Scholarlink

<p>Emotion recognition is the process of detecting the feelings and emotions in humans. The emotion recognition process provides high accuracy in the first attempt when recognizing emotions in normal people. Emotion recognition becomes particularly challenging in children with Autism Spectrum Disorder due to difficulties with speech and social communication, highlighting the need for a robust and effective emotion recognition method. Hence, this research proposes a novel multi-modal temporal autoencoder-based cross-transformer method to recognize emotions in autistic children. For gathering the input electrocardiogram and Galvanic Skin Response data, this research utilizes the electrocardiogram and Galvanic Skin Response signals dataset. Initially, these input data are preprocessed in different preprocessing phases, such as noise reduction, normalization, and augmentation, in order to improve the generalization ability and data quality. Then, the Selective Kernel-based Temporal Autoencoder model is utilized to extract the temporal features from the preprocessed electrocardiogram and Galvanic Skin Response data. After that, the multi-modal cross attention-based transformer acquired the multi-modal emotional intermediate representations through the electrocardiogram and Galvanic Skin Response data’s temporal features. Then, the multimodal fusion is applied in terms of the EmbraceNet to acquire the shared feature representations for electrocardiogram and Galvanic Skin Response data. The EmbraceNet comprises the docking layer and the embracement layer. The docking layer offers the feature vectors, and these vectors are combined with a single vector in the embracement layer. Then, the electrocardiogram and Galvanic Skin Response modalities’ all features are combined, and final multi-modal emotional representations are created. Finally, these emotional representations are classified, and a multi-class output is obtained. The experimental evaluation for the proposed method is carried out on the basis of various evaluation measures. The results demonstrated that the proposed method outperformed the state-of-the-art approaches and gained 98.93% recognition accuracy and 2.7&#xa0;s for computation time.</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

Systematic Approach to Emotion Recognition in Autistic Children Using MTAE-CTrans

  • 期刊论文

  • 发表日期：

    2025年11月12日

查看全文
  • Soshya Joshi,
  • L. N. B. Srinivas

摘要

Emotion recognition is the process of detecting the feelings and emotions in humans. The emotion recognition process provides high accuracy in the first attempt when recognizing emotions in normal people. Emotion recognition becomes particularly challenging in children with Autism Spectrum Disorder due to difficulties with speech and social communication, highlighting the need for a robust and effective emotion recognition method. Hence, this research proposes a novel multi-modal temporal autoencoder-based cross-transformer method to recognize emotions in autistic children. For gathering the input electrocardiogram and Galvanic Skin Response data, this research utilizes the electrocardiogram and Galvanic Skin Response signals dataset. Initially, these input data are preprocessed in different preprocessing phases, such as noise reduction, normalization, and augmentation, in order to improve the generalization ability and data quality. Then, the Selective Kernel-based Temporal Autoencoder model is utilized to extract the temporal features from the preprocessed electrocardiogram and Galvanic Skin Response data. After that, the multi-modal cross attention-based transformer acquired the multi-modal emotional intermediate representations through the electrocardiogram and Galvanic Skin Response data’s temporal features. Then, the multimodal fusion is applied in terms of the EmbraceNet to acquire the shared feature representations for electrocardiogram and Galvanic Skin Response data. The EmbraceNet comprises the docking layer and the embracement layer. The docking layer offers the feature vectors, and these vectors are combined with a single vector in the embracement layer. Then, the electrocardiogram and Galvanic Skin Response modalities’ all features are combined, and final multi-modal emotional representations are created. Finally, these emotional representations are classified, and a multi-class output is obtained. The experimental evaluation for the proposed method is carried out on the basis of various evaluation measures. The results demonstrated that the proposed method outperformed the state-of-the-art approaches and gained 98.93% recognition accuracy and 2.7 s for computation time.