Research on Concealed Dangerous Goods Detection Based on Active Terahertz Active Imaging

Active Terahertz imaging technology exhibits significant potential in security inspection due to its advantages of rapid imaging, strong penetration capabilities, and harmlessness to humans. It is poised to become a mainstream technology in this field. However, the study of detecting concealed dangerous goods in terahertz human imaging using deep learning remains in its infancy, facing challenges such as limited databases, low image resolution and contrast, and inadequate small-object detection capabilities. This paper addresses these challenges by introducing a large-scale dataset with over $\mathbf{1 2, 0 0 0}$ terahertz images from the TGR-23 system, covering 10 categories of hazardous items, and enhancing small-object detection through the improvement of the YOLOv8 model by incorporating a Context Feature Extension Module (CAM) and a Residual Improved CAM (RCAM), which resulted in a 2% increase in detection accuracy.