BITS Faculty Publications

Permanent URI for this communityhttp://localhost:4000/handle/123456789/1867

Browse

Now showing 1 - 5 of 5

Adaptive Radial Basis Functions Neural Network For Motor Imagery Task Classification
(IEEE, 2019) Agarwal, Vandana
Classification of Electroencephalogram (EEG) signals for motor imagery task has been a challenge for researchers due to the complex and highly non-separable structure of the data across different thought classes. For each specific thought in human brain, the EEG signals display a nonstationary behavior. Despite the non-similarity of EEG patterns within a motor imagery class, it is observed that they display some similarity across few samples. In this study, the similar behavior of training patterns of a motor imagery task is captured and patterns are grouped together to form sub-clusters. The sub-cluster centers are obtained using an evolutionary algorithm inspired by the attractiveness of the fireflies. Radial basis functions neural networks, with the sub-cluster centers thus obtained are used for classification. In this study, the convergence of the algorithm is analyzed for BCI Competition IV 2A data set and classification of the two motor imagery classes, left hand and tongue movement, is investigated.
Evolutionary design of Multiquadric radial basis functions neural network for face recognition
(IEEE, 2013) Agarwal, Vandana; Bhanot, Surekha
In this paper, it is proposed to use Multiquadric basis functions at hidden layer of radial basis functions neural networks (RBFNN) for face recognition. The performance of RBFNN depends on the design of the structure of RBFNN, which includes optimal center selection and spread of RBF units, number of neurons at hidden layer, weights etc. Design of hidden layer of RBFNN also includes the choice of basis functions which is proposed to be of Multiquadric basis functions. The shape of Multiquadric basis function plays an important role in the performance of RBFNN in face recognition. A novel evolutionary shape parameter optimization technique inspired by the attractiveness of the natural fireflies is proposed and is used in the design of Multiquadric basis functions for the given face database. The algorithm is tested on two benchmarked face databases ORL and Indian face databases. The proposed technique significantly outperforms the performance of the Gaussian basis functions based RBFNN in terms of face recognition accuracy.
Firefly Inspired Feature Selection for Face Recognition
(IEEE, 2015) Agarwal, Vandana; Bhanot, Surekha
In this paper, an adaptive technique using Firefly Algorithm for feature selection in face recognition is proposed. The artificial fireflies are designed to represent the feature subset and they move in a hyper dimensional space to obtain the best features. The features are extracted using Discrete Cosine Transform (DCT) and Haar wavelets based Discrete Wavelet Transform (DWT). The algorithm is validated using benchmark face databases namely ORL and Yale. The proposed algorithm outperforms various existing techniques. The average recognition accuracy using five randomly selected training samples over four independent runs for the ORL is 94.375%. The accuracy using six training images for Yale face database is 99.16%. The effect of parameter 'gamma', specific to Firefly Algorithm on recognition accuracy is also investigated.
Radial basis function neural network-based face recognition using firefly algorithm
(Springer, 2017-02) Agarwal, Vandana; Bhanot, Surekha
This paper presents an adaptive technique for obtaining centers of the hidden layer neurons of radial basis function neural network (RBFNN) for face recognition. The proposed technique uses firefly algorithm to obtain natural sub-clusters of training face images formed due to variations in pose, illumination, expression and occlusion, etc. Movement of fireflies in a hyper-dimensional input space is controlled by tuning the parameter gamma (γ) of firefly algorithm which plays an important role in maintaining the trade-off between effective search space exploration, firefly convergence, overall computational time and the recognition accuracy. The proposed technique is novel as it combines the advantages of evolutionary firefly algorithm and RBFNN in adaptive evolution of number and centers of hidden neurons. The strength of the proposed technique lies in its fast convergence, improved face recognition performance, reduced feature selection overhead and algorithm stability. The proposed technique is validated using benchmark face databases, namely ORL, Yale, AR and LFW. The average face recognition accuracies achieved using proposed algorithm for the above face databases outperform some of the existing techniques in face recognition.
A Vector-based EEG Signal Feature Extraction Technique for BCI Applications
(IEEE, 2018) Agarwal, Vandana
Brain Computer Interface (BCI) systems offer the ability to effect actuations in the users environment, bypassing the neuro-muscular pathway. The optimal functioning of BCI systems is predicated on two important aspects of the analysis pipeline - informative feature extraction and accurate classification. We propose a simple yet distinct approach to perform the former using a vector-based treatment of signal data and covariance matrices. Our results show a comparable level of performance to certain variants of CSP algorithm. We also present the optimal classifier parameters obtained after parameter-tuning of certain standard classifier models over the BCI Competition III, data-set IVa.