Multi-view domain adaption based multi-scale convolutional conditional invertible discriminator for cross-subject electroencephalogram emotion recognition

作     者：Babu, S. Sivasaravana Venkatesan, Prabhu Velusamy, Parthasarathy Ganesan, Saravana Kumar 

作者机构：Vel Tech High Tech Dr Rangarajan Dr Sakunthala Eng Dept Elect & Commun Engn Chennai 600062 Tamil Nadu India Vel Tech Multi Tech Dr Rangarajan Dr Sakunthala En Dept Elect & Commun Engn Chennai 600062 Tamil Nadu India Karpagam Acad Higher Educ Dept Comp Sci & Engn Coimbatore 641021 Tamil Nadu India Karpagam Coll Engn Dept Elect & Commun Engn Coimbatore 641032 Tamil Nadu India 

出 版 物：《COGNITIVE NEURODYNAMICS》 (Cogn. Neurodynamics)

年 卷 期：2025年第19卷第1期

页      面：1-19页

核心收录：

学科分类：0710[理学-生物学] 1001[医学-基础医学(可授医学、理学学位)] 07[理学] 071003[理学-生理学] 

主　　题：Electroencephalogram Emotion recognition Geodesic flow kernel Multi-scale convolutional neural network Puma optimizer 

摘      要：Cross subject Electroencephalogram (EEG) emotion recognition refers to the process of utilizing electroencephalogram signals to recognize and classify emotions across different individuals. It tracks neural electrical patterns, and by analyzing these signals, it s possible to infer a person s emotional state. The objective of cross-subject recognition is to create models or algorithms that can reliably detect emotions in both the same person and several other people. Accurately predicting emotions poses challenges due to dynamic traits. Models struggle with feature extraction, convergence, and negative transfer issues, hindering cross subject emotion recognition. The proposed model employs thorough signal preprocessing, Short-Time Geodesic Flow Kernel Fourier Transform (STGFKFT) for feature extraction, enhancing classifiers accuracy. Multi-view sheaf attention improves feature discrimination, while the Multi-Scale Convolutional Conditional Invertible Puma Discriminator Neural Network (MSCCIPDNN) framework ensures generalization. Efficient computational techniques and the puma optimization algorithm enhance model robustness and convergence. The suggested framework demonstrates extraordinary success with high accuracy, of 99.5%, 99% and 99.50% for SEED, SEED-IV, and DEAP dataset sequentially. By incorporating these techniques, the proposed method aims to precisely recognition emotions, and accurately captures the features, thereby overcoming the limitations of existing methodologies.