Millimeter-wave concealed foreign object detection based on reconstructed network

Millimeter-wave (MMW) technology is widely used in security screening for its safety and privacy protection. However, concealed object detection based on MMW images still faces challenges: low resolution limits feature discriminability, sample imbalance reduces recognition of rare classes, and high model complexity hinders real-time performance.we propose a reconstructed network for concealed foreign object detection (RCFD-Net), designed for single-class detection of all suspicious concealed objects. RCFD-Net uses a lightweight image reconstruction module to remove foreign objects and locates them by analyzing differences between original and reconstructed images. A feature-reuse fusion classification head further filters detected regions to reduce false positives. Experiments on an MMW security dataset show that RCFD-Net outperforms state-of-the-art methods, achieving an Average Precision (AP) of 86.78.