Scholarlink

<p>The brain's nervous system produces Electroencephalogram (EEG) impulses, which are hard to conceal and useful for identifying emotions. Analysing EEG is complex since it is typically irregular and exhibits poor performance to noise. Current research has enabled with unstructured EEG information in training models for partially supervised learning has shown encouraging results in emotional state classification. Furthermore, the correlation-aided loss function with Concordance Correlation Coefficient (CCC) performs better compared to error-enabled loss functions. The conventional emotion detection algorithm's error and misclassification rate lead to slight differences in the recognition of emotions. This work concentrates on building a hybrid deep learning platform to address the shortcomings of lower accuracy in conventional emotion recognition systems. The input EEG signals are formulated as spectrogram images, from which spectral, temporal, and spatial features are extracted through a weighted feature fusion phase. Here, the weights involved in feature fusion are processed for optimal value through the developed Enhanced Waterwheel Plant Algorithm (EWPA). The developed Atrous Convolution-based One-Dimensional Convolutional Neural Network with Deep Bayesian Learning (AC-1DCNN-DBL) processes the weighted feature set and recognizes different emotions hidden in the signal. Hybrid deep learning can observe a high degree of similarity among the EEG signals by combining two different recognized outputs to make the technique strong and also reduce the misclassification rate. Further, several experimental validations are performed in the developed framework over the classical emotion recognition models to display the effective performance based on the misclassification rate. Extensive experiments have been held to prove the capability of the technique, and it has achieved values of 98.8% and 97.9% for accuracy and precision under the DEAP dataset, indicating a robust model for complex EEG emotion recognition applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced Emotion Recognition Framework Aware of Misclassification Rate Using Atrous Convolution-Based 1DCNN with Deep Bayesian Learning

  • 期刊论文

  • 发表日期：

    2026年1月14日

查看全文
  • C. H. Narsimha Reddy,
  • K. S. Ananda Kumar,
  • K. Manjunathachari,
  • Arif Mohammad Abdul

摘要

The brain's nervous system produces Electroencephalogram (EEG) impulses, which are hard to conceal and useful for identifying emotions. Analysing EEG is complex since it is typically irregular and exhibits poor performance to noise. Current research has enabled with unstructured EEG information in training models for partially supervised learning has shown encouraging results in emotional state classification. Furthermore, the correlation-aided loss function with Concordance Correlation Coefficient (CCC) performs better compared to error-enabled loss functions. The conventional emotion detection algorithm's error and misclassification rate lead to slight differences in the recognition of emotions. This work concentrates on building a hybrid deep learning platform to address the shortcomings of lower accuracy in conventional emotion recognition systems. The input EEG signals are formulated as spectrogram images, from which spectral, temporal, and spatial features are extracted through a weighted feature fusion phase. Here, the weights involved in feature fusion are processed for optimal value through the developed Enhanced Waterwheel Plant Algorithm (EWPA). The developed Atrous Convolution-based One-Dimensional Convolutional Neural Network with Deep Bayesian Learning (AC-1DCNN-DBL) processes the weighted feature set and recognizes different emotions hidden in the signal. Hybrid deep learning can observe a high degree of similarity among the EEG signals by combining two different recognized outputs to make the technique strong and also reduce the misclassification rate. Further, several experimental validations are performed in the developed framework over the classical emotion recognition models to display the effective performance based on the misclassification rate. Extensive experiments have been held to prove the capability of the technique, and it has achieved values of 98.8% and 97.9% for accuracy and precision under the DEAP dataset, indicating a robust model for complex EEG emotion recognition applications.