<p>In EEG emotion recognition, there is a problem of time-consuming and laborious parameter optimization when mapping one-dimensional data to two-dimensional or three-dimensional data for processing. This paper proposes an IDCNN model based on frequency band and region attention mechanisms. Features are extracted from EEG signals, and optimal feature selection is performed using 1test. A novel 1DCNN emotion recognition model is designed based on the extracted features, providing interpretability for parameter selection and convolution operations. Finally, considering the different emotional response capabilities of the left and right brain regions, we propose a brain region attention mechanism combined with frequency band attention mechanisms to better focus on brain regions and frequency bands relevant to emotion. The proposed Self<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7015_Article_IEq1.gif" Format="GIF" Height="11" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$_{AT}$</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow> <mi mathvariant="italic">AT</mi> </mrow> <mrow /> </mmultiscripts> </math></EquationSource> </InlineEquation>-1DCNN model achieves average recognition rates of <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7015_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">$94.01 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>94.01</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7015_Article_IEq3.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">$93.55 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>93.55</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> in two-class experiments on valence and arousal dimensions of DEAP EEG emotion data, and an average recognition rate of <InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7015_Article_IEq4.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="51" /> </InlineMediaObject> <EquationSource Format="TEX">$89.38 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>89.38</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> in four-class experiments, improving by <InlineEquation ID="IEq5"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7015_Article_IEq5.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="43" /> </InlineMediaObject> <EquationSource Format="TEX">$2.96 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>2.96</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq6"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7015_Article_IEq6.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="43" /> </InlineMediaObject> <EquationSource Format="TEX">$3.31 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>3.31</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>, and <InlineEquation ID="IEq7"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7015_Article_IEq7.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="43" /> </InlineMediaObject> <EquationSource Format="TEX">$7.69 \%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>7.69</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>, respectively, compared to existing 1DCNN models.</p>

Self-attention-based 1DCNN model for multiclass EEG emotion classification

期刊论文
发表日期：

2025年2月19日

查看全文

Shrishtika Raikwar,
A. V. R. Mayuri

摘要

In EEG emotion recognition, there is a problem of time-consuming and laborious parameter optimization when mapping one-dimensional data to two-dimensional or three-dimensional data for processing. This paper proposes an IDCNN model based on frequency band and region attention mechanisms. Features are extracted from EEG signals, and optimal feature selection is performed using 1test. A novel 1DCNN emotion recognition model is designed based on the extracted features, providing interpretability for parameter selection and convolution operations. Finally, considering the different emotional response capabilities of the left and right brain regions, we propose a brain region attention mechanism combined with frequency band attention mechanisms to better focus on brain regions and frequency bands relevant to emotion. The proposed Self $_{AT}$ $_{AT}$ -1DCNN model achieves average recognition rates of $94.01 \%$ $94.01 %$ and $93.55 \%$ $93.55 %$ in two-class experiments on valence and arousal dimensions of DEAP EEG emotion data, and an average recognition rate of $89.38 \%$ $89.38 %$ in four-class experiments, improving by $2.96 \%$ $2.96 %$ , $3.31 \%$ $3.31 %$ , and $7.69 \%$ $7.69 %$ , respectively, compared to existing 1DCNN models.