Real Time Ripe Palm Oil Bunch Detection using YOLO V3 Algorithm

The ripeness of the fruit bunch greatly affects the quality of the palm oil. However, to get the matured palm oil bunches, current technology still uses the experience of the harvester in identifying the ripe bunch. The majority of harvesters still use a chisel or long sickle to harvest palm oil bunch from its' tree. They have to determine the ripe bunch from the ground. So, the traditional harvesting method is prone to human error during the determination of the maturity of the palm oil bunch. Thus, this project proposes a real-time ripe palm oil bunch detection using YOLOv3 Algorithm to improve the harvesting process. The preprocessing phase of this project includes data acquisition (collecting palm oil bunch images and videos) and labelling those images based on the respective classes (palm oil maturity level). After the pre-processing phase was completed, the Darknet framework was installed and used for the training and testing phase. The detection model and the dataset were compiled and trained using a pre-trained detection model prepared by the Darknet. A software application was created as a system interface using Python libraries. Tkinter connected it with the Darknet framework using the Darknet command. The outcome of this project's experiment shows that the YOLOv3 Algorithm was able to detect and differentiate the maturity of palm oil bunch's level with learning saturation (overfitting) of 6000th iterations. This project can be integrated with other platforms in the future, such as a mobile application or Internet of Things (IoT).