Hyperspectral image (HSI) classification is often challenged by cross-scene domain shifts and limited target annotations. Existing approaches relying on class-agnostic moment matching or confidence-based pseudo-labeling tend to blur decision boundaries, propagate noise, and struggle with spectral overlap and class imbalance. We propose Spectral Prototype Attention Domain Adaptation (SPADA), a framework that integrates an attention-guided spectral–spatial backbone with dual prototype banks and distance-based posterior modeling. SPADA performs global and class-conditional alignment through source supervision, kernel-based distribution matching, and prototype coupling, followed by diversity-aware active adaptation and confidence-calibrated refinement via prior-adjusted self-training. Across multiple cross-scene benchmarks in urban and inter-city scenarios, SPADA consistently outperforms strong baselines in overall accuracy, average accuracy, and Cohen’s $\kappa$, achieving clear gains on classes affected by spectral overlap or imbalance and maintaining low variance across runs, demonstrating robust and stable domain transfer.

Cross-well acoustic-impedance reconstruction combines seismic observations with retained auxiliary logs whose relationships change between wells. We introduce Multimodal Interaction and Residual Fusion (MIRF), which expresses multimodal fusion as an additive correction to a source-supervised seismic estimate. Three residual heads propose updates, a conditional router determines their composition, and an admission coefficient controls the applied magnitude. Separate anchor supervision retains a seismic reference for examining the effect of auxiliary-log correction. Using the original processed-log protocol on Penobscot L-30 to B-41 transfer, ten matched seeds reduce mean root mean squared error (RMSE) from 1905.5 to 1489.6 $\mathrm{kRayl}$ and mean absolute error from 1659.5 to 1178.4 $\mathrm{kRayl}$ relative to the standalone seismic CNN. MIRF also has the lowest mean RMSE among four matched neural fusion alternatives in this direction. Bidirectional comparisons, ablations, and depth-resolved profiles locate the benefits in reduced broad reconstruction bias and show how component effects change with transfer direction.

Fluid migration screening requires the integration of seismic structure and well log response. This study develops a multimodal framework for ranking an attribute-derived fluid migration favorability proxy from paired local seismic patches and well log windows in the F3 survey. The framework couples frozen seismic foundation model (SFM) features with a trainable well log token encoder and evaluates pooled concatenation, cross attention, and learned query fusion for multimodal integration. Under sample-stratified evaluation across three random seeds, frozen SFM features with learned query fusion achieved the highest observed normalized discounted cumulative gain at rank 20 (nDCG@20) of 85.15%, exceeding the best single-modality baseline by 3.95 percentage points. The results support combining pretrained seismic representations with well log evidence to prioritize intervals in the F3 benchmark.

Cross-material continuous tool wear prediction is difficult because a model must preserve the physical wear scale, not only align high-dimensional sensor features. This limitation is critical in milling, where the target variable is the continuous flank wear width ($\mathrm{VB}$) and material shift can distort the mapping from sensor response to wear magnitude. We address this problem by recasting cross-domain tool wear prediction as monotone wear-scale adaptation. We propose Multi-Physics Monotone Score Transport (MPMST), a monotone score transport framework that constructs a tool-wear-oriented score from sensor-derived candidate cues, transports the target-domain score onto the source-domain wear scale, and then predicts wear through isotonic regression. We also evaluate One-Physics Monotone Score Transport (OPMST), a force-only variant that uses the same score-transport pipeline with a restricted cue family. On Mondragon Unibertsitatea–Tool Condition Monitoring (MU-TCM) with two cross-material transfer tasks, the validation-driven MPMST configuration reduces mean absolute error by approximately 63% relative to Correlation Alignment (CORAL) and by approximately 31% relative to a physics-informed Gaussian process baseline. The results support monotone score construction and score transport as practical mechanisms for continuous tool wear prediction under domain shift, while also showing that MU-TCM is strongly force dominated.