MMW-Det: A Context-Aware Detection Method With Hybrid Attention for Noncooperative mmWave Screening

In millimeter-wave-based noncooperative security screening scenarios, the high density and mobility of people lead to complex imaging backgrounds and diverse body postures, further posing a challenge to the detection of hazardous object carried by people. For the above problems, this article adopts a noncooperative millimeter-wave imaging system to acquire mmWave images in real time. We collect data including hazardous objects carried in various human postures in the simulated screening scenarios. To obtain a complete feature representation, we propose a detection framework for millimeter-wave noncooperative security screening (MMW-Det). The method is based on a real-time detection transformer (RT-DETR), and it significantly improves detection performance through the designed backbone, context module, and multiscale deformable cooperative attention. First, the backbone performs efficient feature extraction on millimeter-wave images. Then, the context module enhances the adaptability to different scale targets and different poses. And the attention module addresses the effects of different human postures and optimizes the feature representation of the target region. Theoretical analysis and experiments show that the proposed MMW-Det achieves the best performance on our noncooperative millimeter-wave dataset, with a mean average precision (mAP) of 79.2% and a single-frame image detection time of 9.3 ms. The method maintains real-time performance while significantly improving the detection accuracy and robustness, and achieves excellent performance in noncooperative millimeter-wave security screening scenarios.