Estimasi Jumlah Buah Kelapa Sawit untuk Mendukung Manajemen Produksi Berbasis YOLOv5 TensorFlow Lite

Abstract

This research aims to develop a mobile application capable of detecting and counting oil palm fruit bunches automatically using the YOLOv5 object detection method. The dataset used in this study consists of 226 images of oil palm fruit bunches collected directly from PT Eagle High Plantation located in East Kalimantan. The dataset was annotated using bounding box techniques and processed using Roboflow to perform data augmentation, increasing the dataset size to 856 images. The model training process was conducted using YOLOv5 to detect oil palm fruit bunches in images and produced a precision value of 0.859, recall of 0.673, mAP@0.5 of 0.767, and mAP@0.5:0.95 of 0.542. The trained model was then converted into TensorFlow Lite format and integrated into a mobile application developed using the Flutter framework. The system allows workers to capture images of oil palm fruit bunches and automatically calculate the number of detected objects directly on mobile devices without relying on server-based processing. Firebase services are also integrated for authentication, data storage, and verification by supervisors. The implementation of this system is expected to improve efficiency and transparency in recording oil palm harvest data.