Single Bean Region of Interest Auto-Detection in a Coffee Beans Image Towards Thorough Quality Inspection

Indonesia was declared the world's fourth-largest coffee bean producer in 2020 and yields consistently increase each year. Indonesia also has the Indonesian National Standard (SNI) 01-2907-2008, which specifies various defects with the defect scores of coffee beans that determine their quality. Common coffee bean defects listed in SNI 01-2907-2008 include black beans, broken black beans, partly black beans, and brown coffee beans. These defects degrade the quality and affect the market value. Previous studies have proposed computer vision systems as a non-invasive testing method for grading coffee beans based on trained images. However, obtaining the detail in a single bean image based on the region of interest could have been more comprehensively explained in those studies, making them less informative and requiring a more detailed conceptual breakdown. This study describes several coffee beans captured as images using a digital camera with a specific field of view, F/25 f-number, 1/60s exposure time, ISO400, and a lighting system reaching 12,000 lux. These images are processed based on the region of interest and centroid, resulting in single bean images with a resolution of 220 x 220 pixels ready for training. The samples used in this study are Robusta Lampung coffee beans, which are commonly found in Indonesia and exhibit various types of defects based on the criteria in SNI 01-2907-2008, such as black beans, broken black beans, partly black beans, and brown coffee beans.