Implementasi Algoritma Yolo Untuk Mendeteksi Jalan Berlubang dan Retak

Abstract

Roads are transportation facilities intended for traffic as well as public infrastructure that supports the mobility and accessibility of road users. Road damage is commonly found and is often identified manually. Since the invention of Computer Vision in the 1950s, object detection and classification have attracted significant interest in various sectors, including industry and medicine. Since then, many studies have been conducted using various Deep Learning algorithms capable of detecting objects. This research utilizes the YOLOv8 (You Only Look Once) algorithm to detect and classify images. This algorithm works by predicting bounding boxes and class probabilities on the entire image in a single capture. In its application, the data is divided into training, validation, and testing datasets, consisting of images of damaged roads categorized into cracks and potholes. By applying specific configurations in the YOLOv8 algorithm, the resulting output includes the Confusion Matrix calculations. The research involves analyzing results from various data train and validation splits: 70%-30%, 80%-20%, and 90%-10%, with training epochs of 25x, 50x, and 100x, to evaluate how training iterations affect performance outcomes. The results indicate that the highest confidence level is achieved when using a 90%-10% split between training and validation data, reaching 97% confidence, with mAP of 93.2%, F1-Score of 88.7%, Recall of 90.8%, and Precision of 86.7% at the 100th epoch.