Improved YOLOv5 Detection Algorithm of Contraband in X-ray Security Inspection Image

Personnel security is particularly important for public places where security screening is required, such as airports, railway stations and subway stations. To assist their security staff in fast security screening, a YOLOv5 network model incorporating Stem module, CGhost module and Mixup data enhancement method improvement is proposed for the features such as cluttered item placement, contraband obscuration and intentional concealment in X-ray security screening images. The Stem module is introduced in the backbone of the model to replace the first layer in the YOLOv5 network to enrich the features and improve the feature expression capability of the network. The CGhost module with stronger feature expression is designed to suppress the influence of redundant features on the network and enhance the network's ability to acquire multi-scale contraband features. At the input side, the Mixup data enhancement method is used to strengthen the network's ability to learn from obscured and hidden contraband learning capability, thus enhancing the recognition capability of the network. Tests were conducted in the Easy, Hard, and Hidden test sets in the PIDray dataset, and the mAP reached 88.2%, 89.2%, and 70.5%, respectively. The experimental results show that the improved YOLOv5 reduced the number of parameters of the model by about 20% and improved the detection accuracy by 2.5%, 1.2%, and 5.0% in the three test sets of contraband identification, respectively.