Multi-modal ai abnormal request detection method and system for rural digitization
By extracting multimodal sequences and environmental fingerprints at edge detection nodes, and combining them with a modality consistency discriminator and dynamic threshold judgment, the problems of poor real-time performance and high false positive rate of multimodal AI requests in rural environments are solved, achieving efficient and accurate abnormal request detection under resource-constrained conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGAN VOCATIONAL & TECH COLLEGE
- Filing Date
- 2026-05-07
- Publication Date
- 2026-07-17
AI Technical Summary
In rural environments, multimodal AI requests suffer from poor real-time performance and a high rate of false positives. Existing technologies cannot effectively address the issues of real-time response and security detection under resource-constrained conditions.
The edge detection node receives the original multimodal AI request stream, extracts the native multimodal sequence and contextual fingerprint, calculates the internal modal consistency confidence and environment adaptation deviation through a pre-trained modal consistency discriminator, generates the edge anomaly suspicion index, and combines dynamic thresholds to judge the feature distillation and compression processing of the request stream, and synchronizes it to the cloud collaborative center for source tracing analysis to generate the final anomaly identification result.
It achieves real-time response and accurate detection of multimodal AI requests under resource-constrained conditions, reduces service latency, and significantly improves the balance between detection accuracy and response speed.
Smart Images
Figure CN122419902A_ABST