Examination of active target detector using in situ data

Abstract

When operating active sonar systems, target-like signals including clutter are detrimental to detecting target signals and induce an increment of false alarms. In this work, various approaches including machine learning and deep learning techniques are applied to in-situ data to investigate their performance. First, one of conventional schemes, constant false alarm rate (CFAR) detector is used to find target signals in acoustic data. It detects target signals with high SNR, but it also captures high-intensity clutter signals, which leads to many false alarms. Subsequently, data-driven decision rules from machine learning of support vector machine (SVM) and deep learning of convolutional neural network (CNN) are applied. Both of SVM using aural features and CNN with spectrogram show better accuracy. In particular, false alarms from the CFAR detector are extremely decreased owing to the data-driven decision rules not exploiting the signal intensity for target detection.