| Summary: | Cracking is one of the concerns that might affect the road condition in a bad way like causing accidents. Early detection of cracks improves the process of road maintenance where undesirable problems might be avoided and prevented. In general, the crack detection process is accomplished by human examination. However, this manual technique is tedious, time-consuming, arduous, and risky. This study proposed an autonomous crack-detecting system using Unmanned Aerial Vehicle (UAV) images using deep learning. It focuses on federal roads in Malaysia which is the road used to connect state capitals. A deep learning approach called You Only Look Once (YOLO) is applied to do the detection. Prior to the detection process, numerous image preprocessing techniques are implemented to prepare and increase the amount of dataset (images) for the training purpose. These stages are crucial for the data preparation in deep learning, to boost the deep learning detection performance and generalization ability. A YOLOv4 is constructed using a MATLAB environment and trained to detect cracks from the provided UAV images. Statistical measures like precision, recall, F1-Score, and Average Precision, are utilized to examine and evaluate the effectiveness of the suggested strategy. Two altitudes were utilized which are 10 m, and 20 m to maneuver the UAV. Simulation findings reveal that the suggested strategy obtained 82.02% of Average Precision (AP) at 10 m height for YOLOv3 and 87.98% at 10 m height for YOLOv4. Therefore, it can be stated that the YOLOv4 outperformed YOLOv3 in detecting pavement cracks from UAV images at 10 m height. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
|
|---|