Papers by Daniel W McMichael
Improving Multispectral Mine Detection Methods By Compensating For Clutter
The detection of signals in clutter depends not only on good signal models and optimal processing... more The detection of signals in clutter depends not only on good signal models and optimal processing, but also on accurate clutter models. An eective approch to land mine detection is the use of infrared imagery to detect thermal perturbations in the soil surface temperature caused by buried land mines. The purpose of this study is to investigate the value of augmenting long- and short-wave infrared imagery by red{green{blue visual imagery, in order to model vegetative clutter and thereby to detect land mines with greater probability and suer fewer false alarms. This study applies the shared mixture classier reported in earlier [14]. 1. HISTORICAL PERSPECTIVE The problem of safely and reliably detecting buried landmines has not yet been solved in any situation. We believe that any satisfactory device will require multiple sensors and high quality algorithms informed by physical and expert information and statistical data. For the last three years we have carried out a number of inves...
Remote Sensing With Spotter
Spotter is a exible extensible software tool for interactive development, application and testing... more Spotter is a exible extensible software tool for interactive development, application and testing of algorithms for detecting objects in multi-spectral imagery. Images from dierent sensors are registered, and low level features are extracted. Feature selection and reduction algorithms generate reduced dimension feature sets. Feature images are soft classied, pixel by pixel, and the output is used to drive a preliminary detection process to identify regions of interest (ROI). Geometric features are extracted from the ROI, which are then soft classied, and a detection decision decision is made for each. Many algorithms are provided, including Gaussian mixture model classiers and shared mixture classiers. Mixture of experts classiers enable physically inspired classiers to be incorporated to complement statistical designs. 1. INTRODUCTION The Spotter package was developed by CSSIP to aid in the creation and evaluation of algorithms for detecting buried land mines. The algorithms ...
Cool Fusion Statistical Modelling for Data Fusion
The consequences of the use of probability in modelling under uncertainty are explored, and it is... more The consequences of the use of probability in modelling under uncertainty are explored, and it is shown how data and parameter independence and likelihood modularity emerge as desirable properties for statistical models. They lead directly to multi-object models which themselves create the association problem. The expectation-maximisation (EM) technique is introduced and then applied to discriminative training of a Gaussian mixture model classi er. Generalised training is introduced which interpolates between the extremes of maximising discriminative and non-discriminative likelihoods. A generalisation of EM, the conditional expectation-maximisation process is presented, and applied to designing an algorithm for estimating the location and rotation parameters for transforming a 3D reference model to generate an observed x-ray image. The EM E-step is derived for the case in which the nuisance variables (\missing data") are divided into sets and integrated out separately. This co...
This paper describes a multiple sensor tracking simulation software package (MUST) developed by t... more This paper describes a multiple sensor tracking simulation software package (MUST) developed by the Data and Information Group of CSSIP which is written in Matlab and C++. The aim of MUST is to assess the performance of a single-target multisensor tracker, the extended Kalman filter-probabilistic data association (EKF-PDA) tracker, under different target/sensor/environment scenarios. The simulator enables the user to specify the parameters of a single-target tracking scenario with multiple sensors. The simulator then generates the target trajectory and estimates the target trajectory from the asynchronous reports such as bearing, range, range rate and elevation angle from multiple sensors. Finally it evaluates the performance of the tracker under the specified scenario. This paper also presents some simulation results for different tracking scenarios.
2007 I D C Information, Decision and Control
Bayesian networks provide a powerful intelligent information fusion architecture for modeling pro... more Bayesian networks provide a powerful intelligent information fusion architecture for modeling probabilistic and causal patterns involving multiple random variables. This paper advances a computable theory of learning discrete Bayesian networks from data. The theory is based on the MAP-MDL principles for maximizing the joint probability or interchangeably miniziming the joint description length of the data and the Bayesian network model including the network structure and the probability distribution parameters. The computable formalisms for the data likelihood given a Bayesian network structure, the description length of a structure, and the estimation of the parameters given a structure are derived. EM algorithms are constructed for handling incomplete and soft data.
1999 Information, Decision and Control. Data and Information Fusion Symposium, Signal Processing and Communications Symposium and Decision and Control Symposium. Proceedings (Cat. No.99EX251), 1999
Fuzziness and randomness are two distinct components of uncertainty. While fuzzy sets are a rigor... more Fuzziness and randomness are two distinct components of uncertainty. While fuzzy sets are a rigorous softening of random sets, many of the operations de ned in fuzzy logic lack a complete formalism, and are not strongly supported by experimental evidence. Causal Probabilistic Networks CPN or Bayesian networks provide an ultimately exible inference m e chanism based on Bayesian probability principles. However, CPNs su er from the overwhelmingly large conditional probability tables with discrete variables. Fuzzi cation of continuous or crisp variables reduces the size of conditional probability tables to practically acceptable levels and these tables exhaustively encompass all the intuitive and fuzzy rules for inference p r oblems. In this way, we reach a new inference engine, called fuzzy causal probabilistic networks, which provides a rigorous formalism for inference under fuzziness and randomness.
A Meta-grammar for CCG
... 1 The authors would like to credit the work of Geoff Jarrad in developing these two combinato... more ... 1 The authors would like to credit the work of Geoff Jarrad in developing these two combinators. ... Thus we hope to benefit from lower coding times and easier debugging, with the option to port to C and re-integrate any mature code that is deemed time-critical. ...
Automatic Complexity Determination of Gaussian Mixture Models with the EMS Algorithm
Journal of Artificial Intelligence Research, 2006
Functional Combinatory Categorial Grammar (FCCG) advances the field of combinatory categorial gra... more Functional Combinatory Categorial Grammar (FCCG) advances the field of combinatory categorial grammars by enabling semantic dependencies to be determined directly from the syntactic derivation under the action of a small set of extraction rules. Predicates are extracted composably and can be used to apply semantic constraints during parsing. The approach is an alternative to that of classical CCG which requires (i) mapping from categories to lambda expressions, (ii) a set of semantic transformation rules for unary combination, and (iii) an explicit β-reduction stage. GFCCG, a generalised form of the grammar, has previously been applied to situation assessment .
Fusing multiple images and extracting features for visual inspection
A Statistical Approach to Situation Assessment
BARTIN: a neural structure that learns to take Bayesian minimum risk decisions
Bayesian growing and pruning strategies for MAP-optimal estimation of Gaussian mixture models
Estimating Gaussian Mixture Models from Data with Missing Features
Fourth International Symposium on Signal Processing and Its Applications, 1996
... 1977. 157, An inequality with applications to statistical estimation for probabilistic functi... more ... 1977. 157, An inequality with applications to statistical estimation for probabilistic functions of Markov processes and to a model for ecology Baum, Eagon - 1967. 101, On convergence properties of the em algorithm for gaussian mixtures Xu, Jordan - 1996. ...
Robust recursive Lp estimation
Control Theory and Applications Iee Proceedings D See Also Iee Proceedings Control Theory and Applications, Apr 1, 1990
Statistical Models for Situation Awareness
1999 Information Decision and Control Data and Information Fusion Symposium Signal Processing and Communications Symposium and Decision and Control Symposium Proceedings, Feb 1, 1999
Information Fusion, Causal Probabilistic Network And Probanet - I: Information Fusion Infrastructure and Probabilistic Knowledge Representation
ABSTRACT
Data fusion for vehicle-borne mine detection
Uploads
Papers by Daniel W McMichael