Article: Faster R-BERT multimodal fusion real-time psychological stress recognition system Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.57 pp.99 - 118 Abstract: The psychological stress problems among college students are increasingly prominent, requiring efficient and objective identification methods. However, existing real-time systems struggle to balance accuracy with processing speed and lack deep integration of multi-source information (such as expressions, voices, and texts). This study proposes a real-time recognition system based on faster robust bidirectional encoder representations from transformers multimodal fusion, significantly improving computing efficiency through an innovative lightweight fusion mechanism. Experiments on public datasets show the system achieves 86.5% accuracy in stress recognition, significantly improving on traditional methods (e.g., 73.2% for single-modal convolutional neural network). Its inference speed meets real-time requirements (30 fps), with the key area under the curve indicator increasing to 0.91 (from 0.82). This study provides an effective approach for non-intrusive, real-time psychological state monitoring in campus environments. Inderscience Publishers - linking academia, business and industry through research

Title: Faster R-BERT multimodal fusion real-time psychological stress recognition system

Authors: Ming Zhang

Addresses: School of Marxism Studies, Chengdu Polytechnic, Chengdu, 610041, China

Abstract: The psychological stress problems among college students are increasingly prominent, requiring efficient and objective identification methods. However, existing real-time systems struggle to balance accuracy with processing speed and lack deep integration of multi-source information (such as expressions, voices, and texts). This study proposes a real-time recognition system based on faster robust bidirectional encoder representations from transformers multimodal fusion, significantly improving computing efficiency through an innovative lightweight fusion mechanism. Experiments on public datasets show the system achieves 86.5% accuracy in stress recognition, significantly improving on traditional methods (e.g., 73.2% for single-modal convolutional neural network). Its inference speed meets real-time requirements (30 fps), with the key area under the curve indicator increasing to 0.91 (from 0.82). This study provides an effective approach for non-intrusive, real-time psychological state monitoring in campus environments.

Keywords: psychological stress; multimodal fusion; real-time system; mental health.

DOI: 10.1504/IJICT.2026.153799

International Journal of Information and Communication Technology, 2026 Vol.27 No.57, pp.99 - 118

Received: 06 Jan 2026
Accepted: 07 Feb 2026
Published online: 26 May 2026 *

Title: Faster R-BERT multimodal fusion real-time psychological stress recognition system

Keep up-to-date