Spectral-Spatial Interpretation

The paper argues that post-hoc image explanations typically show either spatial heatmaps or global channel scores, which is limiting for multispectral data because channel relevance is often spatially localized and physically meaningful. It introduces Spectral-Spatial Interpretation (SSI), described as a full-grid attribution method. The supplied text is truncated before the method's details and any experimental results are described.

Why it matters

Explaining models on multispectral data needs to show which spectral bands matter where, which could make such models easier to inspect in domains that rely on physical measurements.

Read with care

The source text is truncated mid-sentence, so the method's mechanics and performance claims are unavailable. Source claims not independently checked.