Klasifikasi Tingkat Kematangan Buah Sawit Berbasis Deep Learning dengan Menggunakan Arsitektur Yolov5

Abstract

Object identification and recognition in the field of computer vision is undergoing rapid development and is applied to various fields, ranging from industry to the health sector. This is reflected in the amount of research conducted, including a focus on the application and personalization of machine learning, as well as the development of new models to solve specific problems and challenges. In the palm oil industry, fruit maturity is divided into two categories, namely immature and ripe. Traditionally, fruit maturity is determined visually by experienced workers based on the number of fruits falling from the bunch or the color of the bunch. However, this technique has disadvantages such as the reduced amount of oil when many fruits fall from the bunch and the subjective assessment of fruit color. Therefore, the purpose of this research is to create an oil palm fruit maturity classification system based on YOLO v5. The dataset used consists of 1500 photos and the annotation data is created with roboflow. The final result is divided into three categories, namely ripe, immature, and rotten. The YOLOv5s algorithm was used to train the dataset. Based on the model estimation results, mAP reached 92%, accuracy reached 97%, and recall reached 96%. The last step is real-time system testing.