Hybrid Empirical Mode Decomposition and Convolutional Neural Network Framework for Electroencephalography Emotion Recognition via Wavelet-Optimized Images

Abstract

Automatically decoding emotional states through electroencephalography (EEG) poses a significant challenge, largely due to the stochastic and nonlinear dynamics inherent in neural tracking oscillations. To address this, this study presents a novel two-phase framework that integrates empirical mode decomposition (EMD), topographical image generation, and deep convolutional networks to elevate affective state classification accuracy. In the initial phase, multichannel baseline signals from the DEAP repository undergo adaptive decomposition into intrinsic mode functions via EMD, after which singular value decomposition is executed to discard
artifacts and extract the most salient signal traits. These refined feature profiles are subsequently projected onto a 2D spatial grid mirroring the standard 10–20 electrode configuration, yielding a continuous stream of topographic EEG representations. To refine the input space, a discrete wavelet transform routine isolates the most informative frequency coefficients, effectively mitigating the curse of high dimensionality. Ultimately, a convolutional neural network architecture categorizes the target affective responses across both the valence and arousal domains. Empirical evaluations reveal that this EMD-driven topographical scheme, coupled with wavelet-centric refinement, yields highly consistent performance and substantially exceeds the capabilities of conventional classifiers, including support vector machines and shallow ANNs. The proposed system successfully captures concurrent spatial and spectral features of neural dynamics for dependable emotion assessment.

Received: 27 March 2026 | Revised: 26 May 2026 | Accepted: 6 June 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The datasets analyzed during the current study are available in the DEAP repository, which is publicly accessible at http://www.eecs.qmul.ac.uk/mmv/datasets/deap/. For download instructions and access, please refer to http://www.eecs.qmul.ac.uk/mmv/datasets/deap/download.html. Additional data generated in this study are available from the corresponding author on reasonable request.

Author Contribution Statement

Hayder Kareem Saeed: Conceptualization, Methodology, Software, Formal analysis, Investigation, Data curation, Writing – original draft, Visualization. Seyed Enayatallah Alavi: Conceptualization, Methodology, Validation, Formal analysis, Resources, Writing – original draft, Writing – review & editing, Supervision, Project administration, Funding acquisition.