Vigor detection of sweet corn seeds by optimal sensor array based on electronic nose

zhang tingting,sun qun,yang lei,yang liming,wang jianhua

DOI: https://doi.org/10.11975/j.issn.1002-6819.2017.21.034

2017-01-01

Abstract:Nondestructive testing equipment is important for the detection of seed vigor. However, there are few studies based on nondestructive testing equipment in sweet corn seed vigor. Therefore, developing an effective and reliable system for the detection of seed vigor has a certain practical significance. As a bionic electronic system, electronic nose(E-nose) detects the vigor of seed qualitatively and quantitatively through the analysis of sample volatile gas’ s fingerprint information. So it is pretty suitable for sweet corn seed detection, though sweet corn seed’s odor is comprised of complicated compositions and small differences exist among seeds with different vigor, which makes the detection difficult. Given that, this paper proposed a monitoring method of sweet corn seed vigor based on E-nose. Five samples of sweet corn seeds with different vigor(germination percentages were 83.3% , 70.8% , 54.2% , 38.4% and 3.8% ) were detected by E-nose. Principal component analysis(PCA) and linear discrimination analysis(LDA) were used to process the data by Winmuster software. The results showed that E-nose could not distinguish the sweet corns with different seed vigor only by PCA or LDA. Then we tried to use support vector machine(SVM) method to detect the seed vigor. The result was pretty good. To further research the feasibility of E-nose application for testing seed vigor, we used loading analysis to find the most useful sensor array. Loading analysis of E-nose sensors indicated that the sensors of W1W, W5S, W1S, W2S, W2W and W3S were found to be more sensitive than other sensors. These sensors might play an important role in the discrimination of samples, which provided a reference for the development of special-purpose sensor systems for sweet corn seed samples in future. According to this result, we reckoned the sensor array was composed of W1W, W5S, W1S, W2S, W2W and W3S to be the optimized sensor array. To verify the validity of optimization, the classification model and the regression model which were built by SVM method were used to compare the ability of discrimination and forecast between the data before and after optimization. Results indicated that sweet corns with different seed vigor were well classified by the optimized array. The accuracies of the training set and prediction set belonging to the classification model based on the whole sensor array by using SVM were 97.10% and 96.67% , respectively, and the time taken by modeling was 30.75 s. However, the accuracies of the training set and prediction set belonging to the classification model based on the optimized sensor array by using SVM were 98.55% and 96.67% , respectively, and the time taken by modeling was 21.81 s. Meanwhile, the result showed that the regression model based on the optimized sensor array(R~2 was 0.984, root mean square error(RMSE) was 3.01% ) performed better than that based on the whole sensor array(R~2=0.913, RMSE=8.50% ). In addition, there was little difference of prediction parameters between the training set and validation set, which meant that the over-fit phenomenon didn’t exist and the forecast ability of the optimized sensor array was better than the whole sensor array. As a result, E-nose technology could be used as a feasible and reliable method for the determination of seed vigor during the storage. The result can provide the theoretical reference for rapid detection of seed vigor during grain storage using volatile odor information.