Research On Ray Image Recognition System Based On Convolutional Neural Network

The existing X-ray security inspection equipment can only classify the detected objects,such as organic matter,inorganic matter,mixture,metal,liquid,etc.However,the specific functional classification of the detected objects still needs the naked eye observation and recognition of the security personnel.Owing to the visual fatigue or some other factors,the phenomenon of false detection and missed detection often occurs.Therefore,this subject will take X-ray security inspection image as the research object,complete the function of automatic identification and location of dangerous goods in security inspection image,in order to reduce the phenomenon of missed detection and reduce the pressure of security personnel as the goal,to carry out the research of this subject.Firstly,the convolution neural network and target detection technology are introduced,and the characteristics and training methods of convolution neural network are summarized.According to the actual needs of the subject,a ray image recognition system model based on convolution neural network is proposed.Secondly,we focus on the Faster R-CNN model and improve the RPN network in the Faster R-CNN model.Through collecting,summarizing and labeling security inspection images,a database of dangerous goods in security inspection images is established.The data are trained by using the modified Faster R-CNN network model.After adjusting and optimizing parameters continuously,a security inspection image system model based on Faster R-CNN is formed.The recognition accuracy of the model for small target dangerous goods is improved by about 10%.Finally,using the designed security image recognition model,a software recognition system is designed and implemented,which can be applied to the security inspection machine.When X-ray security equipment scans objects,real-time monitoring of dangerous goods in the detected items,and real-time positioning and identification of dangerous goods.