Development and Testing of Green Coffee Bean Quality Sorter using Image Processing and Artificial Neural Network

Abstract

Currently, the Philippines has no commercially available coffee bean sorter to mechanize the manual sorting, which is prone to human errors. Hence, this study aimed to design and develop a green coffee bean (GCB) quality sorter using various electronic materials for the sorting mechanism, a proportional-integral-derivative (PID)-based algorithm and image processing as sorting system control, and other locally available materials for the machine’s framework. The developed prototype was then evaluated through preliminary testing. A series of tests in three trials were conducted with different sets of Arabica GCBs (T1: 120 good GCBs, T2: 120 defective GCBs, T3: 100 good GCBs + 20 defective GCBs, T4: 20 good GCBs + 100 defective GCBs, and T5: 60 good GCBs + 60 defective GCBs) as test materials. It was shown that the machine can separate defective from the good GCBs arranged in linearity using neural network and image processing. Two webcams were installed to take images of both sides of the bean, which were used for determining the GCB quality through a prediction test. The device was found to be functional with an accuracy of 89.17%, which was comparable with manual sorting. Furthermore, the machine can sort 1 kg of GCBs within 2 h and 45 min. The preliminary tests’ results can be used as reference in designing similar equipment.