Abstract:
Identifying the level of maturity of Fresh Fruit Bunches is one of the steps in the palm oil
harvesting process. The level of maturity of fresh fruit bunches is an important factor that
determines the quality of palm oil production. In general, farmers can check FFB
maturity manually by observing directly. Determining FFB maturity manually requires
time and the expertise of a skilled farmer to ensure proper maturity. Faster R-CNN is a
machine learning method that can be used to detect palm oil FFB objects quickly and
accurately. By creating a detection system designed with UML and then using the
Tensorflow 2.4 library for the Faster R-CNN Model, the model can recognize the shape,
color and edges of each TBS training data. Using 8590 TBS image datasets consisting of
6872 training data and 1718 test data trained in 50,000 steps with 64 batch sizes. The
model can identify six levels of maturity of oil palm consisting of unripe, mature 1, mature
2, mature 3, late ripe, and empty bunches. The model training results have an accuracy
of 72% with an IoU of 0.75 then an accuracy of 59% with an IoU of 0.50:0.95 and an
accuracy of 86% with an IoU of 0.50
