Dangerous Objects Detection In Millimeter Wave Images

As the security situation becomes complicated and citizens attach great importance to their lives and property,conducting safety inspections in important places has become a necessary measure.Currently widely used X-rays have radiation hazards to the human body and can only be checked against carry-on baggage.Usually,human body inspections rely on devices such as metal detectors.However,the detection range of metal detectors is limited,and often requires contact search by the staff.Therefore,there are some problems such as easy to miss,inefficient,and offensive.In recent years,rapid advances in millimeter wave imaging technology have enabled humans to be imaged to detect dangerous objects that are carried around.And millimeter wave does no harm to the human body,has already shown the bright application in the security inspection.At the same time,due to technical limitations,the quality of millimeter wave images is not high at present,and the texture of objects in the images is not clear,which brings great challenges to the detection of dangerous objects.In this situation,there is a great need for an efficient and rapid automated detection method.With the rapid development of convolutional neural networks and great progress in image classification and target detection,this paper explores the classification and detection of dangerous objects in millimeter wave images based on convolutional neural networks.This paper first analyzes millimeter wave images.In order to ease the problem of insufficient training data in millimeter wave images,an image fusion algorithm is proposed to generate new images,which is simple and convenient.The validity of the algorithm is verified in millimeter wave image recognition methods.Then,using the convolutional neural network to classify the dangerous objects in millimeter wave images,and compared with the traditional feature extraction methods,it shows that the convolutional neural network has excellent performance in millimeter wave image recognition.Then,for the detection of common knives and guns in millimeter wave images,this paper proposes a rapid detection model that unifies the classification and positioning.Experiments show that both the speed and accuracy have greatly improved.Finally,in order to identify more types of dangerous goods,this paper proposes a GLNet model,using two different networks to identify and analyze the global and local information in millimeter wave images,and finally combines the outputs of the two networks.Compared with other object detection methods,the accuracy of the model has been greatly improved,and can maintain the performance of real-time detection.