Research On Explosive Classification Method Based On Compressed Sensing And Deep Learning

Due to the rampant terrorist activities,relevant technologies in explosives detection and identification have been highly valued by various countries.For explosives detection technology,fluorescence-based explosive detection techniques has been used to detect different types of explosives under a variety of conditions and has demonstrated excellent sensitivity and selectivity for specific materials.However,at present,there is no publicly reported explosives identification and classification technology based on fluorescence method.On the base of fluorescein-based explosives detection technology,the article has studied the method of explosives identification and classification based on fluorescence method.The main work contents are as follows:1)On the base of the explosive detection technology,a compressed identification method based on compressed sensing is proposed,which can classify and identify the micro-trace explosives.Firstly,the signal of the explosive fluorescence is analyzed and preprocessed,and then the features of the signal are extracted by the compressive sensing method.The final classifier uses a support vector machine to realize the training modeling,identification and classification of signal features.The feasibility of this method is initially demonstrated by simulation experiments.2)Using the most popular deep learning techniques to identify and classify explosives,a method for explosives identification and classification based on deep neural network(DNN)was proposed.This method also requires that the signal should be preprocessed,and the preprocessed explosive fluorescence signal sample training set to train a multi-layer deep neural network.Deep neural networks have strong explosives identification and classification capabilities through autonomous learning.The experimental results show that the explosive recognition classification method based on deep learning can have better recognition and classification effect than the traditional pattern recognition method(compressive sensing combined with support vector machine).3)On the base of the method of deep learning to identify and classify explosives,an explosives identification and classification system that can rapidly detect and identify explosives is designed.Using Microsoft's official CNTK toolkit,build a deep hidden neural network(DNN)with three hidden layers to train a given training set of explosive samples.Using the optimal network parameters of the DNN model obtained by training to compile an explosives identification algorithm,the explosives identification and classification system was finally generated in the MATLAB environment.Testing he functional of the system and the test results show that the system has achieved the expected function in the actual application environment.The system can accurately detect and identify the explosives classification.The feasibility and effectiveness of the deep-learning-based explosives identification classification method are verified.This paper applies the deep learning technology in the field of explosives identification and classification,solving the problem that it is impossible to identify and classify every type of explosive after explosives detection.It has important practical significance and good commercial application prospects.