Bidirectional Spatial-Spectral Learning for Efficient Hyperspectral Image Analysis

Abstract

Hyperspectral Image Learning now turns on the ability to balance performance with evidence quality, resource limits, and transfer across settings. The review develops an evidence-centered account of balancing spectral discrimination, spatial context, nonlinearity, and computational cost. The corpus joins 1 focal paper with 12 independently retrieved publications whose authorship and venue fields were independently checked. The analysis is organized around spectral mixing, spatial context, bidirectional fusion, label scarcity, and sensor transfer. Rather than treating reported gains as context-free quantities, the review compares units of analysis, methodological commitments, and evidence limits. Across the literature, the evidence indicates that advances in hyperspectral image learning become credible when representation, objective, and evaluation protocol are evaluated together and when uncertainty about distribution shift is reported explicitly. The proposed reading joins method selection to implementation risk, making transfer failures visible, and proposes a research agenda centered on well-specified controls, robustness tests, and reproducible workflows.