Machine learning based near real-time satellite retrieval precipitation correction method and system
By using a lightweight correction network based on machine learning and spectral fidelity assessment, a quality mask matrix is generated, which solves the problems of high resource consumption and insufficient accuracy in satellite precipitation correction technology, and achieves near real-time, efficient and seamless correction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIAN XINGTUZHIHUA DIGITAL TECH CO LTD
- Filing Date
- 2026-03-03
- Publication Date
- 2026-05-29
AI Technical Summary
Existing satellite-based precipitation correction technology lacks a lightweight network architecture, resulting in high resource consumption, slow processing speed, and inability to adapt to near real-time processing scenarios. Furthermore, the spectral fidelity assessment of the inverted data is not comprehensive enough, making it difficult to form an accurate quality mask matrix, which affects the reliability of data applications.
A lightweight correction network based on machine learning is used to reshape the tensor and reconstruct the channel dimension of the inverted precipitation data in the satellite inversion section, generate network-adaptive feature primitives, and generate a quality mask matrix through spectral fidelity evaluation. Combined with spatiotemporal kriging interpolation and edge smoothing fusion, a seamless correction field is generated.
It significantly improves the correction accuracy and application value of satellite-retrieved precipitation data, meets near real-time processing requirements, and ensures data integrity, consistency, adaptability, and efficiency.
Smart Images
Figure CN122114230A_ABST