ASABE Technical Library - Abstract

Abstract.

An embedded metal oxide semiconductor (MOS) electronic nose was designed for the detection of Chinese pecan quality in this research, and initial sensors were chosen according to GC-MS results of pecan volatiles. After less sensitive sensors were excluded, the remaining sensors were selected to generate the initial sensor array. Then its feature matrix was obtained through three feature extraction methods, namely mean-differential coefficient value, stable value, and response area. Features with poor discrimination were eliminated via the mean analysis, and features with poor stability were then ruled out through the variation coefficient analysis. After that, remaining features were classified through the cluster analysis, and features with less redundancy of each class were selected according to results of the correlation coefficient analysis. To verify the validity of the optimization, principal component analysis (PCA) was used to test the classification ability based on the optimized feature matrix, and also partial least squares regression (PLSR) and back-propagation neural networks (BPNN) were applied to compare between the prediction ability of the non-optimized feature matrix and the optimized one. The result indicated that pecans with different aging time were differentiated distinctly, and each batch of pecans clustered closely in the PCA score plot. Meanwhile, the regression model established by PLSR and BPNN showed a better prediction ability of the optimized feature matrix compared with the non-optimized. As a result, the optimization of the sensor array was not only conducive to simplifying the data dimensionality, but also to improving the performance of the e-nose.