own vocabulary (starting from system definition of the domain up to his personal definitions) | ||
| - In order to better support the search, the system uses in addition spatial or temporal relations between products and the cumulated interest shown by the user in previous queries | ||
| - In absence of input the system even provides suggestions based on accumulated knowldge | ||
|
|
Use storage space in the server in private regions (temporary and permanent) | |
|
|
Transfer results to user's side |
Figure 3: Snap shot of the user HMI mock-up
The 'expert' has access to the above 'user' functionalities and will in addition be capable to
|
|
Make 'definitions': | |
| - Define and/or register algorithms, transformations, … | ||
| - Define data (data sources) | ||
| - Define products | ||
| - Define workflows, connecting data to algorithms in order to create products | ||
|
|
Schedule and condition the execution of defined workflows | |
|
|
Make definitions available to other experts |
The ability of experts to create EO products on the basis of EO data or other already existing EO products is implemented in terms of workflows. Workflows cascade connect data to encapsulated algorithms in order to create products. Workflows are intuitively expressed graphically. A graphical workflow editor will integrate the HMI of the expert.
The 'administrator' has the following capabilities
|
|
System administration (User accounts, server addresses, storage space, …) | |
|
|
System monitoring (workload status, storage space availability, errors, logs, error recovery, …) | |
|
|
Arbitration and content management (prepare system workflow; manage, arbitrate and fine tune experts' contributions; ...) |
| Output |
|
The result of this project will be a prototype of a conceptual breakthrough aimed at:
|
|
Easy and automate information extraction by supporting ways to describe and effectively apply the abilities of experts to the systematic extraction of information (at large archives centres and small companies with limited but specialised data sets): - Experts encapsulate exploitation algorithms, strategies, criteria, … these are the basis for domain 'definitions'. - Experts create and run data and information flows (generally understood as 'workflows') connecting archives and algorithms, thus automatically generating information extraction and classification. |
|
|
|
Easy exploitation by users of EO data, products, and information by helping them in searching, identifying and accessing both domain- and user-dependent value-added information: - The user understanding of the domain is coded in the system on the basis of a general understanding and subsequent, interactive refinements - user semantic. - The system applies coded user's understanding of the domain in order to improve user's exploration and (complex) search of the available (or potentially available) multi-domain data and information. |
The applications used to test the system include:
|
|
Winds and waves (mainly speed and direction) | |
|
|
Oil spill detection | |
|
|
Ship detection | |
|
|
ALgae bloom detection |
The Final Presentation was held at ESRIN on April 6, 2005.
|
|
