2D-ISAR produces images that strictly depend on the geometry of the whole radar-target system and on the relative motion between radar and target. This poses some limits on the use of Automatic Target Recognition (ATR) systems. To overcome this issue, 3D point clouds as a result of 3D-ISAR imaging were proposed as a more complete and reliable representation of the target. Since the acquisition system will output an unknown number of points in a random order, the chosen classifier must be able to process a variable number of input elements to correctly classify the target. After a brief presentation of the state of the art about the 3D classification problem, the architecture of Point Cloud Transformer (PCT) is introduced. PCT is trained and tested on an ad-hoc generated 3D dataset, which in this preliminary experiment contains three different target types: cars, tanks and military trucks. The goal of this work is to show how the transformer is able to correctly manage the recognition of targets, even if the point clouds are made by few points. Lastly, the trained network is tested on some real data.

Keyword: Transformer, Deep learning, Classification, ATR, 3D, ISAR, 3D-InISAR

File: link al file della pubblicazione