Deteksi dan Klasifikasi Jalan Berlubang Dengan Mengunakan Metode Faster Regional Convolutional Neural Network (RCNN)
T. Raden Triolan Wijaya, Dr. Danang Lelono, S.Si., M.T.; Dr. Andi Dharmawan, S.Si., M.Cs.
2024 | Tesis | S2 Ilmu Komputer
Roads play a crucial role in developing the potential of a region. Road damage issues are frequently encountered in various areas, including Bantul Regency and Yogyakarta. However, road maintenance at the national level still relies on manual recording by human workers, which is time-consuming. Although pothole detection has been the focus of several researchers, it has generally only progressed to detecting the presence or absence of potholes. Therefore, developing research on pothole detection can help prioritize road repair and maintenance more effectively, replacing manual methods.
The Convolutional Neural Network (CNN) approach has proven successful in object detection analysis, so this research uses the Faster-RCNN development by combining the Region Proposal Network (RPN) method with the Fast RCNN architecture. The use of RPN in Faster-RCNN can produce region proposals directly from the resulting features to overcome the problem of pothole detection. This research aims to develop a pothole detection classification system using the Faster-RCNN method, which can identify various types of road damage, namely Alligator Cracks (RKB), Block Cracks (RB), Longitudinal Cracks (RM), Low Severity Potholes (LKR), Medium Severity Potholes (LKS), and High Severity Potholes (LKT).
The research results show that the Faster-RCNN method has an average precision class value of 99% for all types of damage, with LKS being the most challenging type to detect compared to the other five types. Model evaluation using loss, mAP, Precision, and Recall metrics indicated that at a learning rate of 0.001, the total loss, mAP, Precision, and Recall were 0.6443, 98.7%, 98.9%, and 97.8%, respectively.
Kata Kunci : Convolutional Neural Network (CNN), Detection, Faster Regional Convolutional Neural Network (RCNN), Pothole, Classification