Innovative road distress detection (IR-DD): an efficient and scalable deep learning approach

Abstract

In the rapidly evolving landscape of transportation infrastructure, the quality and condition of road networks play a pivotal role in societal progress and economic growth. In the realm of road distress detection, traditional methods have long grappled with manual intervention and high costs, requiring trained observers for time-consuming and expensive data collection processes. The limitations of these approaches are compounded by challenges in adapting to diverse road surfaces and handling low-resolution data, particularly in early automated distress survey technologies. This article addresses the critical need for efficient road distress detection, a key component of ensuring safe and reliable transportation systems. Effectively addressing these challenges is crucial for enhancing the efficiency, accuracy, and safety of road distress detection systems. Leveraging advancements in object detection, we introduce the Innovative Road Distress Detection (IR-DD), a novel framework that integrates the YOLOv8 algorithm to enhance the accuracy and real-time capabilities of road distress detection, catering to applications such as smart cities and autonomous vehicles. Our approach incorporates bidirectional feature pyramid network (BiFPN) recursive feature fusion and bidirectional connections to optimize the utilization of multi-scale features, addressing challenges related to information loss and gradients encountered in traditional methods. Comprehensive experimental analysis demonstrates the superior performance, efficiency, and robustness of our integrated approach, positioning it as a cost-effective and compelling alternative to conventional road distress detection methods. Our findings demonstrate the superior performance of our approach compared to other state-of-the-art methods across various evaluation metrics, including precision, recall, F1 score, and mean average precision (mAP) at different intersection over union (IoU) thresholds. Specifically, our method achieves notable results with a precision of 0.666, F1 score of 0.630, mAP@0.5 of 0.650, all while operating at a speed of 86 frames per second (FPS). These outcomes underscore the effectiveness of our approach in real-time road distress detection. This article contributes to the ongoing innovation in object detection techniques, emphasizing the practicality and effectiveness of our proposed solution in advancing the field of road distress detection.