Depression Micro-Expression Recognition Technology Based on Multimodal Knowledge Graphs

According to authoritative surveys, 24.6% of contemporary college students experience varying degrees of mental health issues, with an annual increase of 1-3%. ASevere depression, in particular, can lead to campus crises. Research by experts has demonstrated that micro- expression recognition plays a significant predictive role in depression and holds considerable clinical value. This study first collects multimodal data from conversations between students and psychological counselors using professional equipment, including speech, video, and psychological scale data, to construct a multimodal psychological dataset for college students. The study utilizes a Kinect camera to convert speech into text for analysis and performs micro-expression analysis on video images. Addressing the limitations of traditional expression recognition methods in capturing subtle micro- expressions, this paper proposes a micro-expression recognition model based on a Convolutional Neural Network (CNN)+ Graph Convolutional Network (GCN) transfer learning network. Leveraging the unique advantage of GCNs in automatically updating node information, the model captures the dependencies between image data and corresponding emotional labels in micro-expression sequences. The network model is pre-trained on the CAS(ME)3 3 dataset to obtain initial parameters, followed by transfer learning to retrain the model for application to the college students' multimodal psychological dataset, ultimately producing representation vectors of micro-expressions. By correlating these representation vectors with various emotional categories, a multimodal knowledge graph based on video, speech, and psychological scale data is constructed. Experimental comparisons demonstrate that the proposed model effectively enhances micro-expression recognition performance and accurately identifies students' depressive states when combined with the multimodal