Macaulay Land Use Research Institute, L3-Storm, Redleaf Systems 2000

Landsat TM imagery enhanced by Macaulay

Product: knowledge-controlled spatial processing

Benefits: fast deployment of domain knowledge within spatial processing operations

Example applications: assessment of forest stock damage; land cover change; analysis of reconnaissance data

Macaulay Land Use Research Institute, L3-Storm, Redleaf Systems 2000

Landsat TM imagery enhanced by Macaulay

Macaulay Land Use Research Institute, L3-Storm, Redleaf Systems 2000

Landsat TM imagery enhanced by Macaulay

Satellite imagery and aerial photography are valuable sources of remotely-sensed data. Mature GIS and geographical imaging packages (GIP) have evolved to facilitate the interactive processing of this data, but automating these tasks is difficult: interpreting natural and man-made objects requires more than just data about their appearance. Knowledge of their structure, spatial relationships, expected context, and dynamics are all examples of where domain knowledge facilitates accuracy. This knowledge cannot be readily encoded within GIS/GIP and must therefore be manually applied by the photointerpreter or GIS/GIP operator.

ETORA-II facilitates faster processing times and increased product quality by providing an environment for representing, and automatically reasoning with, your domain knowledge. It works alongside your GIS/GIP software and allows the full functionality of those systems to be used within a rich, knowledge-based environment. Applications constructed using ETORA-II are:

  • Flexible, accommodating the wide range of knowledge and data that could positively contribute to a product;
  • Extensible, allowing the breadth (range of tasks) and depth (accuracy of solutions) of the system to be incrementally developed when new data, knowledge or software resources become available; and
  • Adaptable, allowing complexities such as uncertainty within the system's knowledge, conflicts between possible outcomes, partial and missing knowledge, adverse data quality, data corruption and varying geographical coverage between the available datasets to be accommodated.

Features

  • Reason with both quantitative and qualitative domain knowledge

Encode domain knowledge within modular “agent-like” expert objects;

Represent knowledge using rules, semantic networks, frames, neural networks, object relations, procedures, and others;

Experts can reason with any combination of datasets stored within your GIS/GIP system;

Experts can invoke any GIS/GIP function.

  • Connect to industry standard GIS/GIP packages

ESRI's Arc/INFO and ArcView;

VNI's PV-WAVE;

Any other package or legacy system which publishes an API

  • Reason about uncertainties in the domain

Endorse objects with symbolic uncertainty;

Reason with symbolic uncertainties to derive belief;

Extend the uncertainty scheme to support other representations, such as fuzzy logic and Dempster-Shafer theories.

  • Dynamically form an analysis methodology

Experts can organise their own activity;

Avoid inflexible, hard-coded analysis strategies;

Employ reasoning and processing that that best suits each sub-problem;

Robust to variations in data and knowledge completeness, quality, suitability and uncertainty.

  • Generate explanations of all reasoning and processing performed

Make the system's knowledge explicit;

Generate a reasoned argument for the processing performed;

Associate explanations with the resultant data product;

English-like statements.

The above features combine to permit emphasis upon the nature of each task and how a solution may best be achieved with the available data, knowledge and software resources. This has been dubbed task-orientation, ETORA-II being an Environment for Task-Orientated Analysis.

Figure 1 shows a forestry area in 1988, and Figure 2 the corresponding area in 1994. Figure 3 shows the result of an ETORA-II application to detect changes and the explanations that are produced. Here, agents have utilised various kinds of knowledge and data and have contributed evidence for and against the likelihood of each coloured area being an instance of forest felling. Adequate support was not produced for the blue areas: they do not appear within the resultant product.

Technical Specification

  • COTS product environment

Gensym's G2;

ESRI's Arc/INFO (optional);

ESRI’s ArcView (optional);

VNI's PV-WAVE (optional).

  • FRACE

G2 layered application;

Deploys experts within a blackboard problem solving model;

Sequential and concurrent expert activity;

Multi-threaded expert activation.

  • HYPENDS

G2 layered application;

Implements Endorsement Theory for uncertainty reasoning;

Easily extended to support other uncertainty schemes.

  • G2ARC

G2-Arc/INFO and G2-ArcView bridge;

Provides an Arc/INFO and ArcView client for G2, and a G2 client for ArcView.

  • G2WAVE

G2 - PV-WAVE bridge;

Provides a PV-WAVE client for G2

  • Network environment

TCP/IP

  • Platforms

UNIX (OSF1, Solaris);

Contact

Chris Skelsey,

Alistair Law,

Mayank Patel,

Macaulay Land Use Research Institute, L3-Storm, Redleaf Systems 2000

Landsat TM imagery enhanced by Macaulay

Macaulay Land Use Research Institute, L3-Storm, Redleaf Systems 2000

Landsat TM imagery enhanced by Macaulay