Determinants of Pepper Quality Based on the Percentage of Foreign Objects Based You Only Look Once (YOLO)

Abstract

The presence of foreign objects in pepper seeds is one of the things that affect the quality of pepper seeds. Farmers in Bangka sell pepper to pepper collectors. The collectors in this area still inspect the pepper using manual methods without the help of other tools, so there are still foreign objects such as dry leaves or pepper stalks. This method is often inefficient because the precision of each person is different. In this case, we propose to determine the quality of pepper based on the percentage of foreign objects automatically in accordance with the determination of pepper quality standards regulated in the national quality standard (SNI). The authors use YOLOv3 for object detection which is one of the fastest and most accurate object detection methods, outperforming other detection algorithms. However, YOLOv3 requires a heavy computer architecture. Therefore, YOLOv3-tiny, a lighter version of YOLOv3, can be a solution for smaller architectures. This study found that YOLOv3-tiny model has a reasonably high network performance value: precision value of 0.99, recall value above 70%, and F1 score above 80%. While determining the quality of pepper according to the standard quality of pepper (SNI) the value obtained must be below 2%. Then a comparison was made between the detection system and the manual calculation of objects. It was found that in the sample of 26 pepper seeds, the system detected 8.97 seconds faster than manual calculation.

Keywords

• pepper
• yolov3
• tinyyolov3
• image detection