White-Box mHC: Electromagnetic Spectrum-Aware and Interpretable Stream Interactions for Hyperspectral Image Classification
This work addresses interpretability issues in hyperspectral image classification for researchers and practitioners, offering a partially white-box approach that is incremental in improving transparency.
The paper tackled the problem of opaque deep learning models in hyperspectral image classification by introducing ES-mHC, a framework that models electromagnetic spectrum interactions with interpretable matrices, resulting in coherent spatial patterns and asymmetric behaviors that provide mechanistic insights.
In hyperspectral image classification (HSIC), most deep learning models rely on opaque spectral-spatial feature mixing, limiting their interpretability and hindering understanding of internal decision mechanisms. We present physical spectrum-aware white-box mHC, named ES-mHC, a hyper-connection framework that explicitly models interactions among different electromagnetic spectrum groupings (residual stream in mHC) interactions using structured, directional matrices. By separating feature representation from interaction structure, ES-mHC promotes electromagnetic spectrum grouping specialization, reduces redundancy, and exposes internal information flow that can be directly visualized and spatially analyzed. Using hyperspectral image classification as a representative testbed, we demonstrate that the learned hyper-connection matrices exhibit coherent spatial patterns and asymmetric interaction behaviors, providing mechanistic insight into the model internal dynamics. Furthermore, we find that increasing the expansion rate accelerates the emergence of structured interaction patterns. These results suggest that ES-mHC transforms HSIC from a purely black-box prediction task into a structurally transparent, partially white-box learning process.