Target maneuver onset time estimation via significance test of acoustic feature difference

Abstract

An acoustic feature change detecting based method is proposed in the paper to estimate the target maneuver onset time. Form the statistical analysis of many acoustic features of different targets, the distribution of a certain estimated feature value in a single motion state is approximately normal. Base on the normal distribution, we defined the acoustic feature difference index to select the most sensitive feature to the change of target motion state. Using the feature, we proposed an online method to estimate the target maneuver onset time by testing the significance of acoustic feature difference. Experiment results show that the proposed method is more rapid to estimate the maneuver onset time after the target maneuver occurs. Introduction Rapid and robust target maneuver onset time detection is crucial to obtain comparable target tracking accuracy [1-6]. Many algorithms and techniques have been developed to detect maneuvers. Most of the detectors attempt to identify a maneuver by predicting the expected target position and comparing it to the measured track. The difference between the predicted position and the measured named innovation. Ref. [6] presents six traditional and two novel maneuver detection algorithms. They are all innovation-based. These methods are quite slow. In the radar area, it typically requires a few seconds to detect a maneuver. But in the sonar area the observer is normally far from the target and the bearing of target is changed very slowly, so it typically requires several minutes to detect a maneuver. It is not acceptable for many practical situations in sonar system. Feature-based maneuver detection techniques have been proposed in radar area using radar glint noise [7] and range-Doppler images [8]. In fact, radar target feature retains strong link with target motion mode. Compared with innovation-based techniques, the feature-based technique is fast response and more reliable. In this paper, we exploited the problem using the acoustic features of ship’s radiated noise. Since the features of ship’s radiated noise have direct link with target motion state, the detector is devised by detecting the acoustic feature differences. Form the ship’s radiated noise, we can extract many features. Most of the features are changed during maneuvers, but some of the features are more sensitive to target motion state. We defined the acoustic feature difference index (AFDI) to select the most sensitive feature, and use it to develop the detector by testing the significance of acoustic feature difference. Some experiments were carried out to illustrate the capability of the proposed detector. Experiment results show that the proposed method is more rapid to estimate the maneuver onset time after the target maneuver occurs. Distribution of the estimated feature value In the practical situation, we get the target’s radiated noise ) (t s in real time. The fix length window can be used to get signal segments with certain duration T, such as 0.5s. Using each signal segment, many features can be extracted by different methods. We use l k f representing the lth feature value of these extracted features estimated by the kth signal segment. Assuming the target’s motion state is not change in the K k , , 1 = , the lth feature value set can be given by