基于电子鼻技术的草莓损伤检测系统的开发

[目的]本文旨在研发一款用于草莓损伤检测的专用电子鼻系统。[方法]该系统主要包含软、硬件两大部分，其中硬件部分包含由6个金属氧化物传感器（MOS）构成的传感器阵列、密闭气室、采集电路及洗气净化装置；软件部分则分为数据采集模块和数据分析及模式判别模块。以‘红颜’草莓品种为研究对象，静态提取草莓果实的顶部空气，传感器接触敏感气体分子产生电信号，经模数转换成数值输送至电脑（PC）端后，由数据分析及模式判别模块对试验数据进行预处理、特征提取及模式识别，最终根据所建模型获得判别结果，从而实现对损伤果实的正确识别。[结果]该系统实现了电子鼻小型化、便携式的设计。传感器均对草莓的挥发性气味有不同的响应信号，且损伤草莓与完好草莓的挥发性气味存在差异，表现出不同的"指纹图谱"。利用支持向量机（SVM）方法对损伤草莓进行建模，验证试验表明，该模型具有较好的适用性，其判别准确率为92.5%。[结论]研发的电子鼻系统简易、便携，所建立的草莓损伤检测模型具有较高的识别率，对损伤草莓的区分具有可行性，同时也为草莓损伤的检测提供了一种小型化无损检测装置。
[Objectives] This study was to develop a special electronic nose system to discriminate the bruised strawberries in a quick and non-detective way. [Methods] The electronic nose system included hardware and software. The hardware system was designed using four main components, which included an array of sensor, a sensor chamber, an acquisition circuit and an air washing system. The sensor array was comprised of six metal oxide semiconductor(MOS)sensors. The software system consisted of data acquisition unit, and data analysis and pattern recognition unit. In this study, ‘Hongyan’ strawberry was used as test materials, the headspace volatile of single strawberry was collected and detected by MOS sensors, which was especially selected. Then the reaction between the odorants and sensors triggered electrical signals like voltage change which were further converted to digital output signals using an 8-channel and 12-bit high precision analog-digital converter. After that, the resulting signals were transmitted to a computer data acquisition platform, which provided functions of real-time data storage and display. The data pre-processing and feature extraction were processed by data analysis and pattern recognition unit. The support vector machine(SVM)method was utilized to build the classification model, and the radial basis function was used as kernel function. The unknown pattern of volatiles was compared with this model, and based on model the bruised strawberries could be discriminated. [Results] A novel, small, portable electronic nose system was developed. All six sensors used in this electronic nose system presented positive changes to volatile compounds of strawberry samples. Besides, obvious differences in volatiles between bruised strawberries and intact ones were observed, which illustrated the feasibility of using our designed electronic nose to discriminate bruised strawberries. A total of 26 support vectors were used to build the SVM model, which was much lower than that of samples from the training group. Based on SVM model, an overall of 92.5% classification accuracy was