Abstract: Millimeter wave imaging technology is an emerging direction in the field of security inspection.Researching object detection algorithms that meet this application scenario and improving the corresponding detection speed and detection accuracy have high application significance and value.This paper proposes a real-time object detection algorithm based on the YOLOv5 deep learning model to detect contraband hidden by security personnel.The method uses the GIOU_Loss loss function to improve the ability to measure the intersection of the detection frames.In addition,the network structure is optimized and the improvement and the addition of the millimeter wave data enhancement preprocessing function to accelerate the convergence,thus forming a deep neural network for millimeter wave images for item detection to improve the object detection effect.Experimental results show that this method can effectively detect dangerous goods in millimeter-wave images,and has the advantages of automatic identification and real-time detection.

Key words: Deep learning, Millimeter wave image, Object detection