Scholarlink

<p>Human emotions play a vital role in determining our thoughts, behaviors, and even how we interact with others. In fields like social robotics, mental illness diagnosis, and human–computer interaction, the ability to recognize emotions is essential. However, the existing emotion detection algorithms face challenges such as noise interference, poor feature extraction, and data integration in multimodal contexts that include text, video, and audio. To tackle these problems, this study suggests an “Enhanced Trimodal Emotion Recognition Using Steerable Graph Neural Network with Crayfish Optimization Algorithm.” The suggested approach starts with pre-processing tailored to each modality: text pre-processing refines textual data by removing stop words, stemming/lemmatizing, tokenizing, and normalizing. Audio quality is improved by adaptive frequency-domain filtering, and video data are refined by an Iterative Self-Guided Image Filter (isGIF). Channel-wise Temporal Attention Network (CTAN) for video, Dynamic Memristor-Based Time-Surface Neurons for audio, and Fast Point Transformer for text are used for feature extraction. In the context of data fusion, an Attention-Guided Multiscale Recursive Fusion technique preserves important emotional cues while combining modalities. To ensure high accuracy and efficiency, a Steerable Graph Neural Network with Crayfish Optimization Algorithm (SGNN-COA) is used for classification. The accuracy of the recommended approach for Multimodal Emotion Recognition (MER) exceeds 99.60% on the IEMOCAP dataset and 99.65% on the MELD dataset, demonstrating continuously outstanding performance. These outcomes show off its exceptional capacity to successfully integrate and analyze text, audio, and video input to categorize emotional states with high accuracy.</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 trimodal emotion recognition using Steerable Graph Neural Network with Crayfish Optimization Algorithm

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Devavarapu Sreenivasarao,
  • Shaik Khasim Saheb

摘要

Human emotions play a vital role in determining our thoughts, behaviors, and even how we interact with others. In fields like social robotics, mental illness diagnosis, and human–computer interaction, the ability to recognize emotions is essential. However, the existing emotion detection algorithms face challenges such as noise interference, poor feature extraction, and data integration in multimodal contexts that include text, video, and audio. To tackle these problems, this study suggests an “Enhanced Trimodal Emotion Recognition Using Steerable Graph Neural Network with Crayfish Optimization Algorithm.” The suggested approach starts with pre-processing tailored to each modality: text pre-processing refines textual data by removing stop words, stemming/lemmatizing, tokenizing, and normalizing. Audio quality is improved by adaptive frequency-domain filtering, and video data are refined by an Iterative Self-Guided Image Filter (isGIF). Channel-wise Temporal Attention Network (CTAN) for video, Dynamic Memristor-Based Time-Surface Neurons for audio, and Fast Point Transformer for text are used for feature extraction. In the context of data fusion, an Attention-Guided Multiscale Recursive Fusion technique preserves important emotional cues while combining modalities. To ensure high accuracy and efficiency, a Steerable Graph Neural Network with Crayfish Optimization Algorithm (SGNN-COA) is used for classification. The accuracy of the recommended approach for Multimodal Emotion Recognition (MER) exceeds 99.60% on the IEMOCAP dataset and 99.65% on the MELD dataset, demonstrating continuously outstanding performance. These outcomes show off its exceptional capacity to successfully integrate and analyze text, audio, and video input to categorize emotional states with high accuracy.