Research And System Development Of Intelligent Detection Technology For Safety Of Parcels

With the vigorous rise of mobile Internet and e-commerce industry since 2010,the express logistics industry has ushered in rapid development.And the industry has gradually become one of the leading industries to promote consumption upgrading,promoting high-quality economic development.However,the significant growth of the express delivery business has put a lot of pressure on the security screening process.At this stage,the security inspection process generally uses X-ray image security inspection machine for information collection,and professional security personnel carry out manual identification.Therefore,under large detection pressure,manual identification is prone to the risk of low detection accuracy and greater influence by environmental and subjective factors.In view of the above problems,this paper proposes a detection algorithm for dangerous goods in X-ray security inspection images based on deep learning,and builds a set of intelligent detection system for the safety of sent goods based on the algorithm.The specific research content is as follows:(1)An improved YOLOv3 algorithm is proposed for dangerous goods detection in X-ray security inspection images,so as to improve the detection rate of the original YOLOv3 algorithm.First,the SIXray dataset is selected as the basic training dataset.Aiming at the problem of unbalanced distribution of various samples in the dataset,data enhancement and data self-collection supplement were carried out,so as to form a set of sufficient number and complete X-ray image dangerous goods detection data set.Secondly,based on the data set,YOLOv3 target recognition algorithm was improved adaptively,and the idea of transfer learning was introduced to reduce the difficulty of model training.The initial anchor frame was optimized through K-means clustering,and DIoU was used to replace IoU to optimize the regression function of the frame.The residual block settings of different network depths of the backbone network Darknet-53 were adjusted to improve the network’s recognition accuracy for small and medium-sized targets.Experimental results show that the mAP of the improved object recognition algorithm is increased to 85.47%,which is 2.29%higher than that of the original YOLOv3 model.(2)Aiming at the problems of front and rear occlusion of target objects and complex background in complex detection scenarios,the atrous spatial pyramid pooling and attention mechanism are introduced.Firstly,atrous convolution is used to expand the receptive field and obtain contextual information to improve the recognition of the phenomenon of objects occluding each other in X-ray images.Secondly,the CBAM attention mechanism is introduced to put more recognition weight into the local information containing the target object,which reduces the influence of complex background and noise on the detection accuracy.Experimental results show that the mAP of the algorithm is 88.77%,which is 5.59%higher than before the improvement.(3)Based on the above optimization algorithm,an intelligent detection system for the safety of sent items is built.On the basis of realizing the X-ray image dangerous goods detection function,data management and user management modules are added to form a set of operating systems that can be connected to the real-time detection of security inspection machines.Finally,experimental verification was conducted based on X-ray security inspection equipment.The experimental results indicate that the system performs well in the experimental environment and can effectively detect dangerous goods in real-time.