Road Anomaly Detection Utilizing Swin Transformer and Deep Convolutional Neural Networks with YOLOv8

The detection and timely maintenance of potholes are paramount to ensuring road safety and minimizing transportation-related expenses. Potholes pose a significant hazard to vehicular travel, leading to increased accidents and vehicle damage, which, in turn, demand substantial economic investment for repairs. Despite extensive research and the development of numerous models, challenges persist in achieving high-accuracy detection in diverse environments. This study introduces two neural network architectures to address this issue. The Swin Transformer, which uses shifted window-based self-attention, achieves 97.64% accuracy in pothole detection. In contrast, the Deep Convolutional Neural Network (DCNN) with YOLOv8, a leading real-time object detection algorithm, reaches an accuracy of 98.75%. The analysis highlights the individual strengths of each model and the superior performance of the DCNN-YOLOv8 combination in accuracy, making it a viable candidate for real-world applications in road maintenance and infrastructure assessment. The success of these models underscores the potential of leveraging complex neural network architectures to enhance automated systems in image recognition tasks. Consequently, this research provides a significant stepping stone toward the evolution of automated pothole detection systems, with the potential to greatly influence future advancements in road safety and infrastructure technology.