Development Of Colorimetric Sensor Array System Combined With Image Processing Analyses For Food Volatile Compounds

Odor is an important attribute of food and agricultural products.By detecting the content of volatile organic compounds(VOCs)of food it can help to analyze and distinguish food quality,authenticity,shelf life and so on.The colorimetric sensor array technique is a new method for detecting gases.This technique utilizes the reaction between volatile materials to be detected and an array of chemical dyes to conduct the qualitative and quantitative analysis of the VOCs according to the sensor array's color changes.Besides,the information of gas can be represented intuitively in the form of pictures.In the application of colorimetric sensor array technique,the distribution of VOCs in the reaction chamber and the image processing and analysis of sensors have a high impact on the detection accuracy.In view of this,this study mainly designed and optimized a new reaction chamber,compared some algorithms and programs in the image processing process.Finally,experiments to detect and distinguish vinegar and rice with different storage time were carried out.The main research work is as follows:(1)A reaction chamber with an arc baffle embedded in front of the inlet was designed,and the gas distribution was simulated by COMSOL Multiphysics software.The simulation results showed that the shape of the baffle was a key factor affecting the gas distribution.The position of the baffle could also affect the distribution of the flow,but its influence was not great compared with the former factor.When the curvature of the baffle was too small,the gas velocity in the middle part of the reaction chamber was high as well as the velocity on both sides was low.Meanwhile,gas was mainly distributed in the middle area.The situation would reverse when the baffle arc was too large.The airflow velocity on both sides of the sensor array became smaller,if the baffle was too close to the inlet.The baffle too far away from the exit port could cause the speed in the middle to become smaller and the gas distribution near the outlet wider.The velocity field of the gas in the reaction zone was relatively well-distributed when the curvature was 3.3 and he vertical distance from the front end of the baffle to the inlet was 1.5 cm.The specific velocity values of the regions corresponding to colorimetric sensors were extracted from velocity fields of reaction chambers with COMSOL software.The velocity distribution of the reaction chamber after optimization of the baffle was relatively uniform.At the same time,the size and material of the reaction chamber were designed.To compare new chamber with the chamber for free volatilization of VOCs,bromocresol green was chosen as sensor material to react with acetic acid because of its ability to react remarkably and rapidly with acid gases.Experimental results showed that the mean values of the characteristic components under the new chamber were relatively larger and the coefficients of Variance generally smaller indicating that the reaction in the new chamber was more uniform and rapid.(2)Different methods of localization of the target image center,selection of characteristic region and color space model were compared and optimized in the image processing part.Firstly,methods of one-order moment,minimum enclosing rectangle and ellipse fitting were used to seek center coordinates of target regions.Results showed that the center coordinates obtained by the three methods were very similar and the computation and time consuming of the one-order moment algorithm were less.Secondly,to choose the characteristic region,central and peripheral layers of a sensor domain segmented through flood fill algorithm and some specific-sized circular regions were compared.Results showed that it was more reasonable to choose the peripheral region as the characteristic region.Lastly,feature extractions of the characteristic region in RGB,HSV and Lab color spaces were compared.Results showed that the standard deviations in the Lab color model were significantly smaller,especially those of ?a and ?b components.It can be concluded that feature extraction in the Lab color model had higher stability.(3)The colorimetric sensor array technique was applied to detect and distinguish the product storage time with vinegar and rice as the research object.In order to identify vinegar samples with different ages(new vinegar,and one to four years)in the experiment,the characteristic data from RGB,HSV and Lab color spaces was analyzed with principal component analysis(PCA)and linear discriminant analysis(LDA).The recognition accuracy rate achieved highest in Lab color space.The best identification of the vinegar age also further validated the superior performance of the Lab color space.Then,in the experiment to detect vinegar samples with different ages,the new reaction chamber and the chamber for free volatilization of VOCs were compared.Analysis results of PCA and LDA showed that under the new reaction chamber,the training set and prediction set were both 95%.Meanwhile,the recognition rates were generally higher than those under the chamber for free volatilization of VOCs with the same PCs.It can be concluded that the former model was more stable and accurate.Finally,in the experiment to detect the smell of new rice and stored rice,the two different reaction chambers were both applied.Combined with PCA it can be seen that the new and stored rice samples were well distinguished with the recognition rate of 100%.And the result further validated the feasibility of the new reaction chamber.