Coffee Bean Inspection Machine with Deep Learning Classification

Abstract

In coffee production process, after harvest often find defective coffee beans mixed with normal coffee beans. So a coffee bean inspection machine is made to classify the defects in a short time and high efficiency. The conveyor is controlled by the Raspberry Pi 4 and the camera mounting on the conveyor is for capturing images and uploading on the google drive. Then the images are analyzed by using deep learning 3 models consisting of Enhanced, ResNet-50 and AlexNet model for training image, validation image and test image. The most efficiency model is ResNet-50, which has an accuracy of 93.33%. That means it can classify the defect accurately and save working time to classify the coffee beans. From other research, coffee beans were capture image when the coffee beans place in groups. But in this research, The coffee beans are feed out of the feed machine. Run automatic shooting from a real production line through a moving on belt. and analyzed through the highest accurancy at 93.3%. Based on other literature comparisons, the accuancy was similar in the range.