A hyperspectral image edge detection method and system
By combining ternary physical masks and wavelength-driven feature mapping with context tracking loss, the problems of cross-sensor generalization and edge coarsening in hyperspectral image edge detection are solved, achieving efficient and accurate end-to-end edge detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-06-17
- Publication Date
- 2026-07-17
AI Technical Summary
Existing hyperspectral image edge detection technologies suffer from modal gaps, neglect of mixed pixel characteristics, and reliance on nonmaximum suppression post-processing, resulting in poor cross-sensor generalization ability, label distortion, and rough output edges.
We employ a ternary physical mask construction, wavelength-driven feature mapping, and context tracking loss mechanism. We generate an adaptive transition band through spectral spatial gradient maps and morphological dilation, and combine a lightweight multilayer perceptron and a deep network for end-to-end edge detection.
It achieves seamless generalization across sensors, improves the accuracy of edge detection and end-to-end inference efficiency, and reduces computational complexity and real-time requirements.
Smart Images
Figure CN122415664A_ABST