Visual emotion analysis method and system based on frequency domain enhancement and multi-attribute reasoning
By constructing an attribute node matrix and performing spectral modulation and collaborative perception, combined with graph convolutional layers for multi-layer graph reasoning, the problem of separating global emotional atmosphere from local details in visual sentiment analysis is solved, thereby improving the robustness and discriminative power of sentiment analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-16
- Publication Date
- 2026-07-17
AI Technical Summary
Existing visual sentiment analysis methods struggle to effectively separate global emotional atmosphere from local details in the spatial domain, and multi-attribute reasoning models lack local collaborative relationship modeling and attribute contribution adjustment mechanisms, resulting in limited ability to represent global emotional cues.
By constructing an attribute node matrix, using a spectrum modulation module for frequency domain transformation and modulation, combining an attribute collaborative perception module to extract interaction features of adjacent attribute nodes, adjusting contribution through an attribute attention module, and combining graph convolutional layers for multi-layer graph inference, we can achieve intra-attribute frequency selection and inter-attribute collaborative modeling.
It improves the representation ability of multi-attribute features, enhances the sentiment analysis performance in complex scenarios, solves the problem of global sentiment cue separation, and improves the robustness and discriminative power of sentiment analysis.
Smart Images
Figure CN122416163A_ABST