Research On Signal Processing And Pattern Recognition Based On The Enose

It is well known that, the hardware of electronic nose can not make abreakthrough in the short time. But software can achieve updates or improvement inshort term. The software algorithm can provide high recognition accuracy and lowrecognition time, which can compensate the error caused by hardware andenvironmental uncertainty. Pattern recognition algorithm is the key technology ofelectronic nose software.Electronic nose system usually consists of sensor array module, signalprocessing module and pattern recognition module. In the present case, the systemhardware technology is difficult to get a big break, many researches focused on theelectronic nose signal processing and pattern recognition methods of research.The subject that makes use of TGS8series sensor, U120816model capture card,PC to build an electronic nose hardware system; software system uses MATLAB andLabVIEW to make program. The program has the function of electronic nose sensorsignal acquisition, analysis, and processing and pattern recognition.Stochastic resonance is actually due to the together effect of the input signal,noise and nonlinear systems so that part of the noise energy is converted to the signal.When the input noise reaches a certain value, due to the synergy of noise and inputsignal makes the output signal amplitude is greater than the input signal amplitude. Itplayed an effective role in the output signal amplitude amplification. The signal tonoise ratio in this intensity of noise can be identified as a new features forclassification. In this paper, the theory of stochastic resonance is used for electronicnose signal feature extraction. For the edible oil problems in China, this paper willbe based on the theory of stochastic resonance in an electronic nose system forinferior oil check, to distinguish between different concentrations of poor-quality oil.The experimental results show that the theory of stochastic resonance for the electronic nose signal feature extraction can be overcome the interference caused bythe sensor baseline drift, to avoid signal loss problems results in the noise suppression.Ensure the stability and reproducibility of the signal acquisition, while the results ofthis research also provide a reference for poor-quality oil detection technology.Currently, most part of the research about the pattern recognition in theelectronic nose system based on the hypothesis that the type of testing samples hasbeen appeared in the training samples. For the identification problem in electronicnose which the samples are unknown type, this paper proposes to make use ofK-means clustering algorithm to improve RBF algorithm to achieve the dynamicclassification of the pattern recognition. It has incremental learning and onlineadjustable output node network capabilities to ensure that the network has highgeneralization ability and a certain ability to learn new classification scheme.Through inspect and identify several different types of vinegar, verify the validity ofthe proposed method.