一种基于多模态图像预筛的电厂安全隐患识别方法
By using a cross-modal image pre-screening method, and leveraging the saliency mask of the infrared residual image and the gradient collinearity coefficient to dynamically adjust the trigger threshold, the problem of unreasonable allocation of computing resources in the power plant monitoring system is solved, and efficient and stable hazard identification is achieved.
Patent Information
- Application Number
- CN202610500871.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-17
- Estimated Expiration
- 2046-04-16
AI Technical Summary
Existing power plant monitoring systems have unreasonable allocation of computing resources in multimodal image recognition, resulting in an imbalance between computational load and recognition accuracy. This makes them unable to effectively identify minor hidden dangers and lacks stability under complex operating conditions.
By employing a cross-modal image pre-screening method, a saliency mask is constructed using infrared residual maps. The collinearity coefficients of visible light and infrared image gradients are calculated. Combined with the sliding window statistical signal-to-noise ratio, the trigger threshold is dynamically adjusted to achieve on-demand allocation of computing resources and accurate identification of subtle hidden dangers.
It improves the response efficiency and judgment stability of power plant safety hazard identification, reduces the computational load of edge-side equipment, reduces the false alarm rate, and improves identification accuracy.
Smart Images
Figure CN122049820B_ABST
Abstract
Citation Information
Patent Citations
Image fusion method, device and equipment, and storage medium
CN114519808A
High-temperature high-brightness production environment monitoring system based on deep learning
CN121214047A