Orthogonal Features-based EEG Signal Denoising using Fractionally Compressed Architecture

Abstract

Deep neural networks have gained a lot of prominence over the past years by achieving great results in most of the prediction tasks. This paved way for convolutional neural networks for solving problems in the field of image recognition or classification. However training these networks require millions of parameters which makes it computationally expensive and thus it is not feasible to deploy the models on low memory resources or portable devices. This motivated the researchers to find ways of compressing the parameters trained in neural networks leading to the making of lightweight neural networks. Various compression methods are studied which have produced competitive results. We also study the drawbacks of the proposed methods and the future work to be done in this thesis. The area of fractional calculus is also explored which has been applied in the field of deep learning and has produced competitive results. This work presents a fractional one-dimensional convolutional neural network (CNN) autoencoder for denoising the Electroencephalogram (EEG) signals which often get contaminated with noise during the recording process, mostly due to muscle artifacts (MA), introduced by the movement of muscles. The existing EEG denoising methods make use of decomposition, thresholding and filtering techniques. In the proposed approach, EEG signals are first transformed to orthogonal domain using Tchebichef moments before feeding to the proposed architecture. A new hyper-parameter (a) is introduced which refers to the fractional order with respect to which gradients are calculated during back-propagation. It is observed that by tuning a, the quality of the restored signal improves significantly. Motivated by the high usage of portable low energy devices which make use of compressed deep learning architectures, the trainable parameters of the proposed architecture are compressed using randomized singular value decomposition (RSVD) algorithm. The experiments are performed on the standard EEG datasets, namely, Mendeley and Bonn. The study shows that the proposed fractional and compressed architecture performs better than existing state-of-the-art signal denoising methods.