Feature extraction algorithms for vehicle classification techniques represent a large branch of Automatic Target Recognition (ATR) efforts. Traditionally, vehicle ATR techniques have assumed time series vibration data collected from multiple accelerometers are a function of direct path, engine driven signal energy. If data, however, is highly dependent on measurement location these pre-established feature extraction algorithms are ineffective. In this paper, we examine the consequences of analyzing vibration data potentially contingent upon transfer path effects by exploring the sensitivity of sensor location. We summarize our analysis of spectral signatures from each accelerometer and investigate similarities within the data.

The potential applicability of multiple-channel coherence estimation in situations where one channel contains a noise-free signal replica (as in active radar) or a high-SNR reference signal (as in passive coherent radar) has been proposed in recent work. Invariance of the distribution of M -channel coherence estimate statistics, including recently derived variants optimized for detection of signals having known rank, to the presence of a strong signal on one channel provided all channels are independent and the other M − 1 channels contain only noise enables the desired use of these statistics without altering detection thresholds designed to provide desired false-alarm probabilities. Traditionally, multiple-channel detection using coherence estimates has assumed that time series data from all channels are aggregated at a fusion center. Mitigation of this requirement to demand global aggregation of only scalar statistics that can be computed locally by sharing of data between pairs of nodes has been explored, and the use of maximum-entropy methods to provide surrogate statistics for pairs of nodes that are not in direct communication within a network has been proposed for traditional passive detection problems. This paper examines the applicability of this idea in the presence of a strong reference channel with particular attention to ascertaining relationships between network topology and detection performance.

Coherence estimation is an established approach in multiplechannel detection and estimation, providing optimal solutions in many cases. Recent work has considered the use of maximum-entropy matrix completion when elements are missing from the gram matrix from which the coherence statistics are formed. This is desirable in sensor network settings, for example, where direct communication is not available between every pair of nodes in the network. This paper examines the role of network topology in determining the conditional distributions of the statistic obtained by the matrix completion process under both signal-present and signal-absent hypotheses.

No abstract