EmoSense: AFRN-RAG Framework for FER-Based Facial Emotion Analysis and Mental Health Support

Read full text Summarise with VIRAT
Authors: Peiang Miao, S Li, Cai, Keru, Yuxin Tian, Riju Das, Ifran Rahman Nijhum, Soumyabrata Dev
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Mental health monitoring is becoming increasingly important with the rising prevalence of psychological disorders, yet traditional assessment methods remain subjective and resource-intensive. This paper presents a hybrid framework that combines an Adaptive Feature Recalibration Network (AFRN) for facial emotion recognition with RetrievalAugmented Generation (RAG) for mental health–related feedback. The AFRN architecture incorporates two key components: adaptive spectral recalibration modules that enhance emotion-relevant features through learned attention mechanisms, and a progressive multi-scale fusion scheme that captures both fine-grained local patterns and global facial configurations. On top of the FER backbone, the RAG component leverages domain-specific psychological resources to generate contextually informed textual summaries and guidance based on detected emotion trajectories. Experimental validation on a standard facial emotion recognition dataset demonstrates that AFRN achieves improved classification performance compared to a baseline CNN, while maintaining computational efficiency suitable for real-time or near real-time applications. The overall EmoSense framework illustrates a feasible pathway from automated emotion detection to psychologically grounded, human-readable feedback, and is positioned as an exploratory tool for AI-assisted mental health monitoring and clinical decision support rather than a standalone diagnostic or therapeutic system.

Fields of Study & Subjects

Emotion and Mood Recognition Face recognition and analysis Face and Expression Recognition

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
21 March 2026
Language
EN