Hierarchical superpixel multi-expert Mama network construction method for hyperspectral image classification

By using a hierarchical superpixel multi-expert Mamba network, the problems of insufficient long-distance spatial-spectral dependence and inadequate characterization of local fine-grained changes in hyperspectral image classification are solved. This enables collaborative modeling of multi-scale features, improving classification accuracy and robustness.

CN122452633APending Publication Date: 2026-07-24HUZHOU UNIVERSITY
0 Cites 0 Cited by

Patent Information

Application Number
CN202610863354.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-06-15
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing hyperspectral image classification methods suffer from insufficient modeling of long-distance spatial-spectral dependencies, inadequate characterization of local fine-grained changes, and difficulty in effectively coordinating hierarchical multi-scale features, resulting in inadequate classification performance in complex scenes.

Method used

We propose a Hierarchical Superpixel Multi-Expert Mamba Network (HSMMamba), which combines a learnable hierarchical superpixel mapping module, a hierarchical superpixel multi-expert hybrid module, and a forward-backward-differential Mamba module to achieve adaptive optimization of cross-layer mapping relationships, differentiated modeling of multi-scale features, and collaborative modeling of long-range dependencies and local details.

Benefits of technology

It improves the accuracy and robustness of hyperspectral image classification, especially demonstrating superior generalization and robustness in complex scenes, and significantly improves classification performance.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a hierarchical super-pixel multi-expert Mama network construction method for hyperspectral image classification, belongs to the field of hyperspectral image classification, and particularly relates to the hierarchical super-pixel multi-expert Mama network construction method for hyperspectral image classification. The method aims at solving the problems that long-distance spatial spectrum dependence modeling is insufficient, local fine-grained change description is insufficient, hierarchical multi-scale features are difficult to effectively cooperate and the like when existing hyperspectral image classification is confronted, and aims at significant differences of different scale features in expression ability and semantic emphasis. And the problems that long-range spatial spectrum dependency relationship modeling in the hyperspectral image is insufficient and local boundary and fine grit change information is easy to ignore are solved. The method comprises the following steps: constructing a hierarchical superpixel multi-expert Mama network model for hyperspectral image classification; the model sequentially comprises an LHSM, an HSMoE, an FBDMamba and a classification layer; and classifying the to-be-measured hyperspectral image, and outputting a classification result.
Need to check novelty before this filing date? Find Prior Art