摘要
采用由12个金属氧化物传感器组成阵列的电子鼻,对3种等级碧螺春茶的茶水和茶底气味进行检测。筛选了9个对茶叶挥发香气敏感的传感器阵列,并分别提取了各个传感器所获取茶水和茶底数据的最大值、最小值和平均值作为特征变量,进行主成分分析,再利用K最邻近(KNN)和误差反向神经网络(BP-ANN)对数据进行分析与识别。KNN结果显示,采用茶水和茶底特征融合信息对不同等级碧螺春茶的识别效果较茶水和茶底更佳,KNN模型对独立样本的判别率达到83.33%;设计拓扑结构为10-7-3的BP-ANN模型对信息融合的茶叶样本判别率则达到了100%。
Three grades of Biluochun green tea solution and wet tea-leaf were measured by the electronic nose which was composed of twelve metal oxide semiconductor gas sensors. Twenty-seven feature parameters included the maximum, mini- mum and average of every data curve of sensor arrays were extracted based on the selected 9 gas sensors from all of them. K-nearest neighbors（KNN） and back-propagation neutral networks （BP-ANN）were used in the pattern recognition after principle components analysis （PCA）. The analytical result show that the correct data of the samples based on fusion of the ccharacteristic variables of green tea solution and wet tea-leaf was better than those of others＇ from KNN model and the best correct classification was up to 83.33%. The topology of BP-ANN is designed with 10-7-3. 100% correct classifi- cation was achieved for all the tea samples based on fusion of the ccharacteristic variables using the BP-ANN.
出处
《农机化研究》
北大核心
2012年第11期133-137,共5页
Journal of Agricultural Mechanization Research
基金
国家自然科学基金项目(30971685)
江苏省高校优势学科建设工程资助项目(PADA)
