嗅觉神经网络在电子鼻识别多品牌绿茶中的应用研究

Discrimination of Green Tea with Different Brands by Electronic Nose and an Olfactory Neural Network

生物模式识别机理引入人工嗅觉系统将提高其仿生化程度,并被认为是有前途的传感阵列信息处理方法。本文尝试将一种嗅觉神经网络应用到电子鼻检测和识别多种品牌的绿茶气味。通过包含8个MOS型气敏传感器的自制电子鼻仪器,测量了来着不同地方的5种不同品牌的绿茶样品,在传感阵列信号稳态部分提取特征向量,并使用雷达图考察指纹图谱异同,验证传感阵列及特征提取方法的有效性。采用生物相似性学习算法训练该神经网络,考察了样本训练次数和识别率的关系,发现经过4～7次训练,该网络对这5种绿茶的识别率平均值都在97%以上。

Introducing some principles of biological pattern recognition into artificial olfaction will produce more biomimetic instrument, and they are also considered as promising sensor array information processing methods. This paper presents an electronic nose based on an olfactory neural network to detect and discriminate green tea with different brands. Data of five brands of green tea from different areas were acquired through a homemade electronic nose apparatus containing eight MOS gas sensors. Feature vectors extracted from steady-state phase of array signals, were plotted on radar charts as fingerprint, to inspect the effectiveness of sensor array arrangement and feature selection method. The olfactory neural network trained by a biologically plausibility learning rule was used to classify those samples. The relationship between training cycles and recognition rate was investigated. The results showed that after 4 -7 training cycles the network could correctly discriminate those green teas with average recognition rate of above 97%.