Video depression assessment method and system based on multi-modal graph and collaborative state space

CN122290962APending Publication Date: 2026-06-26HUNAN UNIV

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: HUNAN UNIV
Filing Date: 2026-03-29
Publication Date: 2026-06-26

AI Technical Summary

Technical Problem

Existing technologies suffer from insufficient intermodal interaction, low efficiency in long-term modeling, and lack of time-series dynamic collaboration in video depression assessment, resulting in limited feature representation capabilities and low diagnostic efficiency.

Method used

We construct a multimodal graph and a collaborative state-space model, enhance cross-modal features through the multimodal joint graph, perform temporal modeling using a collaborative bidirectional Mamba2 state-space model, and predict depression scores through a dynamic gating fusion mechanism.

Benefits of technology

It achieves high-precision and high-efficiency long-term depression assessment, improves the robustness and generalization ability of the model, and is significantly better than existing methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122290962A_ABST

Patent Text Reader

Abstract

This invention discloses a video depression assessment method and system based on multimodal graphs and co-state spaces in the field of data processing technology. The method includes: acquiring multimodal depression assessment data containing video and audio; extracting visual and audio features; constructing a multimodal joint graph; and enhancing cross-modal features through a graph message passing mechanism. It also involves combining a co-operative bidirectional Mamba2 state space model with a shared state transition matrix to achieve cross-modal temporal collaborative modeling; fusing visual and audio features through a dynamic gating mechanism to output a predicted depression score; and performing a visual assessment using a heatmap and a prediction comparison map. This invention achieves cross-modal structured information interaction through a multimodal matrix graph neural network and utilizes a co-state space model to efficiently model long-term temporal dependencies with linear complexity, achieving high-precision video depression assessment while maintaining computational efficiency.

Need to check novelty before this filing date? Find Prior Art