Leveraging Complex-Valued Federated Learning for Accurate and Privacy-Respectful Threat Detection Based on Millimeter-Wave Imaging

Millimeter-wave (MMW) radar systems are widely recognized for their affordability, safety, and effectiveness in concealed object (CO) detection. This paper introduces a novel high-resolution pseudo-image-based CO detection framework that leverages complex-valued convolutional neural networks (CV-CNNs) and their federated learning extension (CV-FL) to enhance detection accuracy while preserving user privacy. Additionally, a decision-level fusion (DLF) framework is proposed to further improve CO detection performance. This approach provides a highly accurate and privacy-preserving solution for CO detection using high-resolution MMW radar. The proposed framework preserves privacy by processing complex-valued (CV) data locally to generate CV pseudo-images, which are not transmitted out by leveraging CV-FL, maintaining high detection accuracy and convergence efficiency. In this study, MMW data are collected using a commercial handheld scanner (Vayyar’s ECS2000), focusing on localized scans of the human body. The collected data are complex-valued (CV) high-resolution local 3D pseudo-images over a volume of 13×13×10 cm with spatial resolutions of 1.6 mm, 1.6 mm, and 4.3 mm in the x, y, and z directions, respectively. This paper extends a cutting-edge real-valued (RV) concealed object (CO) detection framework by incorporating its complex-valued (CV) and federated learning (CV-FL) versions, further enhancing its accuracy with a decision-level fusion (DLF) module. The core innovation of this work is the application of CV-FL to CO detection for the first time, enabling collaborative AI training while preserving data privacy. Simulation results reveal a significant performance improvement of up to 17.9% in detecting both metallic and non-metallic objects, highlighting the effectiveness of the proposed approach. In addition, the dataset of the acquired data has been made publicly available.