Detection And Recognition Of Suspicious Objects For Passive Terahertz Human Security Inspection Images

Security inspection and security monitoring of densely populated areas are the most effective way to prevent security incidents.Terahertz imaging technology uses terahertz frequency band for object recognition.It has the characteristics of security,penetrability and fingerprint characteristics.It can distinguish different suspicious objects.Passive terahertz imaging does not require terahertz radiation source.It has a broad and important role in human security inspection.Application prospects.At present,terahertz technology has gradually begun to be used for intensive pedestrian safety inspections in subways,public venues,etc.The passive terahertz imaging speed has reached 10 frames/second.Due to the limitation of production cost and technical process,the terahertz image generated by the passive terahertz imaging system has low signal-to-noise ratio and low resolution.For the identification of suspicious objects,it is difficult to distinguish suspicious objects by security personnel,and the speed is slow.In the aspect of automatic detection,the traditional method uses the detection process of image segmentation,localization and recognition,which is difficult to meet the requirements of rapid security detection accuracy and speed.In recent years,due to the rapid development and application of deep learning technology,the accuracy and speed of image object detection and recognition have been greatly improved.The deep learning-based detection method can not only improve the reliability and efficiency of security inspection,but also quickly identify target suspicious objects that are difficult for human eyes to recognize.The research contents and results of this paper are as follows:(1)The super-resolution reconstruction of passive terahertz security image is studied,and a super-resolution reconstruction model of terahertz security image based on generating confrontation network is proposed and established.Based on the SRGAN algorithm,the algorithm structure and discriminant method are improved for passive terahertz image features to improve the terahertz image quality.(2)Research on passive terahertz human security image detection method based on convolutional neural network,focusing on the Faster-R-CNN algorithm based on region selection and non-region selection SSD algorithm,and comparing them to passive terahertz human security image The performance of object detection and recognition.(3)Design and implement passive terahertz offline modeling and online detection application software system,including passive terahertz human security image detection system architecture,image resource management and offline modeling system,human security image object online detection and recognition system.The passive terahertz offline modeling and online detection application software system developed in this paper,the object detection takes into account the accuracy and speed,the system realizes the online collection and enrichment of the training data set,the offline optimization model and the model update to the online system.closed loop.The experimental results show that with SSD algorithm,the detection accuracy of terahertz human security image objects reaches 74.5%,and the detection speed reaches 12 frames per second.Online detection can meet the requirements of fast security inspection.