Reducing Computational Complexity in Hyperspectral Anomaly Detection: a Feature Level Fusion Approach