人工神经网络对电子鼻性能的影响

电子鼻原型由4个气体传感器组成的阵列和人工神经网络识别软件组成,可识别不同品牌的白酒.以它为例,研究了3种人工神经网络,即反向传输网络(BPN)、学习矢量量化网络(LVQ)和概率神经网络(PNN)对电子鼻性能的影响.结果表明,在需要精细识别时,虽然传感器阵列对白酒的响应谱的差别是电子鼻识别的基础,但是人工神经网络结构和算法包括相关训练参数的选择对决定电子鼻的性能也有重要的作用.比较而言,学习矢量量化网络在分类能力和训练成本方面更胜一筹,而概率神经网络则在计算负载和易用性方面更好一些.

Effect of Artificial Neural Network on Performance of Electronic Nose

A prototype of electronic nose consisting of an array based on four gas sensors and artificial neural network(ANN) recognition software was constructed for identifying different brand of spirits.As an example,the effect of three type of ANN including back propagation network (BPN),learning vector quantization network (LVQ) and probabilistic neural network (PNN) on the performance of the electronic nose was investigated.The results indicate that for a fine recognition,the difference (among) the response spectra of sensor array to spirits is the base of recognition,but the ANN architecture and corresponding algorithm including the choice of relative parameters while training also play an important role in determining the performance of electronic nose.By comparison,LVQ network was better than the others in classification ability and training cost,and PNN network in computation load and easy use.