Research On Cigarette Detection System Based On Wavelet Packet RBF Neural Network

Due to the special nature of tobacco control in China,criminals often make fake brand cigarettes and sell them in order to obtain large profits.In order to improve the efficiency of counterfeit and smuggled cigarettes,cigarette detectors came into being.In this paper,based on the research of eddy current identification metal theory,combined with wavelet transform and RBF neural network,this paper proposes an automatic identification method of metal types using eddy current and wavelet packet RBF neural network:1)The principle of eddy current generation and the main factors affecting eddy current are analyzed.Firstly,the principle of eddy current generation on metal conductors is analyzed.Secondly,since eddy currents are formed only in a thin layer on the surface of the conductor,its distribution range is analyzed.Finally,the effects of different frequencies,different metals,different coil-to-metal distances on eddy currents were studied.2)In order to improve the recognition rate of metal species,a method based on wavelet packet RBF neural network and metal species recognition based on RBF neural network is proposed.Wavelet packet RBF neural network is divided into two types of combined wavelets:"loose" and "compact" Neural network,using three different kinds of metals to carry out MATLAB simulation of these three algorithms,the results show that the use of wavelet packet based RBF neural network method can better identify the type of metal,because the method extracts wavelet transform The eddy current feature quantity is used as the input of the neural network to build the RBF training model,so the recognition rate is better.3)Design a set of eddy current metal type recognition system based on single chip computer,which includes hardware and software system.The hardware system part includes STM32F407ZGT6 single-chip minimum system,power supply circuit,oscillation circuit,amplifying circuit,frequency measuring circuit and ranging circuit design.The software system part is completed in two parts.First,the benchmark data is obtained in the MATLAB platform,then input to the single-chip program,and finally the single-chip program completes the function of the detector.In this paper,the use of wavelet neural network when the simulation results are obtained can greatly improve the recognition rate of metal species,and also make the efficiency of cigarette detectors in work.