تشخيص هدف مجموعه باز در تصاوير رادار روزنه مصنوعي با استفاده از يادگيري عميق

This research addresses the problem of automatic target recognition in synthetic aperture radar (SAR) imagery under an open-set scenario, in which, in addition to correctly classifying known targets, the ability to detect an‎d reject unknown samples is also required. Challenges such as speckle noise, severe variability of radar backscattering signatures with respect to viewing angle, an‎d the scarcity of labeled training data have rendered traditional methods an‎d even convolutional neural networks unreliable in real-world operating conditions. To overcome these limitations, a novel unified architecture is proposed in this study, in which a Vision Transformer (ViT) is employed to extract global image dependencies, capsule networks are utilized to model the spatial structure of targets an‎d preserve part–whole relationships, an‎d the PS-OpenMax classifier is incorporated to identify unknown samples. In addition, the particle swarm optimization (PSO) algorithm is adopted to automatically an‎d effectively tune the hyperparameters of the PS-OpenMax classifier. The performance of the proposed model is eva‎luated on the widely used MSTAR, SAMPLE, an‎d SAR-ACD datasets. Experimental results demonstrate that the proposed approach achieves higher accuracy, greater robustness against increasing openness, an‎d superior generalization capability on unseen data compared to existing methods. Specifically, the proposed method attains an accuracy of 95.59% an‎d an F1-score of 96.45% on the SAMPLE dataset, outperforming previous studies. Furthermore, it achieves an accuracy of 96.13% an‎d an F1-score of 96.49% on the SAR-ACD dataset. These quantitative results indicate that integrating transformer-based architectures with capsule networks, together with advanced open-set recognition techniques, provides an effective an‎d reliable framework for the development of automatic target recognition (ATR) systems in SAR imagery.

محمدرضا احمدزاده

مجتبي بهشتي

محمدرضا تابان , حامد نريماني