Concurrent detection and classification of targets with multistage signal-processing algorithms

Concurrent detection and classification (CDAC) of targets stands as the goal in littoral mine-hunting missions. CDAC systems commonly apply model-based algorithms that include a priori known features of the target inside the detection algorithm. If the models are accurate, then this approach significantly reduces the false-alarm rate inherent in detection-only methods. When the possible targets are unknown, as may be the case in tactical situations, then these model-based methods not only fail to reduce the false-alarm rate, but may also reduce the probability of detection. Simultaneous optimization of detection and classification presents a challenge due to competing criteria; detection seeks to integrate signals to improve signal-to-noise ratio, while classification seeks to preserve small features of distinction within the signals. In this work, a method for robust CDAC is demonstrated that exploits the capabilities of autonomous underwater vehicles (AUVs) and multistage signal-processing algorithms to systematically investigate targets of interest in a single mission. The deformable geometry of the AUV-borne sonar network is exploited to provide favorable views of targets to achieve multiple objectives in series, and on-board computational facilities allow the implementation of multiple signal-processing regimes.