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

Area of Science:

• Transportation infrastructure
• Computer vision
• Artificial intelligence

Background:

• Traditional road distress detection methods are manual, costly, and struggle with diverse surfaces and low-resolution data.
• Efficient and accurate road distress detection is vital for safe and reliable transportation systems.
• Existing automated methods face challenges with information loss and gradient issues.

Purpose of the Study:

• To introduce a novel framework, Innovative Road Distress Detection (IR-DD), for enhanced road distress detection.
• To improve accuracy and real-time capabilities in road distress detection for smart city and autonomous vehicle applications.
• To address limitations of traditional methods by optimizing multi-scale feature utilization.

Main Methods:

• Integration of the YOLOv8 object detection algorithm.
• Implementation of bidirectional feature pyramid network (BiFPN) for recursive feature fusion.
• Utilization of bidirectional connections to optimize multi-scale feature utilization and mitigate information loss.

Main Results:

• The IR-DD framework demonstrates superior performance, efficiency, and robustness compared to conventional methods.
• Achieved a precision of 0.666, F1 score of 0.630, and mAP@0.5 of 0.650.
• Operates at a high speed of 86 frames per second (FPS), enabling effective real-time detection.

Conclusions:

• The IR-DD framework offers a cost-effective and compelling alternative for road distress detection.
• The approach significantly advances object detection techniques for practical road infrastructure monitoring.
• The findings highlight the effectiveness of IR-DD in enhancing the safety and reliability of transportation systems.