Adaptive underwater target classification with multi-aspect decision feedback

This paper presents a new scheme for underwater target classification in a changing environment. An adaptive target classification system is developed that uses the decisions of multiple aspects of the objects. The system employs a decision feedback mechanism to map the changed feature vector to a new feature space familiar to the classifier. Results on an acoustic backscattered data set, namely the 40 kHz data collected at Coastal Systems Station (CSS), are presented. This data set contains returns from six different objects at 72 aspect angles with 5 degrees separation and with varying signal-to-reverberation ratio (SRR). The results are then benchmarked with those of a neural network-based multi-aspect fusion system.