Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Development of Convolutional Neural Network Based Models for the Prediction of Specialty Coffee Aroma Using Gas Chromatography-Mass Spectrometry
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: 2021 ASABE Annual International Virtual Meeting 2100865.(doi:10.13031/aim.202100865)
Authors: Tzu-Kuan Yu, Yu-Tang Chang, Shu-Ping Hung, Juin-Ming Lu, Jia-Hung Peng, Shih-Fang Chen
Keywords: Specialty coffee, Aroma prediction, Machine learning, Convolution neural network, Gas chromatography-mass spectrometry
Abstract. Aroma is an important quality indicator of specialty coffee. Currently, the aroma quality of specialty coffee is assessed by well-trained personnel, introducing potential subjective cognitive bias in the assessment procedure. This study aimed to investigate the feasibility of building aroma predictive models using machine learning methods on analytical chemical data of coffee aromas. Based on 183 specialty coffee data sampled using headspace solid-phase microextraction (HS-SPME) and gas chromatography-mass spectrometry (GC-MS), this study developed the aroma classification models to identify five targeted coffee aromas using artificial neural network (ANN), convolution neural network (CNN), and residual network (ResNet). The results showed that the model that coupled CNN with the normalized GC-MS data demonstrated the best predictive performance and obtained a 0.799 average accuracy and a 0.536 average recall. The proposed methods provided a promising practice to predict specialty coffee aroma in a relatively systematic and objective way. Further studies will be focused on improving model performances by enlarging the current dataset, and model interpretation methods will be applied to deduce the relation between aromatic chemical compounds and the model prediction mechanisms.
(Download PDF) (Export to EndNotes)