A multi-channel satellite cloud image convection initial segmentation method based on a segmentation-all model

By improving the SAM2 model for segmenting everything, a multi-channel satellite cloud image convection initiation segmentation method is constructed, which solves the problems of accuracy and reliability in detecting convection initiation targets in satellite cloud images, and realizes efficient convection initiation region segmentation and early warning.

CN122415648APending Publication Date: 2026-07-17NANJING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-24
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing technologies are insufficient to effectively detect nascent convective targets at extremely small scales in satellite cloud images, and traditional methods are prone to misjudgment in complex cloud backgrounds. Furthermore, the lack of specificity in multispectral information fusion leads to insufficient detection accuracy and reliability.

Method used

Based on the SAM2 model for segmenting everything, a multi-channel satellite cloud image convection initial segmentation method is constructed. By using a spectral grouping cross-attention module, a physical perception adaptation module, and a high-resolution feature connection module, combined with end-to-end training and a combined loss function, the segmentation boundary is optimized and the sample imbalance problem is alleviated.

Benefits of technology

It significantly improves the detection accuracy and computational efficiency of the initial convection region, provides reliable technical support for early warning of severe convective weather, reduces the false alarm rate, and improves the accuracy of detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415648A_ABST
    Figure CN122415648A_ABST
Patent Text Reader

Abstract

本发明公开了一种基于分割一切模型的多通道卫星云图对流初生分割方法,包括以下步骤:步骤 1,获取卫星云图图像数据与对流初生标签数据,构建图像‑标签数据对,在此基础上构建输入样本;步骤 2,基于预训练的SAM2模型构建主干模型,构建多通道卫星云图对流初生分割模型;步骤 3,采用端到端的训练方式,将预处理后的训练集图像数据输入到多通道卫星云图对流初生分割模型;步骤4,将训练完成的模型应用于所述数据集的测试集,实现对卫星图像对流初生的精准分割。针对性地解决了多光谱通道异质性、通用视觉特征与气象物理先验的语义鸿沟以及极小目标空间细节丢失等核心问题,整个方法体系高效而实用。
Need to check novelty before this filing date? Find Prior Art