Multichannel Target Detection in Heterogeneous Sea Clutter using Online Dictionary Learning

Traditional coherent detectors are designed to work in complex Gaussian distributed clutter and rely on the availability of homogeneous training data. However, in the maritime domain, the dynamic nature of the sea clutter can lead to significant performance degradation. This is due to the clutter statistics being non-Gaussian and slowly varying over time and range. Dictionary learning is a data driven approach that can be used to estimate the background sea clutter and better distinguish targets. In previous work, this was applied to the range Doppler domain with improved detection performance when compared to traditional coherent processing techniques. We now extend the dictionary learning approach by exploiting the space-time spectrum achieved with a multichannel radar. The optimal power spectrum relies on a covariance estimate which is typically averaged over a number of range bins. However, by training the dictionary to learn what the power spectrum looks like at a single range bin, the approach becomes robust to variations over range, and ideally suited for detection in heterogeneous clutter.