Research And Development Of Embedded Millimeter Wave Detection System

Similar to the principle of optical imaging,millimeter wave imaging is realized by reconstructing the millimeter wave echo received by the millimeter wave radar.Since millimeter wave can penetrate clothing and be reflected by skin,and the required radiation dose is harmless to human,active millimeter wave imaging and target detection technology can be organically combined to realize the automatic detection of hidden contraband on human body surface.With the continuous development of millimeter wave imaging and target detection technology,millimeter wave detection system,as a safe and efficient non-contact intelligent security system,has gradually entered the airport,high-speed rail and other application fields.At present,millimeter wave detection systems developed by domestic and foreign companies all adopt deep learning schemes.However,due to the characteristics of millimeter wave images,the actual generalization ability of the detection algorithm is poor,which requires a large scale of training samples.Therefore,the training set needs to be constructed at high cost.In addition,this kind of solution requires the detection terminal to be equipped with GPU and related hardware facilities with high computing power,which leads to high cost and large size of the equipment,which is not conducive to the maintenance and promotion of the detection system.In view of the above software and hardware problems,this paper carries out the research and development of embedded millimeter wave detection system by relying on actual scientific and technological cooperation projects.The main contents are as follows:(1)In view of the problems of weak generalization ability and high dependence on sample size of existing deep learning based millimeter wave image contraband detection algorithms,this paper reverts to classical image target detection ideas and proposes a millimeter wave image contraband detection model guided by saliency detection,which is composed of "contraband detection module" and "detection correction module" cascade.The contraband detection module consists of four cascading submodules: image noise reduction,balance enhancement,saliency detection and confidence processing.The Image noise reduction submodule removes the noise in the image background;The saliency detection submodule firstly estimates the position and size of the image foreground by using the saliency detection method based on compressed sensing and frequency domain analysis.Then combined with the balance enhancement submodule,a balance enhancement method based on visual attention is proposed to improve the detection rate of contraband.The confidence processing submodule completes the confidence processing by integrating the contraband detection results from multiple angles of human body,and finally obtains several candidate regions and outputs them to the detection correction module.The detection correction module is composed of two submodules: specificity correction and correctness correction.The specificity correction submodule uses the proposed specificity correction method based on the binary mask of human body region to correct the possible black background of the image in the candidate region to improve the effectiveness of correctness correction.The correctness correction submodule uses the reliability calculation method of the proposed candidate region to calculate the probability that the input candidate region contains the contraband target,and selects the candidate region according to the reliability threshold,and outputs the final detection result.(2)Based on the above algorithm model,an embedded millimeter wave detection system based on ARM is designed and developed in view of the problems of high equipment cost and huge volume of the existing millimeter wave detection system.The hardware of the system was developed based on TI AM5728,with A dual-core ARM Cortex A-15 processor as the processing unit,and the output and interaction were completed by an LCD touch screen.The software of the system is based on C++ 14 and Open CV 3,and Qt Creator 5.6.2 is used to develop the system graphical user interface.Due to the small amount of computation in the proposed algorithm model,the real-time requirement of detection tasks can be satisfied only by using CPU calculation,thus realizing the lightweight of detection equipment and significantly reducing the cost of system hardware.Experimental data and actual tests of the system show that the system has low dependence on sample size and strong generalization ability,and can complete the task of detecting contraband in millimeter wave image in real time under the condition of small sample size.