CD Skripsi
Identifikasi Tandan Buah Segar Kelapa Sawit Berdasarkan Varietas Dura Dan Tenera Menggunakan Pencitraan Komputer Dan Deep Learning
Oil palm is a primary commodity in Indonesia’s plantation industry, with the production of fresh fruit bunches (FFB) increasing annually. However, the process of identifying FFB varieties such as Dura and Tenera is still performed manually using axes in a destructive manner, which is inefficient and damages the surface of the fruit. Non-destructive imaging methods have emerged to address this issue. This study aims to develop a system for identifying oil palm FFB varieties using computer vision and the YOLOv8 algorithm to distinguish between Dura and Tenera varieties. A total of 40 FFB samples were used, categorized based on two ripeness levels. The system was developed through image acquisition using an RGB camera and a conveyor integrated with a laptop. Dataset preparation began with annotation and image segmentation using Roboflow software, followed by data augmentation. Model training and testing, along with RGB intensity analysis of the FFB, were conducted using the YOLO object detection algorithm in Python. The model was trained on 1,200 augmented images derived from the 40 samples. The results showed a precision of 0.97 and a recall of 0.955. Model detection accuracy validation using a confusion matrix achieved a [email protected] of 0.91. RGB analysis indicated that unripe fruits have lower intensity compared to ripe ones. This system demonstrates strong potential for real-time FFB sorting in the palm oil processing industry.
Keywords: Computer Vision, Oil Palm Fresh Fruit Bunch, Dura and Tenera variety, ripenes level, YOLOv8 algorithm
Tidak tersedia versi lain