A Novel Detection Method for Low-Frequency Active Sonar based on Joint Distribution Background Fitting and Multiple Feature Fusion

The background of the active sonar deviates far from the Rayleigh model when encountering reverberation or clutter, which causes unexpected false alarm rates in conventional detectors. Especially for the low-frequency active sonar, due to its relatively low measurement resolution, there are relatively few features that can be utilized for target recognition. In this article, a Joint-Distribution (Joint-D) based fitting method is proposed for non-Rayleigh sonar data, to enhance the performance of conventional constant false alarm rate (CFAR) detection. Subsequently, other than single energy criterion, multi feature fusion based on Joint-D is implemented for both the local and spatial subregions of the detection results. Finally, target recognition is performed using a classifier trained by dual-stage support vector machine (SVM), where the first stage utilizes local features and the second stage utilizes spatial features. Receiver operating characteristic (ROC) curves show that the proposed method outperforms conventional counterparts across SRR from 0-25 dB. When achieving the same detection rate, particularly the 90% threshold, it requires an average of 5 dB less SRR than conventional counterparts. In the case study using sea trial data, the detection rate reaches 100%, and the false alarm rate is consistent with the theoretical value.