Rapid determination of the roasting degree of cocoa beans by extreme learning machine (ELM)-based imaging analysis

Abstract

The determination of the levels of roasting of cocoa relies on expensive analytical equipment, sensory panel, and, in the cases of small processors and growers, empiricism. In this study, cocoa beans were roasted for 10–40 min to obtain different levels of roasting, and the images of the beans were captured by a smartphone camera. An extreme learning machine (ELM)-based algorithm was developed to predict the roasting degree of cocoa beans using the images of the cocoa bean cross-sections. A 22-dimension feature vector, including color and texture features, is extracted from each sample, and a total of 350 samples are used to train an ELM network. A majority rule-based voting method was used to make the decision. Experimental results showed that the proposed method achieved a classification accuracy of 93.75%. GC-MS analysis was conducted to determine the chemical compounds in the raw and roasted cocoa beans, and enrichment analysis, principal components analysis, partial least-squares–discriminant analysis, and Pearson correlation analysis were conducted to identify major chemicals respond to roasting time and classify the cocoa beans samples. Caffeine and theobromine were identified as primary chemical compounds that responded to roasting time, and cocoa beans with different levels of roasting were successfully classified.

Keywords

• Extreme learning machine
• Cocoa bean
• Roasting
• Classification