| Stations,airports,subways,stadiums,as an important part of public places,use Xray security inspection machines for contraband inspections,which can effectively avoid the threat or interference of prohibited items such as controlled knives,explosives or chemicals to passengers or means of transportation.The necessary way to ensure the safety of personnel and property.At present,the detection results of contraband security inspections are still subject to the human judgment of the security inspectors,which affects the detection accuracy of contrabands,which will lead to the occurrence of some unstable factors,and deep learning technology has made extraordinary achievements in various fields.Therefore,this paper uses a deep learning target detection algorithm to automatically detect and identify contraband during the security inspection process,so as to improve the detection efficiency during the security inspection process and prevent safety accidents in public places with a large flow of people,which has important research significance.The main research contents of the thesis are:First,analyze the SIXray security inspection image data set,then select a variety of image processing methods,and compare the extraction capabilities of several segmentation methods for the main body of various contraband images based on traditional edge detection algorithms,combined with X-ray The characteristics of the security inspection image,an improved Canny edge detection method is proposed to remove the background noise interference to obtain the main image of the contraband.After determining the most suitable image segmentation method,the data set was effectively enhanced.Secondly,on the basis of YOLO-v3,the SPP module and a new 104×104 scale are added,and the network is improved to 4-scale detection to improve the identification ability of small target contrabands.The K-means clustering algorithm is used to multiscale prior frame.Adapt to different sizes of all kinds of contraband.An improved target detection algorithm is used to detect and identify the processed X-ray contraband images.Finally,using Recall,Precision,m AP,etc.as evaluation indicators,the experimental results of the data set before and after image processing and before and after the improved network are compared.The experimental results of the image processed security image data set in the improved YOLO-v3 network are 84.94%.,Compared with the original paper,m AP increased by 5.38%.In addition,compared with the current research,this article is better than other researches under the same data set. |
