Research On X-ray Image Detection Method Of Prohibited Item Based On Deep Learning

In recent years,terrorist attacks and violent incidents occur frequently,which have a tremendous impact on personal safety.Security issues have become a concern of leaders of various countries.Safety inspection is an important measure to reduce terrorist incidents.X-ray security inspection equipment is widely used in various public places.However,the current security inspection work is highly dependent on labor.In the case of large passenger flow and long working hours of security personnel,it is prone to missed inspection and misdetection.In this context,X-ray prohibited item detection has become a research hotspot,using artificial intelligence technology to prohibited item in real time and assist security personnel to check.Based on the characteristics of X-ray images of prohibited item,this thesis proposes a Xray image detection network based on deep learning.The main research contents are as follows:First,we establish a X-ray dataset and process sample.Firstly,in the view of the problem that the public X-ray prohibited item dataset is less and the image quality is low,this thesis establishes the common contraband dataset PIXray,which consists of 5148 X-ray images containing six classes of prohibited items.Secondly,image denoising and sharpening are performed.Finally,the true value information is manually annotated to obtain a high standard dataset.Second,we design a multi-scale feature extraction network based on attention mechanism.In this thesis,a multi-scale feature extraction network based on attention mechanism is proposed for the different sizes of prohibited item and the chaotic background of security images.Introduce the channel attention module,assign different weights to each feature channel,highlight the useful information channel;fuse the feature maps of each scale extracted by the convolutional neural network to combine the underlying visual information and high-level semantic information,and the experiment shows that the design of feature extraction network effectively improves the detection accuracy of prohibited item.Third,we design a detection network for dense occlusion problems.Aiming at the phenomenon of stacking and occluding objects in security check,we improve the nonmaximum suppression algorithm,the score of the frame suppressed by non-maximum value is reduced,and the missed detection phenomenon is reduced to some extent;the repulsion loss is added in the loss function,so that the prediction box is far from the ground-truth objects of the non-target,and reduces the interaction between the two targets that are closer.Experiments show that the detection network can detect prohibited item more accurately under the phenomenon of dense occlusion.Research on X-ray image detection method of prohibited item based on deep learning proposed in this thesis has theoretical and practical value,and can assist security person to complete common contraband detection tasks and improve security inspection efficiency.