ABSTRACT Electroencephalogram (EEG) signals are commonly used in analysing the brain activity. The EEG signals have small amplitude and hence, they are often affected by the artefacts. For the efficient processing, it is necessary to remove the artefacts from the EEG signals. This paper develops a technique through deep learning scheme for removing the artefacts present in the EEG signal. Initially, the EEG signals are pre-processed and provided to the feature extraction process, where the wavelet features are extracted from the signal by applying the wavelet transform. The extracted features are provided to the proposed classifier, namely deep-ConvLSTM, for removing the artefacts from the EEG signal. Here, the deep learner is trained based on the proposed Cat Swarm Fractional Calculus Optimisation (CSFCO) algorithm, which is the integration of Cat Swarm Optimisation (CSO) and Fractional Calculus (FC). Experimentation of the proposed technique is carried out by introducing artefacts, such as ECG, EMG, EOG and random noise on the EEG signal. Simulation results carried out on the proposed deep-ConvLSTM depict that the proposed framework has better performance than the comparative techniques with the values of 3888.362, 62.356, and 69.939 dB, for the MSE, RMSE, and SNR, respectively.

中文翻译：

---

基于 Cat Swarm 分数阶微积分优化的深度学习从 EEG 信号中去除伪影

摘要 脑电图 (EEG) 信号通常用于分析大脑活动。脑电信号的幅度很小，因此，它们经常受到人为因素的影响。为了有效处理，有必要从 EEG 信号中去除伪影。本文开发了一种通过深度学习方案去除 EEG 信号中存在的伪影的技术。最初，对 EEG 信号进行预处理并提供给特征提取过程，其中通过应用小波变换从信号中提取小波特征。提取的特征提供给提出的分类器，即 deep-ConvLSTM，用于从 EEG 信号中去除伪影。在这里，深度学习器基于提出的 Cat Swarm 分数阶微积分优化 (CSFCO) 算法进行训练，这是 Cat Swarm Optimization (CSO) 和 Fractional Calculus (FC) 的集成。通过在 EEG 信号上引入人工制品，例如 ECG、EMG、EOG 和随机噪声，对所提出的技术进行实验。对所提出的 deep-ConvLSTM 进行的仿真结果表明，所提出的框架比 MSE、RMSE 和 SNR 的值分别为 3888.362、62.356 和 69.939 dB 的比较技术具有更好的性能。