Target Classification for 3D-ISAR Using CNNs
Chow Yii Pui, Brian W.‐H. Ng, Luke Rosenberg, Tri–Tan Cao
Abstract
In maritime surveillance, inverse synthetic aperture radar (ISAR) is a technique for imaging non-cooperative targets, with classification typically performed by the radar operator. By automating the target classification process, the operator workload will be reduced significantly and the classification accuracy can be improved. Traditional classification approaches use geometric features extracted from images of known targets to form a training dataset that is later used to classify an unknown target. While these approaches work reasonably well, deep learning-based techniques have recently demonstrated significant improvements over conventional processing schemes in many areas of radar. Classification of traditional 2D-ISAR imagery is difficult due to the motion of the sea causing a wide range of imagery. The three dimensional (3D)-ISAR technique was developed as an alternative representation with the target represented by a 3D point cloud. In this paper, we investigate how 3D-ISAR can be used for classification of maritime targets. The proposed scheme makes use of features extracted from the 3D-ISAR generated point cloud of the target from different perspectives (i.e. side, top and front views) to form three point density images (PDI). These are then fed into a convolutional neural network (CNN) to classify the targets.