Researches On The Technology Of RMB Paper Money Identification And Value Recognition Based On TMR Sensor

With the economic development and the international status improvement of our country,RMB note issuance and circulation is increasing at home and abroad, but counterfeit money alsoappears and the number and types of it is increasing in recent years. There are a lot of cash inflowsand outflows everyday in banks, railway stations and other places. In order to discriminate a widerange of high imitation counterfeit money, as well as to improve banknote value recognitionaccuracy and rapidity, on the on hand, national mint departments continue to improve and increasethe RMB banknote machine-readable features, it is urgent for financial machinery manufacturingenterprises to upgrade and improve banknote identification and value recognition technology. Asthe important machine-readable feature, the reasearches on the techonlogy of RMB banknoteidentification and value recognition based on magnetic signals is great of importance in the currentand long time.The TMR sensor is a new magnetoresistive sensor which has a high sensitivity, temperaturecharacteristics and strong output signal, it is suitable for detecting banknote magnetic signals. Inview of this case, TMR magnetic sensors are used to detected magnetic banknote magnetic signalson the detector platform. Banknote identification and value recognition technology researchprogram which is based on magnetic signals is designed. TMR magnetic sensors are used to detectand collect magnetic signal of characters and safety line, in order to achieve data compression andreduce data redundancy, wavelet transform technology is used to reduce noise, the method ofenergy difference and signal threshold are used to extract useful signal, according to the charactersof two signals and research goals, by analyzing magnetic signals features, features of characters andsafety line are extracted respectively, the nearest neighbor algorithm and BP neural networkalgorithm which are based on magnetic signal of characters are designed to identify true banknotes,the support vector machine algorithm that is based on features of safety line is designed to identifythe value of banknote, and on this basis, grip research method, particle swarm optimization, andgenetic algorithm is used to optimize support vector machine parameters to improve recognitionaccuracy. According to GB requirements about banknote identification and value recognition, thefollowing research programs are designed and made:(1) Banknotes magnetic signal preprocessing and feature extraction technology researchIn view of the fast and high target about about banknote identification and value recognition onthe detector, noise, redundant and useless data will lead the original signal to be distorted, this leadsto accuracy issues, banknotes magnetic signal preprocessing and feature extraction technology research program is to be designed, the program reduce noise and eliminate redundant and uselessdata, retain and implement effective signal feature extraction and data compression, greatlyimproving the accuracy and rapidity of banknote identification and value recognition. Firstly, thenoise performance of mean filter, median filter and wavelet transform technology is studied;Secondly, the accuracy of the energy ratio method, the energy difference and the threshold methodis studied; Finally, the time-domain characteristics of characters and safe line magnetic signals areextracted, statistics threshold range, according to threshold to select data simple characteristics,providing a data source for subsequent banknote discrimination and value recognition.(2) Banknote identification algorithm design based banknote characters magnetic signalGiven some financial sensors can not detect obvious magnectic signals and some detectors cannot discriminate counterfeit banknotes such as “patchwork money” and “added magnetic substance”,the TMR sensors are used to detect magnetic signals of character, and banknote discriminationalgorithms are designed on this basis, this method can discriminate these banknotes in offlinesimulation. Firstly, the nearest neighbor algorithm and BP neural network algorithms are designedto identify banknote; Secondly, the accuracy and reliability which are based on character magneticsignals are demonstrated in linear and nonlinear angles. Finally, simulation results show that: undercurrent conditions, the nearest-neighbor algorithm and neural network algorithm can be used toidentify banknotes, but given the limitations of the experimental data，the diversity of counterfeitmoney and the advantages of BP neural network algorithm, the preferred neural network algorithmis used to identify notes.(3) RMB banknote value recognition algorithm based on the magnetic signal of safety lineGiven financial TMR sensors can detect stable and clear security thread magnetic signals,some detectors can not count banknotes and display banknote value exactly, and some detectorswhich are based on image recognition technology run slowly and cost much, TMR sensors are usedto detect security thread magnetic signals and banknote vaue recognition algorithm is designed andimproved, this research program solves these above problems and achieves rapid and accuraterecognition effect. Firstly, to train SVM classification model and set parameters based onexperience, this model is to be used to recognize banknote value; Secondly, analysis theclassification performance of grid search algorithm, particle swarm optimization algorithm andgenetic algorithm; Finally, the experimental simulation results shows that the GAalgorithm not onlyhas a good convergence, high accuracy, short computing time and high robustness, but also hashigher prediction accuracy rate.Banknote identification and value recognition technology research program which is based onmagnetic signals is designed. According to this program, a magnetic signal detection data acquisition experimental platform is built, the TMR sensors are used to detected magnetic signalson the exprimental platform, wavelet transform technology is ursed to reduce noise, effective signaland features are to be extracted, the threshold of these features is to be stored, the feature sampledatabase is to be established according to the characteristic features, the BP neural networkalgorithm based on the magnetic signal of character is designed to discriminate banknotes, itsaccuracy is100%, the SVM algorithm based on security thread magnetic signal is designed torecognize vulue, its accuracy is98.3%. The simulation results show that banknote identification andvalue recognition technology research program which is based on magnetic signals is feasible andcorrect, meanwhile it provides theoretical and experimental basis for nanknote identification andvalue recognition technology based on magnetic signal which is in the detector and other financialmachineries.