Track: Modeling and Simulation

Abstract

The harvest of oil palm fresh fruit bunches (FFB), especially in Indonesia, is determined manually, mainly based on the color and number of fallen fruit. Information technology innovation can help determine harvest time quite accurately and consistently. Computer visuals are one method that predicts can detect oil palm fruit maturity. The importance of the ability to detect objects directly, especially in oil palm plantations, is certainly a challenge. YOLOv4 can observe any changes indirect objects properly and quickly and even has a very deep model structure making it very suitable for detecting bunch maturity. FFB has a maturity level which is referred to as a fraction. There are 3 fractions analyzed in YOLOv4 to fraction 1, fraction 2, and fraction 3. Based on the study results, the YOLOv4 method was able to detect the maturity level of bunches of fraction 1, fraction 2, and fraction 3 with an accuracy of mAP@0.50 of 99.17% mAP@0.75 of 97.08% at the checkpoint weight of 6000. In study has disadvantaged unbalanced data set due to the difficult terrain in the field and the lack of resources due to the pandemic. The result model learning can develop to predict ripeness fruits bunch real-time with smartphone or Raspberry.