ERU - 2016
http://dl.lib.uom.lk/handle/123/19525
2024-03-28T11:42:00ZA simplified epilepsy classification technique utilizing svd
http://dl.lib.uom.lk/handle/123/19550
A simplified epilepsy classification technique utilizing svd
Prabhakar, SK; Rajaguru, H
Jayasekara, AGBP; Amarasinghe, YWR
EEG signals represent both the brain function and
also the status of the whole body, i.e. a simple action as blinking
the eyes introduces oscillation in the EEG records. The EEG is
a direct way to measure neural activities and it is important in
the area of biomedical research to understand and develop new
processing techniques. EEG signal pre-processing and postprocessing
methods include EEG signal modeling,
segmentation, filtering and de-noising, and EEG processing
methods which consist of two tasks, namely, feature
extraction/dimensionality reduction and classification. In this
paper, the performance analysis of Independent Component
Analysis (ICA) is considered as a dimensionality reduction
technique followed by Singular Value Decomposition (SVD) as
a Post Classifier for the Classification of Epilepsy Risk Levels
from EEG Signals. The analysis is done in terms of bench mark
parameters such as Performance Index (PI), Quality Values
(QV), Sensitivity, Specificity and Time Delay.
2016-04-01T00:00:00ZShort-term traffic prediction with visitor location registry data
http://dl.lib.uom.lk/handle/123/19549
Short-term traffic prediction with visitor location registry data
Dilhasha, F; Fernando, K; Godahewa, R; Ossen, S; Perara, AS; Walpola, M
Jayasekara, AGBP; Amarasinghe, YWR
Increasing road traffic is a major issue in current
world. In this paper, we propose a set of prediction models
that can perform short term traffic prediction for a given road
segment. These prediction models have been developed using
Neural Networks (NN), Bayesian Networks, Hidden Markov
Models, variations of Regression and ensemble approaches of
these models. CCTV records are used for validation of the results
based on which a maximum accuracy of 85% was achieved.
2016-04-01T00:00:00ZModeling and forecasting opec reference basket crude oil prices using artificial neural networks
http://dl.lib.uom.lk/handle/123/19548
Modeling and forecasting opec reference basket crude oil prices using artificial neural networks
Mahanthege, SR; Chandrasekera, NV; Jayasundara, DDM
Jayasekara, AGBP; Amarasinghe, YWR
With Organization of petroleum exporting countries (OPEC) being the dominant player in the crude oil industry, it is of great importance to accurately forecast OPEC reference basket (ORB) crude oil prices. This study focuses on forecasting ORB crude oil prices using two Artificial Neural Network (ANN) models: Feedforward Neural Network (FFNN) and Time-Delay Neural Network (TDNN). Training of the networks were done by changing several parameters of the networks. Levenberg-Marquardt training algorithm has been used in training ANN. Results indicates that the TDNN model outperforms the FFNN model in forecasting daily ORB crude oil prices.
2016-04-01T00:00:00ZHuman action detection using space-time interest points
http://dl.lib.uom.lk/handle/123/19547
Human action detection using space-time interest points
Sriashalya, S; Ramanan, A
Jayasekara, AGBP; Amarasinghe, YWR
The bag-of-features (BoF) approach for
human action classification uses spatio-temporal
features to assign the visual words of a codebook. Space
time interest points (STIP) feature detector captures the
temporal extent of the features, allowing distinguishing
between fast and slow movements. This study compares
the relative performance of action classification on KTH
videos using the combination of STIP feature detector
with histogram of gradient orientations (HOG) and
histograms of optical flow (HOF) descriptors. The
extracted descriptors are clustered using K-means
algorithm and the feature sets are classified with two
classifiers: nearest neighbour (NN) and support vector
machine (SVM). In addition, this study compares actionspecific
and global codebook in the BoF framework.
Furthermore, less discriminative visual words are
removed from initially constructed codebook to yield a
compact form using likelihood ratio measure. Testing
results show that STIP with HOF performs better than
HOG descriptors and simple linear SVM outperforms
NN classifier. It can be noticed that action-specific
codebooks when merged together perform better than
globally constructed codebook in action classification on
videos.
2016-04-01T00:00:00Z