Oil Palm Fresh Fruit Branch Ripeness Detection Using YOLOV6 Algorithm

Malaysia is one of the largest palm oil producers and exporters in the world. As a result, the efficient production of palm oil plays a major role in the Malaysian economy. Currently, oil palm fresh fruit branches (FFBs) are harvested by human graders at oil palm plantations based on the fruit surface colour and the number of loose fruits on the ground as a measure of the ripeness level. However, solely relying on human graders would result in misclassified bunches due to factors such as the position of the FFBs on the trees, the height of the trees and incorrect count of the number of loose fruits on the ground. Therefore, this study proposed an automated oil palm FFBs ripeness detection via a deep learning algorithm, namely You Only Look Once Version 6 (YOLOV6). YOLOV6 is used to detect oil palm FFB ripeness levels based on the surface colour of the oil palm fruits. Real images of FFBs at oil palm plantations are taken to be classified into four classes: unripe, underripe, ripe, and overripe. Each class differs from other classes of oil palm FFBs based on their colour. The datasets are trained using two YOLOV6 models which are YOLOV6s (small) and YOLOV6m (medium). The proposed algorithm performance is analysed based on the precision, recall, F1 score, mean average precision, training time, inference speed, frames per second and non-maximum suppression time. The YOLOV6m model outperformed the YOLOV6s model, according to the investigation. For the YOLOV6m model using 100 training epochs, the achieved precision, recall, f1 score, mAP(50), and mAP(50–95) is 36.9%, 30%, 33.1%, 36.9%, and 16.5%, respectively. The YOLOV6m model automatically detected oil palm FFBs accurately with a predicted bounding box including objectness score for all the four classes of palm oil fresh fruit branches which includes unripe FFB, underripe FFB, ripe FFB, and overripe FFB, respectively. It is expected that the finding of this study can be applied as a real implementation at an oil palm plantation in future.