Novel Target Detection in Clutter using Joint Bayesian & Generalized Likelyhood Tests in Neyman-Pearson Detection

Hyperspectral sensors have proven capable of distinguishing targets that are otherwise very difficult to discern. However, such sensors also tend to produce false alarms at a significant rate. The Battle Management, Command and Control (BMC2) architecture will deal with most difficult detection scenario: targets may be sub-pixel in size, with a cluttered Earth background including clouds, land, water, and spectral noise, in a constantly shifting scene. The ultimate objective is to use the scene to extract background statistics and ultimately, to extract the target. The Generalized Likelihood Ratio Test (GLRT) and the Bayesian Likelihood Ratio Test (BLRT) are widely used as standard detection techniques for target signals that are corrupted by structured clutter and unstructured broadband noise. The choice between GLRT and BLRT depends on the availability of statistical or physical a-priori information about the unknown parameters in the data mixing model. We propose a new approach for combining the GLRT and the BLRT into a cascaded detector by deriving a general form of the BLRT. In this detector, the GLRT's role is to determine the background pixels' statistics by subtracting the target's pixels (or including sub-pixels) from the pixels of the entire image. Using the estimated statistical information about the background pixels, the BLRT is then applied to reduce the probability of false alarms. In the process, we also introduce a new way to discern between target and background structures in vector space, which is one of the most difficult problems in practice.