Automating the analysis of remotely-sensed data 1

Automating the analysis of remotely-sensed data 1 © Macaulay Land Use Research Institute, L 3 -Storm, Redleaf Systems, 2000 Landsat TM imagery enhanced by Macaulay

ETORA-II • A toolkit to facilitate the automatic analysis of remotely-sensed imagery – Faster application development through the use of Commercial Off-the. Shelf (COTS) products; and – Cheaper application deployment through automation • An Environment for Task ORientated Analysis 2

Presentation Overview • • • An ETORA-II application The case for automating the analysis of remotely-sensed data Some hard problems, and the limitations of available software What is required for automation, and why L 3 -Storm and Redleaf? ETORA-II features Summary 3

ETORA-II • Developed for the Macaulay Land Use Research Institute, Aberdeen, Scotland • Problem: the need to update the Land Cover of Scotland (1988) dataset – Census of Scotland’s land cover (>1300 classes); – 20 person years to interpret and digitise aerial photography; – ~£ 2 m • Solution: SYMOLAC-II, constructed using ETORA-II – Complex rules/regulations, involving multiple datasets requiring a wide range of expertise; – Leading to an automated information system for Scotland’s land cover 4 The LCS 8 dataset

ETORA-II • A collaborative effort – MLURI as an end user; – Redleaf Systems as a software developer; – L 3 -Storm as a software developer/provider of COTS products; – IPR agreement in place The LCS 8 dataset 5

The Case for Automation • There is an industry-wide need to increase automation RS data markets • Why? An increasing need for human expertise. . . … and an increasing need for automation RS data volume and variety, availability, awareness – The cost of delivering information to end-users 6

Some Hard Problems • There is more to automation than chaining together a series of operations – many different approaches may be possible; – these approaches may not be linear; – the data and knowledge available for one geographical area may not exist for another, or could be of lesser quality; – the results may be conflicting; – non-mathematical knowledge can improve results and increase efficiency; – processing explanations must be easily accessible; and – new data, knowledge and software resources can become available at any time. • GIS/GIP packages have not been designed to support such processing 7

Achieving Automation • Automation must involve – Knowledge of the problem domain • A dedicated reasoning component is required – Command control of GIS and GIP • The COTS approach; • Reuse legacy systems • A system capable of such automation must be – Flexible • to accommodate the varied data, processing, knowledge, and reasoning strategies necessary to solve a problem; – Extensible • to allow new data, knowledge, and software resources to be readily and cheaply utilised; and – Adaptable • enabling the system to work around typical complexities 8

Achieving Automation • What are ETORA-II’s underlying features, those that make it suited to supporting automated applications – Flexible • to accommodate the varied data, processing, knowledge, and reasoning strategies necessary to solve a problem; – Extensible • to allow new data, knowledge, and software resources to be readily and cheaply utilised; and – Adaptable • enabling the system to work around typical complexities 9

ETORA-II Flexibility • COTS design – re-use of commercial software Arc. View G 2 Arc/INFO PV-WAVE COTS Products 10

ETORA-II Flexibility: COTS SYMOLAC-II, … ? Project ETORA-II Product EDP; RPC; CORBA; COM; Java Arc/INFO G 2 Arc. View Further bridges. . . PV-WAVE Legacy systems ER-Mapper Imagine PCI 11 COTS

ETORA-II Flexibility: COTS SYMOLAC-II, … ETORA-II Arc/INFO Project Product • Contributions Arc. View EDP; RPC; CORBA; development; COM; Java – Eliminate excessive G 2 – Reduce risk; and – Minimise programme costs Future bridges. . . PV-WAVE Legacy systems ER-Mapper Imagine PCI 12 COTS

ETORA-II Flexibility • COTS design – re-use of commercial software • Experts – agent-like collections of knowledge; Experts 13

ETORA-II Flexibility: Experts • Collections of application-specific knowledge, represented within the G 2 component – Blackboard problem-solving model • • Experts can utilise G 2’s powerful knowledge representation and reasoning capability, and command external software Planning and scheduling experts – Solution methodology is dynamic • Concurrent and/or sequential responses • Contributions – – domain knowledge can be modularised; many types of representation and reasoning are possible; iterative, opportunistic reasoning, and “good enough” solutions; others. . . 14

ETORA-II Flexibility • COTS design – re-use of commercial software • Experts – agent-like collections of knowledge • Uncertainty handling – hypotheses and evidence Hypotheses and Evidence 15

ETORA-II Flexibility: Uncertainty • • Derived domain knowledge can be represented as hypotheses, with zero or more evidence items to believe or disbelieve them; Hypotheses and evidence are both created by experts; Uncertainty is a function of the belief; Based on Endorsement Theory (Cohen, 1986) • Contributions – uncertainty considered throughout an application; – can support different uncertainty representations. 16

ETORA-II Flexibility • COTS design – re-use of commercial software • Experts – agent-like collections of knowledge • Uncertainty handling ? – hypotheses and evidence • Explanations – HTML statements produced by experts 17 Explanations

ETORA-II Flexibility: Explanations • Experts associate statements with evidence – explanations do not just record expert activity • Accessible outwith ETORA-II using any browser • Contributions: – solution development; – value-added products 18

Achieving Automation • What are ETORA-II’s underlying features, those that make it suited to supporting automated applications – Flexible • to accommodate the varied data, processing, knowledge, and reasoning strategies necessary to solve a problem; – Extensible • to allow new data, knowledge, and software resources to be readily and cheaply utilised; and – Adaptable • enabling the system to work around typical complexities 19

ETORA-II Extensibility • New knowledge and data resources must be readily accessible – This property emerges from the use of experts – Adding new data – Adding new knowledge • New software resources must be readily accessible – PCI, Imagine, GRASS, ER-Mapper, etc. – G 2 supports: data, object, and RPC connectivity over TCP/IP and DECnet; Active. X, Java, CORBA • The COTS advantage 20

Achieving Automation • What are ETORA-II’s underlying features, those that make it suited to supporting automated applications – Flexible • to accommodate the varied data, processing, knowledge, and reasoning strategies necessary to solve a problem; – Extensible • to allow new data, knowledge, and software resources to be readily and cheaply utilised; and – Adaptable • enabling the system to work around typical complexities 21

ETORA-II Adaptability • Common complexities that require adaptation – – • uncertainties exist within the reasoning processes; there is more than one interpretation of an area; not all areas of interest can be analysed using the most effective data; not all areas of interest can be analysed using the most effective knowledge The property emerges from the use of experts 22

Achieving Automation • This capability is described as task-orientation: the ability to focus on each analysis task, employing the most effective data, method, and software resources to each; • The specific features of a system capable of building taskorientated applications are: – The ability to use multi-source data; – The ability to represent and reason with disparate knowledge; – The ability to dynamically adapt analyses to the specific nature of each task, and the “real-world” aspects that might introduce uncertainty; – A bridged COTS environment; and – The ability to generate detailed reasoning explanations. 23

Summary • • • Automation is not a trivial process; GIS/GIP packages do not have the requisite capability; ETORA-II achieves this capability via – – • A COTS design; Experts; Hypotheses and evidence; and Explanations The toolkit is – Flexible; – Extensible; and – Adaptable • This capability is termed task-orientated 24

Commercialisation • Our partnership – MLURI as an end user – Redleaf as a software developer – L 3 -Storm as a software developer, and provider of COTS products • We believe that task-orientation is necessary to facilitate greater automation within the analysis of remotely-sensed data; • Redleaf and L 3 -Storm are seeking a complementary partner to further the development of ETORA-II towards a commercially viable product – A commercial data/application provider with interests in global markets 25