Classroom interaction is an important indicator for evaluating teaching quality. It automatically extracts educational big data on the emotional state and changes of teachers and students in classroom interaction, thereby gaining insight into students' personality traits, learning interests, and mental health status, as well as teachers' interaction awareness, skills, and affinity performance. It achieves non cognitive level classroom teaching and interaction evaluation, and breaks through the limitations of traditional listening evaluation that are difficult to scale and daily. This project aims to use high-definition audio and video big data in the classroom to conduct three research projects: (1) Overcoming the limitations of existing technologies in single modality, single subject, and short duration, proposing a multimodal One Stage classroom interaction detection method suitable for long time series and multiple subjects using LSTM based on Attention mechanism; (2) To compensate for the shortcomings of discrete emotion models in expressing emotional intensity and continuous emotions, as well as existing methods in modeling emotional propagation, a multi-agent continuous emotion recognition network for classroom interaction is constructed using graph networks Improve the shortcomings of existing methods in static time constrained semantic information extraction and single-layer encoding expression, and propose a video description method using prior knowledge for multi granularity semantic extraction and multi-layer encoding expression in elastic and dynamic time scales. This project will provide intelligent technology for automated classroom interactive evaluation.

Developmental dyslexia (DD) is prevalent in the children population with an estimated prevalence of 5% to 10%, it not only undermines individual development but weakens the competence of the national population. The current study focuses on the fusiform gyrus and adopts a three-group design which includes the DD group, age-control group, and reading-control group. With the adoption of fMRI and representative similarity analysis (RSA), qMRI measurements, and surface-based approaches to explore the impairments of the gyrification, the morphometry of the cerebral cortex and the cytoarchitecture of the fusiform gyrus. The study attempts to reveal the etiology of the atypical neural representations of orthographical and phonological information of fusiform gyrus in DD, and the mechanisms underlying the emergence of the atypical neural representations. The findings will shed light on the early identification, precision diagnosis, and efficient intervention for Chinese developmental dyslexic children.

Nonsuicidal self-injury (Nonsuicidal self-injury, NSSI) refers to the behavior that individuals intentionally and directly injure their own body tissues without suicidal intentions, and are not recognized by society and culture. Timely and effective prevention and intervention of NSSI behaviors among adolescents is needed to reduce the harm caused by NSSI behaviors to individuals, families, schools and society. Therefore, exploring the mechanism of adolescent NSSI behavior and providing health management measures has extremely important scientific significance and practical value for the mental health of adolescents in my country. The ecological instantaneous assessment method can capture the emotional state and physiological state of the individual in time and give feedback in time. It can obtain real-time data of the individual state at low cost, high efficiency and accuracy, and provides a new direction for the intervention management of NSSI behavior.