Human-Centered Multimodal Affective-Care Product-Design Method Integrating YOLOv8-CBAM

The growing aging population highlights the need for intelligent systems that address emotional as well as physical care. However, most facial expression recognition (FER) methods treat emotion as an isolated classification task and provide limited support for care-oriented interaction. This study proposes a human-centered affective care framework that integrates elderly FER with context-aware and design-driven responses. An enhanced YOLOv8-CBAM model is developed to improve sensitivity to subtle and low-contrast facial expressions in older adults, forming a stable perception layer. On this basis, a perception–mapping–response (PMR) mechanism converts recognized emotions into interpretable action tokens that guide value-aligned multimodal feedback. A new elderly facial expression dataset, AFE-8, comprising 5,735 images across eight emotion categories, is constructed for evaluation. Results under a fixed held-out test protocol demonstrate stable recognition performance, while system-level and expert-based evaluations suggest improved interaction coherence compared with conventional FER-centered pipelines. Overall, the proposed framework extends elderly FER beyond classification toward emotionally aware and human-centered care systems.