RARN: Lightweight Deep Residual Learning with Attention for Human Emotions Recognition

Zhenyuan Zhu

2024

Abstract

Human emotion identification represents a formidable challenge within computer vision research. This study endeavours to classify human emotions across seven discrete categories: anger, disgust, fear, happiness, neutral, sadness, and surprise. To address this challenge, this paper introduces the Rotation-Aware Residual Network (RARN), a novel framework leveraging convolutional neural networks (CNNs) and spatial attention mechanisms. Notably, this approach is designed to excel in accurately discerning facial emotions amidst complex real-world contexts. Experimental validation conducted on the FER-2013 Dataset underscores the efficacy of our proposed model, demonstrating notable improvements in emotion recognition accuracy. Crucially, the Rotation-Aware Residual Network's innovative integration of multi-scale fusion and angle-sensitive spatial attention modules underscores its unique capacity to capture nuanced facial expressions. This breakthrough has significant implications for diverse applications, including human-computer interaction, psychological health assessment, and social signal processing. Moving forward, future research endeavours will focus on further refining the network architecture and expanding the diversity of datasets to enhance the model's performance across various scenarios.

Paper Citation

in Harvard Style

Zhu Z. (2024). RARN: Lightweight Deep Residual Learning with Attention for Human Emotions Recognition. In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI; ISBN 978-989-758-713-9, SciTePress, pages 127-134. DOI: 10.5220/0012911100004508

in Bibtex Style

@conference{emiti24,
author={Zhenyuan Zhu},
title={RARN: Lightweight Deep Residual Learning with Attention for Human Emotions Recognition},
booktitle={Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI},
year={2024},
pages={127-134},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012911100004508},
isbn={978-989-758-713-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI
TI - RARN: Lightweight Deep Residual Learning with Attention for Human Emotions Recognition
SN - 978-989-758-713-9
AU - Zhu Z.
PY - 2024
SP - 127
EP - 134
DO - 10.5220/0012911100004508
PB - SciTePress