Beyond Reconstruction: Reconstruction-to-Vector Diffusion for Hyperspectral Anomaly Detection
Reconstruction-to-Vector Diffusion (R2VD) moves beyond simple image reconstruction for anomaly detection by treating it as a manifold purification process. It replaces ambiguous scalar errors with high-dimensional vector dynamics, using four stages—PPE, GMP, RSM, and VDI—to robustly isolate targets from complex backgrounds. This new paradigm achieves state-of-the-art performance by evaluating vector consistency instead of relying on flawed scalar residuals.