Signal subspace processing of 4D remote sensing data

The thrust toward high resolution multispectral staring array sensors for a variety of remote sensing applications has produced a need for efficient processing techniques that fully exploit the time, space and spectral correlations of the data. Over the past two decades, the techniques of signal subspace processing have been especially effective in processing 2 and 3 dimensional data. These include approaches such as TMF (Temporal Matched Filter), MUSIC (MUltiple SIgnal Classification), Principal Components and STAP (Space Time Adaptive Processing). This paper extends the basic theory and algorithms to handle 4D (spectral, temporal and two spatial dimensions). Suboptimal formulations are presented that reduce the computational requirements while still taking advantage of the critical correlations. The background estimation and target detection performance of these techniques are demonstrated using (a) high fidelity simulated scene sets corresponding to hyperspectral and multiband imaging sensors (b) measured data sets from a current multiband space sensor The paper concludes with recommended algorithm investigations and measured data set evaluations