Research Of Intelligent Identification Algorithm For Express Contraband And System Development

With the rapid development of the express logistics industry,the circulation of express parcels is growing at a blowout rate,which has brought huge challenges to the express security inspection process.However,at this stage,manual express delivery security checks have problems such as low detection accuracy,slow speed,and high environmental factors,which can easily cause missed or false detections of contraband,and cause safety accidents in the express logistics and transportation link.In response to this situation,this paper conducts research on the detection algorithm of contraband in lightweight X-ray images based on deep learning,and develops an intelligent identification system for express contraband to realize automatic identification and positioning of contraband in express X-ray images.The main research contents are as follows:(1)Aiming at the problem of low accuracy of manual identification of contraband in X-ray images,a detection algorithm for contraband in Xray images based on YOLOv5 is proposed.First of all,for the detection of small target contraband,design a multi-scale output detection layer,optimize the initial anchor frame based on K-means++,add a channel and spatial attention mechanism in the backbone network to reduce the interference of background information,and improve the model’s ability to detect small targets.Secondly,for the detection of stacked contraband,a DS-NMS algorithm is proposed to reduce the mistaken deletion of the prediction frame of stacked contraband at the output end,and improve the detection accuracy of stacked contraband;finally,by designing a multiscale feature fusion network and improving The bounding box loss function further improves the positioning accuracy of the model for contraband.The experimental results show that the average accuracy of the algorithm in the SIXray_DA data set after data enhancement reaches 92.24%,which is 6.7%higher than that before the improvement.(2)Considering that express security inspection has high requirements for real-time,the proposed X-ray image contraband detection algorithm is designed as a lightweight model.Firstly,the lightweight structure design is carried out for the network model before training,and the lightweight backbone network and neck network modules GHC,C3 Ghost and C3GS are constructed,and the lightweight space pyramid pooling Light-SPPF is proposed,so that the model can maximize the accuracy while reducing the amount of parameters and calculations.Secondly,pruning is performed for the trained network model to remove the necessary convolution channels,which further improves the detection speed of the model to meet the requirements of real-time.The experimental results show that under the premise of maintaining the average accuracy of 90.22%,the number of parameters and the model volume are reduced by 69%and 68%,respectively,and the detection speed is increased by 18%,which achieves the effect of lightweight model design.(3)Building an experimental platform for intelligent identification of express contraband based on X-ray security inspection equipment,design and implement an intelligent identification system for express contraband,and deploy and perform performance tests on the algorithm model proposed in this paper.The experimental results show that the average detection rate of contraband in various scenarios has reached 91.9%,and the average detection time is 0.4s.The system can detect contraband in express X-ray images in real time and accurately,and is feasible in practical application scenarios and effectiveness.