食品挥发气味嗅觉可视系统的研制及图像处理研究

发布时间：2018-03-01 01:14

本文关键词： 挥发气味可视化传感器检测反应室图像处理　出处：《江苏大学》2017年硕士论文　论文类型：学位论文

【摘要】：气味是食品、农产品的重要属性,可以通过挥发性气体的检测实现食品的品质、真伪、货架期等的分析和区分。嗅觉可视化传感器技术是检测气体的新技术,它利用色敏材料制作的传感器阵列与待检测气体发生反应时传感器的颜色变化,对挥发性有机气体(VOCs)进行定性或定量分析,并且将气味信息以直观可视的图像方式表达出来。在嗅觉可视化技术检测食品品质中,挥发气体在反应室内与传感器的分布与接触状况以及传感器的图像信号处理和分析对检测精度有很高的影响。基于此,本研究设计并优化了气体分布反应室,对图像处理流程中的部分算法和方案进行了比较研究,以实现不同存储时间食醋和大米气体的检测和区分。论文的主要研究内容如下:(1)设计了一个在进气口前嵌入弧形挡板以起到引流效果的反应室,利用COMSOL Multiphysics仿真软件对反应室进行流场模拟。仿真效果显示挡板的弯曲程度是影响气体分布比较关键的因素,挡板位置虽然也会影响反应室中气流的分布,但是与前者相比,其影响并不是很大。当挡板的弯曲程度大时,会导致反应室中间的气体流速大,两侧的气体流速小,且气体主要分布在中间通道区域,四周的气体分布较少;当挡板的弯曲程度过小时,情况相反。当挡板靠近进气口,会导致传感器阵列两侧气体流速变小;当挡板远离出气口,则会使色敏传感器阵列中间的气体流速变小,同时,气体在靠近出口部分的分布较广。经过反复仿真模拟本研究确定当挡板曲率为3.3,挡板前端距离进气口的垂直距离为1.5cm时,仿真的效果最好。通过COMSOL Multiphysics提取传感器阵列各区域的速度进行比较,同样表明挡板形状和位置优化后的反应室更能够保证反应室中各区域流场均匀。本研究同时对反应室的部件和整体尺寸、材料进行了设计。利用溴甲酚绿和乙酸的显色反应,将加工制作完成后的反应室,与模拟气体自由挥发环境的反应室进行试验比较,结果显示有挡板反应室下,提取的颜色区域相应各分量特征的平均值一般较大,变异系数偏小,说明前者中乙酸气体与传感器阵列的反应更加充分,其均衡性也较好。(2)对图像处理流程中的图形中心点定位、特征区域选取、颜色空间模型的选择进行了比较和分析。首先,应用一阶矩、最小外接矩形、椭圆拟合等方法,来求目标图形的中心点坐标,结果显示,三种方法都能满足实际应用时的需要,而第一种方法的计算量和耗时相对较少。其次,分别对应用中心定位取圆形区域为特征区域,与漫水填充法分割传感器不同颜色分布层为特征区域进行了对比,结果显示靠近传感器中心的颜色层并不能很好反应传感器反应前后的特征,而选择剔除中心颜色层的外围区域为特征区域更为合理。最后,对基于RGB、HSV和Lab三种颜色模型提取的传感器显色区域特征数据进行了比较分析,结果显示,Lab颜色空间模型时,特别是在颜色分量Δa和Δb下,其特征变量的标准差较小。相较于RGB和HSV颜色模型,在Lab颜色模型下提取的特征值具有更高的稳定性。(3)应用设计的嗅觉可视化系统及图像处理方法,分别以食醋和大米作为液态和固态食品的代表,对产品的储藏时间进行了检测和区分。首先,在检测新醋和醋龄1~4年的食醋时,分别在RGB、HSV和Lab颜色空间下提取特征数据,并结合主成分分析(PCA)和线性判别分析(LDA)等,比较了不同颜色模型下醋龄检测的识别率。结果显示,在颜色模型为Lab,LDA模型对食醋年限的鉴别效果最好,这也进一步验证了Lab颜色空间的较优表现。其次,在利用嗅觉可视化系统检测不同年限食醋时,分别应用了新反应室和模拟气体自由挥发环境的反应室,PCA和LDA分析的结果显示,在新反应室下,当主成分数为8时,模型的训练集和预测集的识别率均达到95%,且相同主成分数时,识别率一般高于在模拟气体自由挥发环境的反应室,这说明前者的模型稳定性和准确性更高。最后在新反应室和模拟气体自由挥发环境的反应室下,对新米和陈米气味进行检测,结合PCA可知,前者可以100%地区分出两种不同的大米,这也进一步验证了所设计的新反应室的可行性。
[Abstract]:The smell is an important attribute of food, agricultural products, through the detection of volatile gas to achieve the food quality, authenticity, analysis and distinguish the shelf life of the sensor. The olfactory visualization technology is a new technology for gas detection, it uses a sensor array making color sensitive material changes color sensor and the tested gas react, to volatile organic compounds (VOCs) qualitative or quantitative analysis, and the odor information in the way of image visual expression. The olfactory visualization technology in food quality detection, gas distribution in the reaction chamber and the sensor and the contact condition and the image sensor signal processing and analysis have great influence on the detection accuracy based on. This, the research on the design and optimization of the distribution of gas reaction chamber, in the process of image processing algorithm and scheme are studied, in order to realize Detection and differentiation of vinegar and rice with storage time of gas. The main research contents of this paper are as follows: (1) a design of the front of the air inlet baffle to embedded into the reaction chamber drainage effect, the flow simulation of reaction chamber using COMSOL Multiphysics simulation software. The simulation results show that the factors which affect the bending degree of baffle the key of gas distribution, baffle position although also can influence the gas flow distribution in the reaction chamber, but compared with the former, the impact is not great. When the bending degree of baffle is large, will cause the gas flow velocity in the middle of a large reaction chamber, on both sides of the gas flow rate is small, and the main gas distribution area in the middle channel. The distribution of less surrounding gas; when the bending degree of the baffle is too small, the opposite is the case. When the baffle near the inlet gas flow rate on both sides of the sensor array can lead to smaller; when the baffle is away from the outlet. The gas flow rate intermediate color sensor array becomes small, at the same time, the gas distribution in the near exit portion is wide. The study determined when the baffle curvature is 3.3 repeated simulation, the vertical distance from the front end of the air inlet baffle distance is 1.5cm, the best simulation results. Through the COMSOL extraction of Multiphysics sensor array in each region the speed comparison also shows that the reaction chamber baffle shape and position optimized more to ensure the regional reaction chamber. The uniform flow field and the research on the reaction chamber components and the overall size, materials are designed. Using the chromogenic reaction of bromocresol green and acetic acid, will be processed after the completion of the reaction chamber, for compared with the reaction chamber to simulate the gas free volatile environment, results show that the baffle reaction chamber, the corresponding color region extraction component average value is generally large, variation The small coefficient that acetic acid gas sensor array and the former reaction more fully, the balance is also good. (2) the image graphics processing center location in the process of feature selection area, color space selection of models were compared and analyzed. Firstly, based on the first order moment, the minimum bounding rectangle. The ellipse fitting method, coordinates, Center for target figure shows that the three methods can meet the needs of practical application, and the computation of the first method and time consuming less. Secondly, the application of positioning center take circular area for feature regions, different color segmentation sensor distribution layer feature compared with the regional flood fill method, results show that near the center of the color sensor layer is not very good characteristics before and after reaction and the choice of sensor response, peripheral region excluding center color layer for the feature area Is more reasonable. Finally, based on RGB, HSV and Lab three kinds of color model to extract the sensor data are compared and analyzed, the color region shows that the Lab color space model, especially in color components a and B, the characteristics of the variable standard deviation is smaller than RGB and HSV. Color model, feature extraction in the Lab color model value has higher stability. (3) the olfactory visualization system and image processing method is applied to design, vinegar and rice were used as liquid and solid food, the product storage time were tested and identified. First of all, in the detection of new vinegar vinegar at the age of 1~4 years and vinegar, respectively in RGB, feature extraction data of HSV and Lab color space, and combined with principal component analysis (PCA) and linear discriminant analysis (LDA), compare the recognition models of different colors under the old vinegar detection rate. The results showed that the color model For Lab, the best identification effect LDA model for vinegar age, this further proves the better performance of Lab color space. Secondly, in the detection of different years of vinegar using olfactory visualization system, are applied to the reaction chamber and the reaction chamber of new simulation of gas free volatile environment, analysis of the PCA and LDA results showed that in the new reaction chamber, when the concentration is 8, the model training and prediction sets, the recognition rate reached 95%, and the same number of principal components, the recognition rate is generally higher than that in the reaction chamber to simulate the gas free volatile environment, it shows that the stability and accuracy of the former is higher. Finally in the new reaction chamber the reaction chamber and Simulation of gas free volatile environment, to detect the new meters and the smell of rice, with PCA shows that the former 100% can distinguish two different types of rice, which further proves that the new reaction chamber design is feasible .

【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS207.3;TP391.41

【参考文献】