Security Detection And Segmentation Of Baggage X-ray Image Based On Deep Learning

In recent years,with the improvement of Chinese transportation network,it is more and more convenient for residents to travel by rail,plane,subway and other public transport modes.X-ray security detection is very important as a common means of security detection in public transportation and public places.At present,the X-ray security inspection adopts the method of machine imaging and manual identification.Due to the low efficiency of manual identification,it can’t meet the requirements of high-precision security inspection.Therefore,this paper proposes a security detection and segmentation algorithm based on deep learning.The main contents of this paper are as follows:According to the current situation of baggage X-ray image security detection and segmentation,this paper analyzes the purpose and significance of the research,and determines the use of deep learning method as the image detection method in this paper.The origin of deep learning and the structure and development of classical algorithms are discussed in detail.This paper introduces the principle of threshold segmentation,region segmentation and edge operator detection segmentation of the current classical image segmentation algorithm in detail,and designs the contrast experiment on the data set GDX-Ray of luggage X-ray image.Experimental results show that the traditional image segmentation algorithms have their own advantages and disadvantages,which can not fully meet the needs of baggage Xray image segmentation,and the segmentation time is slightly longer,which is not conducive to practical application.The development of deep learning in the field of image segmentation is briefly introduced,and the experimental results show that the full convolution network based on deep learning has a good application in the research field of this paper.In order to integrate real-time image detection and segmentation,Mask R-CNN is selected as the main task of image detection and segmentation in this paper.The structure of Mask R-CNN network is analyzed,and we give a detailed introduction to its structure innovation points,ROI Align layer and mask division branch.According to the common problems of multi-scale target detection and overlapping target detection in baggage X-ray image detection and segmentation,the feature extraction network and calibration frame screening network are optimized respectively.Firstly,the FPN network is combined with Res Net,which is the backbone network of Mask R-CNN.The image position information and semantic information are fully utilized for the target to be tested in different scales by layer sampling and horizontal connection.Then,the NMS network which generates and filters the calibration frame is improved to soft NMS,so that adjacent targets or overlapped targets can be detected without being missed.According to the relevant regulations of the traffic control department,guns,knives,blades and small target nails are selected as the objects of experimental detection and segmentation.The baggage X-ray image which contains these four kinds of dangerous goods in GDX-Ray data set is selected to test the improved algorithm.Due to the small amount of data,the pre trained model in the large-scale image detection data set MS COCO data set is selected as the initial model,and the data set in this paper is trained by the fine-tuning method in migration learning.The experimental results show that the MAP value of the improved Mask R-CNN model is 0.973,and the detection speed is 4.55 fps.Compared with the improved Mask R-CNN,the MAP value is increased by 0.026,and a high-quality segmentation mask is generated,which can basically meet the accuracy and speed requirements of baggage X-ray security detection and segmentation.Considering the practical application situation,in order to facilitate the operation of security inspectors,a system of baggage X-ray image security detection and segmentation is designed.Through requirement analysis and interface design,the algorithm of detection and segmentation of baggage X-ray image proposed in this paper can be applied.