Rn Williams

Texture-based classification of sub-Antarctic vegetation communities on Heard Island

This study was the first to use high-resolution IKONOS imagery to classify vegetation communities on sub-Antarctic Heard Island. We focused on the use of texture measures, in addition to standard multispectral information, to improve the classification of sub-Antarctic vegetation communities. Heard Island’s pristine and rapidly changing environment makes it a relevant and exciting location to study the regional effects of climate change. This study uses IKONOS imagery to provide automated, up-to-date, and non-invasive means to map vegetation as an important indicator for environmental change. Three classification techniques were compared: multispectral classification, texture based classification, and a combination of both. Texture features were calculated using the Grey Level Co-occurrence Matrix (GLCM). We investigated the effect of the texture window size on classification accuracy. The combined approach produced a higher accuracy than using multispectral bands alone. It was also found that the selection of GLCM texture features is critical. The highest accuracy (85%) was produced using all original spectral bands and three uncorrelated texture features. Incorporating texture improved classification accuracy by 6%.

An RFID Attacker Behavior Taxonomy

A taxonomy of system attacker behavior reveals security vulnerabilities in RFID authorization and monitoring systems. RFID systems are classified by their informational goals-typically, authorization and monitoring. Authorization systems replace the more traditional approaches of granting an entity access to a particular zone, whereas monitoring systems establish an entitys location in that zone. Although their informational goals differ, the underlying hardware is identical for both types of systems; consequently, attacks at the hardware level are the same. However, because attacker behavior invalidates each subsystems informational goals differently, RFID security requirements should consider these goals individually.

A new model for classifying DNA code inspired by neural networks and FSA

This paper introduces a new model of classifiers CL(V,E,l,r) designed for classifying DNA sequences and combining the flexibility of neural networks and the generality of finite state automata. Our careful and thorough verification demonstrates that the classifiers CL(V,E,l,r) are general enough and will be capable of solving all classification tasks for any given DNA dataset. We develop a minimisation algorithm for these classifiers and include several open questions which could benefit from contributions of various researchers throughout the world.

A technique for the identification and analysis of icebergs in synthetic aperture radar images of Antarctica

This paper describes an image analysis technique developed to identify icebergs depicted in synthetic aperture radar images of Antarctica and to determine the outlines of these icebergs. The technique uses a pixel bonding process to delineate the edges of the icebergs. It then separates them from the background water and sea ice by an edge-guided image segmentation process. Characteristics such as centroid position and iceberg area are calculated for each iceberg segment and placed in a file for input to appropriate statistical data analysis software. The technique has been tested on three ERS-1 SAR sub-images in which it succeeded in identifying virtually all segments containing icebergs of size 6 pixels or larger. The images were first passed through an averaging filter to reduce speckle. This process produced a pixel size of 100 m x 100 m. As implemented, the technique over-estimates iceberg areas by about 20% on average and the detection rate falls off rapidly for icebergs less than six pixels in size. Performance in these areas is expected to improve when additional stages, based on a more detailed analysis of pixel intensity, are implemented.

DynamicWEB: Adapting to Concept Drift and Object Drift in COBWEB

Examining concepts that change over time has been an active area of research within data mining. This paper presents a new method that functions in contexts where concept drift is present, while also allowing for modification of the instances themselves as they change over time. This method is well suited to domains where subjects of interest are sampled multiple times, and where they may migrate from one resultant concept to another due to Object Drift. The method presented here is an extensive modification to the conceptual clustering algorithm COBWEB, and is titled DynamicWEB.

An automated image analysis system for determining sea-ice concentration and cloud cover from AVHRR images of the Antarctic

The Australian Bureau of Meteorology operates a meteorological centre at the Casey station in East Antarctica. This centre is able to receive Advanced Very High Resolution Radiometer (AVHRR) data directly from the NOAA satellites and the Bureau uses these data for operational and research applications. Currently the AVHRR images are interpreted manually. However, the ICEMAPPER analysis system is able to automate this process. The system uses one set of rules to identify areas of high cloud, low cloud, open water, land ice and sea ice and another to determine sea-ice concentration. Some of the rules in ICEMAPPER were derived using information obtained from published and unpublished research, augmented by details given by practising meteorologists. Others were created by having an expert image interpreter analyse a set of representative AVHRR images and processing these analyses, using a statistical package, to produce rules which closely duplicate the manual analyses. The system was tested on six AVHRR images of the East Antarctic coastline, acquired late in the 1997/1998 summer season. It successfully identified 85% of pixels, selected from the images using a regular grid, as belonging to one of the five surface classes: high cloud, low cloud, open water, land ice or sea ice.

Application of the Particle Filter to Tracking of Fish in Aquaculture Research

The analysis of fish movement as an indicator of fish behaviour plays an important role in aquaculture research. Currently observations are carried out manually using video recordings. In this paper we describe a tracking system which can automatically detect and track two fish in a video sequence in a small aquaculture tank. The system is based on the particle filter tracking algorithm augmented by an adaptive partition scheme and using a global nearest neighbour approach for data association. Results show that this method is sufficient for simple interactions where fish bypass each other without significant changes in velocity. However, more complex scenarios involving occlusions, loss of tracks and fish manoeuvres can cause ambiguity during data association.

Midwater acoustic modeling for multibeam sonar simulation

Simulation and modeling software has been developed to generate synthetic midwater multibeam data. Essentially, the simulator can be considered as a virtual test tank. In order to develop multibeam data analysis methods for fisheries research, it is essential to have a variety of test data sets available, which are ground truthed, georeferenced and corrected for vessel motion. Since equipment and ship time are expensive and data quality not always guaranteed, the simulator provides an effective alternative. The seabed and any objects in the water column such as fish and fish schools can be defined in a 3‐dimensional space. A specification for a generic linear array multibeam sonar and its position in space and time can be chosen. The acoustic model implements the technique of acoustic ray‐tracing to obtain the pressure at the transducer face, which is converted to individual samples by modeling the working of a digital multibeam system. Beamforming is performed on the fly, and both raw and beamformed comp...

DynamicWEB: profile correlation using COBWEB

Establishing relationships within a dataset is one of the core objectives of data mining. In this paper a method of correlating behaviour profiles in a continuous dataset is presented. The profiling problem which motivated the research is intrusion detection. The profiles are dynamic in nature, changing frequently, and are made up of many attributes. The paper describes a modified version of the COBWEB hierarchical conceptual clustering algorithm called DynamicWEB. DynamicWEB operates at runtime, keeping the profiles up to date, and in the correct location within the clustering tree. Further, as there are a number of attributes within the domain of interest, the tree also extends multi-dimensionally. This allows for multiple correlations to occur simultaneously, focusing on different attributes within the one profile.

Automated intelligent abundance analysis of scallop survey video footage

Underwater video is increasingly being pursued as a low impact alternative to traditional techniques (such as trawls and dredges) for determining abundance and size frequency of target species. Our research focuses on automatically annotating survey scallop video footage using artificial intelligence techniques. We use a multi-layered approach which implements an attention selection process followed by sub-image segmentation and classification. Initial attention selection is performed using the University of Southern Californias (USCs) iLab Neuromorphic Visual Toolkit (iNVT). Once the iNVT has determined regions of potential interest we use image segmentation and feature extraction techniques to produce data suitable for analysis within the Weka machine learning workbench environment.