Pavement Disease Detection through Improved YOLOv5s Neural Network

Area of Science:

• Computer Vision
• Artificial Intelligence
• Road Engineering

Background:

• Traditional pavement disease detection methods suffer from high computational costs and low recognition rates.
• Efficient and accurate automated systems are needed for road maintenance and safety.

Purpose of the Study:

• To develop an improved Ghost-YOLOv5s detection algorithm for enhanced pavement disease recognition.
• To reduce the computational load while improving the accuracy of pavement disease detection.

Main Methods:

• Integration of Ghost modules and C3Ghost into the YOLOv5s network to minimize floating-point operations (FLOPs).
• Implementation of Mosaic data augmentation to enhance feature expression.
• Reconstruction of a public road disease dataset for model verification.

Main Results:

• The proposed Ghost-YOLOv5s model achieved an average accuracy of 88.17%, a 4.01% increase over existing YOLOv5s.
• The model's frames per second (FPS) reached 12.51, an 184% improvement, indicating faster processing.
• Experimental deployment on NVIDIA Jetson Nano confirmed practical applicability.

Conclusions:

• The enhanced Ghost-YOLOv5s algorithm provides a computationally efficient and highly accurate solution for pavement disease detection.
• The method meets the practical requirements for real-time road surface inspection.
• This advancement contributes to improved road infrastructure management through AI.