Millimeter-Wave Imaging for Concealed Object Detection

Millimeter-wave imaging harnesses electromagnetic radiation in the 30–300 GHz band to form high-contrast representations of objects concealed beneath clothing or within baggage. Unlike X-rays, millimeter waves are non-ionising and pose no health risk, making them suitable for routine security screening at airports, public venues and border crossings. Systems may operate in active mode, transmitting a controlled signal and analysing its reflection, or in passive mode, detecting naturally emitted radiation from the body and concealed items. Advances in antenna arrays, synthetic-aperture radar and lens design have improved spatial resolution to the millimetre scale, while novel reconstruction algorithms mitigate speckle noise and diffraction artefacts. In parallel, machine-learning approaches—ranging from traditional feature extraction to deep convolutional networks—have accelerated detection and classification of metallic and non-metallic threats, achieving real-time performance. Integration with complementary sensors (visible, infrared or terahertz) and privacy-preserving rendering techniques further enhances detection accuracy without exposing anatomical detail. The global significance of millimeter-wave imaging lies in its ability to deliver rapid, non-invasive, stand-off threat detection, thereby strengthening public safety and counter-terrorism efforts.

Research from Nature Portfolio

Recent studies have demonstrated that single-shot detectors can be adapted to passive sub-THz imagery for faster and more accurate threat identification. One approach replaced the conventional backbone of the Single Shot MultiBox Detector with a residual network to facilitate deeper feature learning and reduce training complexity. A multi-scale feature-fusion module was introduced to enhance detection of small or low-contrast targets, while a hybrid attention mechanism sharpened spatial and channel responses to foreground objects. The application of a Focal Loss function further improved robustness to class imbalance. Experimental results on passive terahertz security datasets achieved a mean average precision exceeding 99.9% at 17 frames per second, outperforming established models such as Faster R-CNN, YOLO and RetinaNet.

Research from all publishers

Innovations in passive millimeter-wave screening have employed real-time object detectors originally developed for visible imagery. One study adapted the You Only Look Once (YOLOv3) framework to passive MMW images, achieving detection speeds above 30 FPS and mean average precision near 95% on small training sets. Another investigation combined human-pose segmentation with a convolutional neural network to reduce background clutter and transform object recognition into a binary classification task, improving detection reliability for low-reflectivity items. In active W-band systems, semantic-segmentation networks have been tailored for precise pixel-level localisation of concealed objects. By stacking dilated convolution blocks to preserve spatial resolution and applying connected-component analysis, researchers reported a 38% uplift in average precision and a 27% rise in intersection-over-union metrics for small targets against complex body backgrounds.

Millimeter-Wave Imaging for Concealed Object Detection publication trend

The graph below shows the total number of articles in millimeter-wave imaging for concealed object detection across all publications each year (not limited to Nature Index journals).

Technical terms

Millimeter-wave imaging: Non-ionising radar technique using 30–300 GHz waves to penetrate clothing and form images of concealed items.

Passive imaging: Mode that senses naturally emitted or ambient millimeter-wave radiation without transmitting a signal.

Active imaging: Mode that transmits millimeter-wave signals and analyses reflected energy to reconstruct object profiles.

You Only Look Once (YOLO): Real-time object detection framework that predicts bounding boxes and class probabilities in a single forward pass.

Single Shot MultiBox Detector (SSD): Single-stage detection network that discretises output space of bounding boxes into default boxes over different aspect ratios and scales.

Semantic segmentation: Pixel-wise classification method that assigns each image pixel to a predetermined category, enabling precise object boundaries.

Hybrid attention mechanism: Neural module that adaptively weights spatial and channel features to highlight relevant image regions and suppress background noise.

Focal Loss: Loss function designed to address class imbalance by down-weighting well-classified examples and focusing training on hard, misclassified samples.

References

1. Real-time Concealed Object Detection from Passive Millimeter Wave Images Based on the YOLOv3 Algorithm. Sensors (2020).
2. Improved SSD network for fast concealed object detection and recognition in passive terahertz security images. Scientific Reports (2022).
3. CNN with Pose Segmentation for Suspicious Object Detection in MMW Security Images. Sensors (2020).
4. Precise Localization of Concealed Objects in Millimeter-Wave Images via Semantic Segmentation. IEEE Access (2020).