<p>Brain-Computer Interface (BCI) is a system that facilitates interaction between the brain and external devices by processing cerebral activity. BCIs are widely used in applications in neuro rehabilitation, neuro care, and prosthetics. Among the various methods, Electroencephalography (EEG) is noted as one of the widely used non-invasive techniques for measuring electrical brain impulses. However, it is highly prone to artifacts from sources like eye movements, muscle activity, and heartbeats, which can distort the signals and potentially lead to degradation in BCI performance. As a result, efficiently eliminating these artifacts continues to be a major challenge in EEG signal processing. To tackle this challenge, many deep learning methods have been explored. However, identifying the optimal network architecture and tuning its parameters remains time-consuming and demands substantial domain expertise. To address the challenge, we propose a novel Dense block centered Convolution Neural Network built using the method of Differential Evolution (DEDensNet). In the proposed model, EEG noisy signals given as the input to the dense block based neural network to automatically remove artifacts, in which both the architecture and initial parameters are optimized using evolution process. This makes it possible to eradicate artifacts efficiently. The effectiveness of the model is compared with that of existing models on three different datasets, yielding promising results and a reduced root mean square error along with improved signal-to-noise ratio and an improvement of 23.77% in the classification task.</p>