Research On Automatic Detection Of Contraband Based On X-ray Images

Based on passenger luggage X-ray images,automatic detection of guns,equipment and other contraband in passenger luggage can not only assist security inspectors in maintaining aviation and transportation safety,but also have important significance for the sustainable and rapid development of civil aviation smart airports in the future.At present,the task of passenger luggage security inspection mainly relies on the manual identification of X-ray images of passenger luggage by security personnel,which is prone to misdetection and missed inspection due to problems such as visual fatigue,which is extremely unstable.In order to solve this problem,based on the characteristics of passenger luggage X-ray images,the thesis combines the deep learning method with the contraband detection task.When faced with the serious imbalance of the number of contraband in the passenger luggage in the real scene,the image classification method is used to identify the problem.Various types of contraband that may be contained in passenger luggage X-ray images.At the same time,the thesis uses SwinTransformer to extract the characteristics of contraband in the X-ray image containing contraband,and completes the task of contraband detection and positioning of X-ray image by the method of target detection.The specific research contents are as follows:Aiming at the problems of unbalanced samples of passengers’ baggage contraband and missed detection and false detection caused by different scales of contraband in real scenarios,the convolutional neural network Res Net101 is used,and the multi-scale feature extraction capability of the model is improved based on the Res2 Net network,and a feature fusion module is designed.The shallow network with more edge and texture features in the model is spliced with the deep network to improve the expression information of contraband in the feature map.Based on the cost-sensitive idea,the thesis uses the class balance loss function to reduce the interference caused by a large number of negative samples to the generative model,and uses the Focal Loss loss function to improve the model’s classification accuracy of contraband in the sample imbalance and difficult sample classification problems.The experimental results on the passenger baggage public data set SIXray show that the proposed method improves the m AP index by 4.5% compared with the current optimal end-to-end classification model,and the AP index of difficult-to-classify samples such as scissors and pliers is significantly improved Effect.Aiming at the problems of poor effect and low robustness of the target detection model of passenger baggage X-ray image,this thesis proposes a passenger baggage X-ray image enhancement strategy,which uses median blur,random blur and self-defined blur to remove part of the noise to improve the X-ray image.Considering the influence of the robustness of the target detection task model,considering the interference caused by the X-ray background to the feature extraction of contraband,the original image and the background-removed image pixels are superimposed.Through the use of the data augmentation module,the accuracy of the model in the contraband detection task of the SIXray dataset is improved by 0.8%.The thesis cites Swin-Transformer to extract the features of contraband in a targeted manner,and uses the FPN network to combine Swin-Transformer with the target detection model Cascade RCNN.The detection result of the model in the SIXRay dataset reaches 91.2%,and the detection result in the OPIXray dataset reaches 89.4%,and the detection accuracy has been greatly improved.