<p>An electroencephalogram (EEG) based emotion recognition framework is developed for applications in human–computer interaction, mental health assessment, neuromarketing, and adaptive learning. The framework employs discriminative neural patterns within specific EEG rhythms through a PSO-optimized infinite-impulse-response (IIR) bandpass filter, ensuring minimal reconstruction loss. The decomposed <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$\delta $</EquationSource> <EquationSource Format="MATHML"><math> <mi>δ</mi> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq2"> <EquationSource Format="TEX">$\theta $</EquationSource> <EquationSource Format="MATHML"><math> <mi>θ</mi> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq3"> <EquationSource Format="TEX">$\alpha $</EquationSource> <EquationSource Format="MATHML"><math> <mi>α</mi> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq4"> <EquationSource Format="TEX">$\beta $</EquationSource> <EquationSource Format="MATHML"><math> <mi>β</mi> </math></EquationSource> </InlineEquation>, and <InlineEquation ID="IEq5"> <EquationSource Format="TEX">$\gamma $</EquationSource> <EquationSource Format="MATHML"><math> <mi>γ</mi> </math></EquationSource> </InlineEquation> rhythms are transformed using the Adaptive Superlet Transform (ASLT) to obtain high-resolution time–frequency representations. These are classified using deep learning models, where Convolutional Neural Network (CNN) and ResNet-18 achieved superior performance. Consequently, a hybrid CNN–ResNet model (HCRNet) is developed, featuring a dual-path fusion mechanism that integrates local CNN features with ResNet’s hierarchical representations. Experiments on the DEAP and DREAMER datasets demonstrate that <InlineEquation ID="IEq6"> <EquationSource Format="TEX">$\beta $</EquationSource> <EquationSource Format="MATHML"><math> <mi>β</mi> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq7"> <EquationSource Format="TEX">$\theta $</EquationSource> <EquationSource Format="MATHML"><math> <mi>θ</mi> </math></EquationSource> </InlineEquation> rhythms exhibit high discriminative power, while the combined rhythms yield the best accuracy of 93.89% and 97.11%, confirming the effectiveness of optimized rhythm-specific analysis for reliable EEG-based emotion recognition.</p>

Decoding Human Feelings: A Dynamic Fusion of Adaptive IIR Filtering and HCRNet for Emotion Recognition

期刊论文
发表日期：

2026年2月3日

查看全文

Amit Kumar Dwivedi,
Om Prakash Verma,
Sachin Taran

摘要

An electroencephalogram (EEG) based emotion recognition framework is developed for applications in human–computer interaction, mental health assessment, neuromarketing, and adaptive learning. The framework employs discriminative neural patterns within specific EEG rhythms through a PSO-optimized infinite-impulse-response (IIR) bandpass filter, ensuring minimal reconstruction loss. The decomposed $\delta $ $δ$ , $\theta $ $θ$ , $\alpha $ $α$ , $\beta $ $β$ , and $\gamma $ $γ$ rhythms are transformed using the Adaptive Superlet Transform (ASLT) to obtain high-resolution time–frequency representations. These are classified using deep learning models, where Convolutional Neural Network (CNN) and ResNet-18 achieved superior performance. Consequently, a hybrid CNN–ResNet model (HCRNet) is developed, featuring a dual-path fusion mechanism that integrates local CNN features with ResNet’s hierarchical representations. Experiments on the DEAP and DREAMER datasets demonstrate that $\beta $ $β$ and $\theta $ $θ$ rhythms exhibit high discriminative power, while the combined rhythms yield the best accuracy of 93.89% and 97.11%, confirming the effectiveness of optimized rhythm-specific analysis for reliable EEG-based emotion recognition.