Improved YOLOv8 Detection Algorithm of Contraband in X-Ray Security Inspection Image

Personnel safety is particularly important in public places where security checks are required, such as airports, train stations and subway stations. In view of the features of disorderly placement of items, blocking of contraband and intentional hiding in X-ray security inspection images, this research proposes an improved YOLOv8 network model combined with Mixup data augmentation, Biformer dual-layer routing attention modules, and C2f_f modules to assist security personnel in conducting efficient security inspections. At the input stage, the Mixup data augmentation method is employed to enhance the model's ability to learn to detect obstructed and concealed prohibited items. The Biformer dual-layer routing attention modules are introduced into the backbone network to enhance the model's focus on foreground information and improve its recognition ability. The C2f_f modules are incorporated into the backbone network and Neck section to improve the model's feature representation capability and enhance overall network performance. Testing on the Easy, Hard, and Hidden datasets of PIDray yields mAPs of 85.2%, 86.2%, and 68.3% respectively. Experimental results demonstrate an increased detection accuracy of 2.3%, 1.4%, and 2.2% respectively on the three prohibited item recognition test sets for the improved YOLOv8 model.