A novel manifold discriminant extreme learning machine combined with an E-nose for chili pepper identification via aroma analysis

Abstract

The electronic nose (E-nose) is a bionic sensing technology that simulates the biological olfactory system and is currently applied in various fields. To identify chili pepper varieties and origins conveniently and accurately, in this study, we developed a novel manifold discriminant extreme learning machine (MDELM) classification model combined with an E-nose to analyze the aroma of chili peppers. First, we collected flavor information from different chili pepper varieties and chili peppers of the same variety from different origins via an E-nose. Second, an MDELM classification model is designed by integrating manifold learning, linear discriminant analysis and maximum variance theory into a unified extreme learning machine framework. Third, we conducted extensive comparative experiments on three chili pepper odor datasets. The experimental results showed that MDELM achieved classification accuracies of 90.40 %, 87.60 %, and 98.80 % on the three datasets, outperforming the other six comparison models, which exhibited excellent performance in identifying chili pepper varieties and origins. Finally, ablation experiments and early recognition experiments were conducted, which indicated that each module of the model improved the model classification performance and that the MDELM can effectively complete early identification tasks for chili pepper odors via an E-nose.