Research On Detection And Recognition Algorithm For Hidden Target In Passive Detection Imaging

Because the terahertz passive detection technology uses the difference in the radiated brightness temperature of the terahertz frequency band caused by the different material characteristics to image,the detection system itself does not actively radiate electromagnetic waves,and it has the advantages of no radiation,transparent imaging,and non-contact detection.This technology can provide high-resolution imaging to detect hidden small targets,so it has important application value in the field of security inspection and monitoring.Target detection and recognition technology,as one of the key technologies,plays an important role in the overall performance of the terahertz passive detection system.Based on actual scientific research projects,this thesis has carried out research on the detection and recognition algorithm of hidden targets in terahertz images.First,the preprocessing research of terahertz image is carried out,which mainly includes temperature drift correction and denoising processing of terahertz image,which provides a good image basis for subsequent detection and recognition tasks;furthermore,for target detection and recognition in security inspection and surveillance applications The technology requires high real-time and high accuracy,and the corresponding detection and recognition methods have been proposed.The main contents of the thesis are as follows:(1)The basic theory of Planck's blackbody radiation is introduced.On this basis,the theory of radiation measurement and imaging principle for passive detection of terahertz are analyzed and discussed.(2)Analyzed the general structure of the terahertz passive detection imaging system;combined with the working mode of the imaging system,analyzed and discussed the causes of temperature drift and fringe noise in terahertz imaging,and carried out related preprocessing of terahertz images accordingly Methodological research.An improved temperature drift correction algorithm based on two-point calibration is proposed,which effectively solves the temperature drift phenomenon caused by the long imaging time of single-channel scanning.Aiming at the problem of fringe noise in terahertz images,several traditional de-striping algorithms are compared and analyzed,and on this basis,a de-striping noise algorithm based on mixed domain is proposed,and it is verified that the algorithm has better performance in de-streaking noise.Effect.(3)Aiming at the high real-time and high accuracy requirements of target detection and recognition in the field of terahertz security inspection and monitoring,a target detection algorithm based on the attention mechanism(AM-RSSD-Single Shot Multi Box Detector with Res Net based on Attention Mechanism).The algorithm is improved based on the "SSD-Single Shot Multi Box Detector"(SSD-Single Shot Multi Box Detector).Aiming at the characteristics of terahertz images,it has carried out lightweight feature extraction network,introduced attention mechanism,improved loss function and advanced Test the adaptive adjustment of the bounding box generation strategy;finally,the problem of small sample set training is solved through transfer learning.Experiments have verified that the algorithm can still meet the requirements for rapid detection and recognition of targets in terahertz images under the configuration conditions of lower computing power,and has a higher accuracy rate and a lower false alarm rate.