Implementation of Slicing Aided Hyper Inference (SAHI) in YOLOv8 to Counting Oil Palm Trees Using High-Resolution Aerial Imagery Data
Naufal Najiv Zhorif, Rahmat Kenzie Anandyto, Albrizy Ullaya Rusyadi, Edy Irwansyah
Abstract
Palm oil is a commodity that contributes significantly to Indonesia's national economic growth, with a total plantation area of 116,000 hectares. By 2023, Indonesia is projected to produce approximately 47 million metric tons of palm oil. One of the major challenges in the manual counting of oil palm trees in a large area of a plantation is the labour-intensive, time-consuming, costly, and dangerous nature of the work in the field. The use of aerial imagery allows for the mapping of large areas with comprehensive data coverage. This study proposes a method of mapping oil palm plantations for the counting of oil palm trees using high-resolution aerial images taken with drones. Furthermore, the use of artificial intelligence (AI) methods and deep learning (DL) with the You Only Look Once (YOLO) model for object detection has demonstrated good accuracy in previous studies. This research will utilize the YOLOv8m object detection model and the slicing method, namely Slicing Hyper Aided Hyper Inference (SAHI), which is anticipated to enhance the precision of object detection models on high-resolution aerial imagery. The study concluded that the use of the SAHI slicing method can significantly enhance the accuracy of the model, as evidenced by a Mean Absolute Percentage Error (MAPE) value of 0.01758 on aerial imagery equivalent to 73.2 hectares, with a detection time of 5 minutes and 45 seconds.