Multimodal visual health risk assessment enhanced by temporal features and spatial attention

The rapid advancement of artificial intelligence has provided convenient technical approaches for visual health risk assessment. Traditional health risk assessment methods fail to consider the comprehensive utilization of data's temporal characteristics and spatial information, resulting in insufficient accuracy and reliability of assessment outcomes. To address this, this study proposes a multimodal visual health risk assessment framework that integrates temporal features with spatial attention enhancement. A bidirectional long short-term memory (LSTM) network is employed to extract temporal dynamic features from facial video sequences, capturing age information from the subject's facial images. Experiments on the UTKFace dataset demonstrate that the model achieves 79.61% accuracy in age recognition tasks, reducing the mean absolute error by approximately 3.6 years. The system consistently generates personalized screen time recommendations, enabling quantitative assessment of eye fatigue risk. This visual health risk assessment system employs non-contact facial analysis to provide tailored eye protection strategies for test subjects across different age groups, offering a viable technical pathway for intelligent interventions in public health.