Theory And Implementation Of Security Inspection Image Detection And Segmentation System Based On Deep Learning

In recent years,there have been many terrorist attacks at home and abroad.Each search resulted in a large number of casualties.The place where the incident occurred was almost crowded,such as train station squares,park entrance areas,airport entrance areas,shop entrance areas,and school entrances.Area,etc.The development of X-ray fluoroscopy and digital image processing technology is widely used in the field of security inspection.The X-ray security inspection image provides security inspectors with various forms of object perspective images,and provides a reliable basis for security personnel to carry out detection of dangerous goods,thereby making the security inspection work more convenient and faster.However,the X-ray security inspection equipment used in some important places such as modern subway entrances,airports,ports,etc.,can only perform simple classification on the packaged items,such as organic,inorganic,metal,and liquid,while the functional classification of the detected objects is still Visual inspection by professional security inspectors is required.Manually observing security inspection images relies heavily on the experience and technology of security inspectors.Due to the subjective differences,misdetection and omission of some dangerous objects often occur.Therefore,this thesis aims to study the functional difficulties of security inspection scenes and the limitations of current detection methods,using digital image processing and deep learning theories and methods.This article first systematically elaborates the professional knowledge theories such as digital image processing,convolutional neural networks,and case segmentation algorithms involved in the research,and lays a theoretical foundation for subsequent experimental research.Secondly,through the use of image preprocessing technology,the data is marked and specially processed to provide a large-scale data resource center for the security inspection system in this paper.Then,based on the analysis and research of target detection and semantic segmentation in the field of deep learning,based on the Mask RCNN instance segmentation algorithm,an instance segmentation algorithm Faster Mask RCNN(FMRCNN)using semantic contour information is proposed.The structure of the model is adjusted for the RPN network,and the Softer NMS algorithm is used instead of the NMS algorithm in terms of attention mechanism.The comparison results of several experiments show that the optimized FMRCNN model significantly improves the detection accuracy,and is higher than the accuracy of the intelligent security inspection dangerous goodsidentification algorithm based on Intel Open VINO technology applied in the second China International Import Expo.Finally,a mature FMRCNN model is used to analyze and design a system based on deep learning for security inspection image detection and segmentation to interface with security inspection machines.By combining artificial intelligence,big data,and security,on the one hand,people’s traffic safety coefficient is greatly improved,providing more effective guarantees for the public’s personal safety,and automated detection greatly shortens the time to enter and exit public places.It is more convenient for citizens to travel;on the other hand,automatic detection of dangerous goods assists security personnel to complete security tasks more efficiently,reduces the pressure on security personnel,and has a positive impact on saving labor costs and work efficiency.In the long run,this gives The security field brings new thinking and new methods,and the application of artificial intelligence is more extensive,bringing a new round of revolution to traditional industries.