Realization Of Display And Control Unit And Research On Target Recognition Algorithm Of Terahertz Passive Imaging System

Based on the passive imaging mechanism,the terahertz passive detection imaging security system can detect hidden dangerous objects in the human body.During the detection process,the system receives the electromagnetic waves of the frequency band emitted by the human body and the background through the antenna,instead of emitting electromagnetic waves,to use the bright temperature difference to image.Although the resolution of the image is lower than that of the optical image,it is safer and clearer than that of the X-ray image.As the central hub of the imaging system,the display and control unit undertakes the responsibility of coordinating the work of each part in the system and human-machine interaction.Besides,the research of the recognition algorithm is the embodiment of the intelligence of the system.Taking the actual scientific research project as the support point,this paper carried out the research on the realization of the display and control unit and the target recognition algorithm based on deep learning,focusing on the working principle of the terahertz imaging system and the problems such as low resolution and small number of samples in terahertz images.The main research contents are as follows:(1)Based on the imaging principle of passive imaging system,the structural framework of terahertz passive detection imaging system is analyzed.The communication interface between the display control unit and each module of the system is determined by analyzing the connection and communication mode between each part of the system,and the functional modules of the display control unit are divided according to the practical application requirements.This paper realizes the graphic user interface design of the display control unit and the software development of each functional module,and verifies the feasibility of the display control unit in the engineering application through simulation and project stage experiment.Therefore the friendly man-machine interaction is realized.(2)In order to solve the problem of insufficient samples in the target identification sample base of terahertz images,a Target Identification Based on Transfer Learning(TI-TL)method was proposed.The network pre-training was conducted through handwritten data sets,After that,apply the established target samples to the pre-training network to start the formal training.By this way,the problem of overfitting caused by insufficient samples is reduced and effectively improve the accuracy of target recognition.(3)Aiming at the real-time demand of recognition algorithm in security check scene,lightweight target recognition networks based on fully connected network and convolutional neural network are proposed respectively.In the fully connected recognition network,effective means such as dropout and batch normalization layer are adopted to reduce overfitting,thus achieving high accuracy and fast target recognition and classification.The convolution recognition network uses the idea of small convolution instead of large convolution to optimize the number of network parameters.At the same time,to solve the problem of overdose parameters of the full connection layer in the traditional convolutional neural network,a convolution kernel of 1?1 is proposed to replace the traditional full connection layer,which significantly reduces the number of network parameters,effectively improves the network operation speed,and finally realizes accurate and efficient identification of samples.The feasibility of the display control unit and the recognition algorithm is verified by simulation results and experimental data.