Null Space Analysis for Detecting Unknown Objects During Automatic Target Recognition Tasks in Sonar Data

Abstract:

During automatic target recognition once a detector has found points of interest the classifier is then tasked with identifying target objects from non-target objects. However, occasionally the detector may find something that is neither known false alarm nor expected target. In these cases what is the classifier to do? In this paper we define a process using Null Space Analysis (NuSA) to detect the unexpected points of interest so that a classifier can provide notification that it has encountered data that requires additional review. NuSA provided a mechanism to manipulate the null space basis to minimize the amount of information present in the null space, this allows the projection to be measured so that anomoly detection can be performed. Analysis are results are shown that show NuSA outperforms similar methods in detecting unexpected data within sonar automatic target recognition tasks.

Links:

Citation:

M. Cook, A. Khoury, A. Zare and P. Gader, "Null Space Analysis for Detecting Unknown Objects During Automatic Target Recognition Tasks in Sonar Data," OCEANS 2023 - MTS/IEEE U.S. Gulf Coast, Biloxi, MS, USA, 2023, pp. 1-5, doi: 10.23919/OCEANS52994.2023.10337218.
@inproceedings{cook2023null,
title={Null Space Analysis for Detecting Unknown Objects During Automatic Target Recognition Tasks in Sonar Data},
author={Cook, Matthew and Khoury, Anthony and Zare, Alina and Gader, Paul},
booktitle={OCEANS 2023-MTS/IEEE US Gulf Coast},
pages={1--5},
year={2023},
organization={IEEE}
}