Human vision uses mechanisms of covert attention to selectively process interesting information and overt eye movements to extend this selectivity ability. Thus, visual tasks can be effectively dealt with by limited processing resources. Modelling visual attention for machine vision systems is not only critical but also challenging. In the machine vision literature there have been many conventional attention models developed but they are all space-based only and cannot perform object-based selection. In consequence, they fail to work in real-world visual environments due to the intrinsic limitations of the space-based attention theory upon which these models are built. The aim of the work presented in this thesis is to provide a novel human-like visual selection framework based on the object-based attention theory recently being developed in psychophysics. The proposed solution – a Hierarchical Object-based Attention Framework (HOAF) based on grouping competition, consists of two closely-coupled visual selection models of (1) hierarchical object-based visual (covert) attention and (2) object-based attention-driven (overt) saccadic eye movements. The Hierarchical Object-based Attention Model (HOAM) is the primary selection mechanism and the Object-based Attention-Driven Saccading model (OADS) has a supporting role, both of which are combined in the integrated visual selection framework HOAF. This thesis first describes the proposed object-based attention model HOAM which is the primary component of the selection framework HOAF. The model is based on recent psychophysical results on object-based visual attention and adopted grouping-based competition to integrate object-based and space-based attention together so as to achieve object-based hierarchical selectivity. The behaviour of the model is demonstrated on a number of synthetic images simulating psychophysical experiments and real-world natural scenes. The experimental results showed that the performance of our object-based attention model HOAM concurs with the main findings in the psychophysical literature on object-based and space-based visual attention. Moreover, HOAM has outstanding hierarchical selectivity from far to near and from coarse to fine by features, objects, spatial regions, and their groupings in complex natural scenes. This successful performance arises from three original mechanisms in the model: grouping-based saliency evaluation, integrated competition between groupings, and hierarchical selectivity. The model is the first implemented machine vision model of integrated object-based and space-based visual attention. The thesis then addresses another proposed model of Object-based Attention-Driven Saccadic eye movements (OADS) built upon the object-based attention model HOAM, ii as an overt saccading component within the object-based selection framework HOAF. This model, like our object-based attention model HOAM, is also the first implemented machine vision saccading model which makes a clear distinction between (covert) visual attention and overt saccading movements in a two-level selection system – an important feature of human vision but not yet explored in conventional machine vision saccading systems. In the saccading model OADS, a log-polar retina-like sensor is employed to simulate the human-like foveation imaging for space variant sensing. Through a novel mechanism for attention-driven orienting, the sensor fixates on new destinations determined by object-based attention. Hence it helps attention to selectively process interesting objects located at the periphery of the whole field of view to accomplish the large-scale visual selection tasks. By another proposed novel mechanism for temporary inhibition of return, OADS can simulate the human saccading/ attention behaviour to refixate/reattend interesting objects for further detailed inspection. This thesis concludes that the proposed human-like visual selection solution – HOAF, which is inspired by psychophysical object-based attention theory and grouping-based competition, is particularly useful for machine vision. HOAF is a general and effective visual selection framework integrating object-based attention and attentiondriven saccadic eye movements with biological plausibility and object-based hierarchical selectivity from coarse to fine in a space-time context.

Genetic epidemiology is the study of the joint action of genes and environmental factors in determining the phenotypes of diseases. The twin study is a classic and important epidemiological tool, which can help to separate the underlying effects of genes and environment on phenotypes. Twin data have been widely examined using traditional methods to genetic epidemiological research. However, they provide a rich sources information related to many complex phenotypes that has the potential to be further explored and exploited. This thesis focuses on two major genetic epidemiological approaches: familial aggregation analysis and linkage analysis, using twin data from TwinsUK Registry. Structural equation modelling (SEM) is a conventional method used in familial aggregation analysis, and is applied in this research to discover the underlying genetic and environmental influences on two complex phenotypes: coping strategies and osteoarthritis. However, SEM is a confirmatory method and relies on prior biomedical hypotheses. A new exploratory method, named MDS-C, combining multidimensional scaling and clustering method is developed in this thesis. It does not rely on using prior hypothetical models and is applied to uncover underlying genetic determinants of bone mineral density (BMD). The results suggest that the genetic influence on BMD is site-specific. Haseman-Elston (H-E) regression is a conventional linkage analysis approach using the identity by descent (IBD) information between twins to detect quantitative trait loci (QTLs) which regulate the quantitative phenotype. However, it only considers the genetic effect from individual loci. Two new approaches including a pair-wise H-E regression (PWH-E) and a feature screening approach (FSA) are proposed in this research to detect QTLs allowing gene-gene interaction. Simulation studies demonstrate that PWH-E and FSA have greater power to detect QTLs with interactions. Application to real-world BMD data results in identifying a set of potential QTLs, including 7 chromosomal loci consistent with previous genome-wide studies.

This study concerns the use of particular technologies and of Geographical Information System (GIS) as tools to aid effective planning and decision making in the context of a Developing Country such as Malaysia. As in other Developing Countries, the main problems of using GIS within small organisations such as local authorities and planning authorities in Malaysia are associated with the availability of data, the availability of skilled personnel and the constraints of budgets. A low-cost digital data acquisition method that includes the use of low-cost scanners and a low-cost digital photogrammetric workstation (DPW) are proposed for generating some of the necessary data. The proposed low-cost method includes the use of low-cost A4 format desktop publishing scanners (DTP) and the Desktop Mapping System (DMS) version 3.1, which is a PC-based digital photogrammetric workstation (DPW). The ARC/INFO version 6.1 software has been employed as the data integration, data visualisation and data analysis tool. A series of tests on DTP scanners and digital photogrammetric products generated from a low-cost DPW have been carried out to confirm that these low-cost data acquisition methods are suitable for generating the required digital data accurately, quickly and without the need for highly skilled personnel. Results from accuracy assessment of DTP scanners have indicated that, while the distortion errors introduced by DTP scanner imperfection are significant, they can be minimised using proper calibration procedures. Results for accuracy assessment have further indicated that high quality DEMs and orthoimages can be produced with a low-cost DPW provided high quality aerial photo images and ground control points are available. The Petaling District in the Klang Valley Region, Malaysia has been used as the study area to demonstrate the role of digital photogrammetry, satellite remote sensing and GIS in the stage-by stage site selection process.

In order to incorporate inductive learning techniques into a knowledge-based design model and an integrated knowledge-based design support system architecture, the computational techniques for developing a knowledge-based design support system architecture and the role of inductive learning in AI-based design are investigated. This investigation gives a background to the development of an incremental learning model for design suitable for a class of design tasks whose structures are not well known initially. This incremental learning model for design is used as a basis to develop knowledge-based design support system architecture that can be used as a kernel for knowledge-based design applications. This architecture integrates a number of computational techniques to support the representation and reasoning of design knowledge. In particular, it integrates a blackboard control system with an assumption-based truth maintenance system in an object-oriented environment to support the exploration of multiple design solutions by supporting the exploration and management of design contexts. As an integral part of this knowledge-based design support architecture, a design concept learning system utilising a number of unsupervised inductive learning techniques is developed. This design concept learning system combines concept formation techniques with design heuristics as background knowledge to build a design concept tree from raw data or past design examples. The effectiveness of this knowledge-based design support architecture and the design concept learning system is demonstrated through a realistic design domain, the design of small-molecule drugs one of the key tasks of which is to identify a pharmacophore description (the structure of a design problem) from known molecule examples. In this thesis, knowledge-based design and inductive learning techniques are first reviewed. Based on this review, an incremental learning model and an integrated architecture for intelligent design support are presented.

The understanding of how processes in natural phenomena interact at different scales of time has been a great challenge for humans. How information is transferred across scale is fundamental if one tries to scale up from finer to coarse levels of granularity. Computer simulation has been a powerful tool to determine the appropriate amount of detail one has to impose when developing simulation models of such phenomena. However, it has been proved to be difficult to represent change at many scales of time and subject to cyclical processes. This issue has received little attention in traditional AI work on temporal reasoning but it becomes important in more complex domains, such as ecological modelling. Traditionally, models of ecosystems have been developed in imperative languages. Very few of those temporal logic theories have been used to the specification of simulation models in ecology. The aggregation of processes working at different scales of time is very difficult (sometimes impossible) to do reliably. The reason is because these processes influence each other, and their functionality does not always scale to other levels. Thus the problems to tackle are representing cyclical and interacting processes at many scales and to provide a framework to make the integration of such processes more reliable. We propose a framework for temporal modelling which allows modellers to represent cyclical and interacting processes at many scales. This theory combines both aspects by means of <I>modular temporal classes</I> and an underlying special temporal unification algorithm. To allow integration of different models they are developed as agents to run within a certain range of autonomy in a multi-agent system architecture. This <I>Ecoagency </I>framework is evaluated on ecological modelling problems and compared to a formal language for describing ecological systems.

The results of scientific activities are often made available on the Web. Large volumes of information are typically kept and managed in an ad-hoc manner, and a substantial effort is required from scientists to discover data with which they can work. This issue can be seen as a part of the vision for the Semantic Web in its attempt to bring structure and meaningful relationships between concepts and objects that exist in the Web. A system has been developed to discover relevant resources given a small set of training documents. The approach is to discover the most discriminatory terms and their relationships, then to use these to develop a domain ontology. Several series of experiments were undertaken to determine the most appropriate preprocessing steps and weighting scheme. The ontology is represented using RDF and mapped into relational structures. It contains semantic information about meteorology. It is used in information retrieval as a query expansion step in computing the ontology context for the users' requests. The experiments we conducted show the effectiveness of query expansion with terms selected from domain ontology. The system presented here is not specifically exclusive to meteorology, but it can be applied to any information domain providing that an initial set of relevant documents forming the training set exists. We have shown that satisfactory results can be achieved with a small initial set of documents. The document base for information retrieval grows with the discovery of new documents.

No description available.

No description available.

Transformer fault diagnosis through Frequency Response Analysis (FRA) has been receiving a great deal of attention in recent years. As a comparative technique, FRA has good capability and sensitivity in detecting mechanical faults that are difficult to identify by conventional condition assessment techniques. Power transformers are among the most expensive equipment owned by electric utilities, and it is not reasonable to produce deformation on actual transformers and carry out measurement sensitivity studies. On the other hand, simulation models, which can accurately reproduce transformer high frequency behaviours, are flexible tools for performing FRA deformation type sensitivity studies for deriving FRA interpretation rules. The main objective of this thesis is to develop appropriate simulation models for use in FRA diagnosis and to improve the interpretation of FRA responses through simulation studies. The transformer models developed at the University of Manchester (then UMIST) were by far the best representation of state-of-art modelling techniques; the inductance and the capacitance of the basic model unit were calculated using winding geometry and material properties, the frequency dependent conductive and dielectric losses were also included. In addition, mutual capacitive and inductive couplings between units were carefully considered to ensure the accuracy of the model. However, there is still some room for improvement on these models and during this PhD research, major contributions are made on as. follows: firstly take core effect into consideration to reproduce valid FRA characteristic representation in the low frequencies, secondly status of network terminal nodes are uniformed represented by externally connecting an impedance so that during FRA deformation sensitivity study, it is flexible to change the terminal condition, thirdly reconfigure the network node and unit relationship so that tap winding connection are precisely represented as the design, finally convert the single-phase model to a three-phase model and by developing a reduced matrix model, keep the simulation accuracy intact for a three-phase transformer up to 2 MHz, at the same time reduce computational time significantly. In detail, this PhD thesis describes the following three parts of my research: Firstly a transformer model incorporating a magnetic core based on the Principle of Duality is established to interpret low frequency characteristics of FRA responses (from 10Hz to up to 1 kHz). This model includes leakage inductances and capacitances of windings and can explain FRA low frequency differences caused by asymmetry of magnetic paths in three-limb and five-limb core transformers. Secondly, FRA characteristics were studied systematically using a component-system approach through building models for single windings, a one-phase winding set and finally the three-phase transformer. In this way the effects of winding structure, inductive and capacitive coupling among windings, among phases and terminal connection effect on FRA characteristics were studied. FinaUya complete three-phase transformer reduced matrix model is built, that can flexibly represent winding terminal connection and precisely describe tap positions. Using this modelling strategy, transmission power transformers at 2751132 kVand 275/33 kV voltage levels are simulated and numerous deformation sensitivity studies are performed, in order to gain better understanding on their FRA characteristics and to identify FRA features of different winding deformation types on these transformers. The research indicates that the overall approach used to develop these simulation models has helped in improving interpretation of FRA responses. The transformer modelling techniques being developed, with further refinement, can be a useful tool for FRA diagnosis and benefit the test engineers from the industry.

This thesis considers the guidance and control of unmanned vehicles within complex environments. A systems engineering approach was adopted where significant effort was directed towards defining a high level capability requirement and subsequent problem exploration, decomposition and definition, prior to addressing the technical focus. The goal of this approach was to ensure that technical work was directed towards realistic end user requirements and operational scenarios. As the complexity of an operational environment increases, so does the requirement to consider the local obstacle space continually, and this is aided by splitting the motion planning functionality into distinct global and local layers. The technical focus of this thesis is on the development and simulation-based testing of a new local motion planning and control framework, where knowledge of i) feasible vehicle manoeuvre constraints ii) local obstacle map iii) current environment conditions are all combined into a continuous receding horizon approach. This framework separates the output and control space elements of the problem, reducing the complexity of the local motion trajectory optimisation and therefore enabling faster design and increased horizon length. Bezier polynomial functions are used to describe local motion trajectories which are constrained to vehicle performance limits and optimised to achieve a specified goal. The primary problem addressed is ‘situation-aware’ trajectory tracking, but other local motion planning modes are also considered. Development and testing of the new framework is undertaken within simulation (Matlab), based on a nonlinear 6 degree of freedom model of a quadrotor unmanned air vehicle. Situation-aware trajectory tracking is demonstrated in the presence of static and dynamic obstacles, as well as the presence of realistic turbulence and gusts. The immediate-term deconfliction of multiple unmanned vehicles, and multiple formations of unmanned vehicles, is also demonstrated, including the provision of rules-of-the-air type behaviour