A hyperspectral remote sensing image ground object classification method

CN122289760APending Publication Date: 2026-06-26SHAN DONG HUI JIE DI XIN KE JI YOU XIAN GONG SI

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: SHAN DONG HUI JIE DI XIN KE JI YOU XIAN GONG SI
Filing Date: 2026-03-17
Publication Date: 2026-06-26

AI Technical Summary

Technical Problem

Existing hyperspectral remote sensing image land cover classification methods suffer from connectivity distortion and difficulty in achieving accurate classification when dealing with real-world scenes with uneven land cover distribution density due to fixed-scale neighborhood selection strategies.

Method used

By acquiring local spectral density and neighborhood scale, multiple graph groups are constructed and sparsely fused. Combining geodesic distance matrix groups and manifold embedding feature matrices, two-layer clustering and residual analysis are used to construct a skeleton graph and perform anchor point cyclic updates. Finally, classification is performed based on a four-layer refinement mechanism.

Benefits of technology

It achieves more comprehensive hyperspectral remote sensing image land cover classification, improves the accuracy and reliability of classification, can identify high-confidence samples and refine difficult-to-classify samples layer by layer, and generate reliable skeleton maps.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122289760A_ABST

Patent Text Reader

Abstract

This invention provides a method for classifying ground features in hyperspectral remote sensing images. The method includes obtaining local spectral density and neighborhood scale from the original hyperspectral image; constructing multiple map groups based on the local spectral density and neighborhood scale; obtaining a sparse fused map based on a fusion strategy; obtaining a geodesic distance matrix group based on the sparse fused map; obtaining a manifold embedding feature matrix through a manifold embedding strategy; employing two-layer clustering to obtain initial clustering results and membership vectors; obtaining five-dimensional residuals and one-dimensional comprehensive residuals; performing residual clustering and adaptive partitioning to obtain residual recognition results and partitioned sample sets; constructing a skeleton map using the sparse fused map and partitioned sample sets; and iteratively updating the anchor points of the skeleton map to obtain the final skeleton map; and processing the partitioned sample sets based on a four-layer refinement mechanism to obtain the final classification result. This provides a method for more refined classification of ground features in hyperspectral remote sensing images.

Need to check novelty before this filing date? Find Prior Art