An AI Power Defect Detect In Rail Surface

P.Rama Krishna; J.Diwakar, G.Sai Krishna, Chekuri Dange Manikanta

doi:10.62647/

Authors

P.Rama Krishna Assistant Professor ,Dept of AIML,NRI Institute of Technology Agiripalli,AP, India. Author
J.Diwakar, G.Sai Krishna, Chekuri Dange Manikanta B.Tech Students Dept of AIML,NRI Institute of Technology , Agiripalli ,AP. India. Author

DOI:

https://doi.org/10.62647/

Keywords:

Rail surface defect detection, YOLOv8, YOLOv9, YOLOv5x6, deep learning, SPD-Conv, EMA attention, Focal-SIoU loss, computer vision, real-time detection, railway safety, Flask framework, user authentication

Abstract

Ensuring the safety and reliability of railway transportation requires accurate and real-time detection of rail surface defects such as cracks, scars, and fractures. In this study, we present an improved YOLOv8-based defect detection model enhanced with SPD-Conv building blocks, an Efficient Multi-scale Attention (EMA) module, and a Focal-SIoU loss function, enabling robust recognition of small and densely occluded defects without increasing network complexity. Experimental results demonstrate significant improvements in precision, recall, and average accuracy compared to the baseline YOLOv8 model. To further strengthen the system, advanced YOLO variants including YOLOv5x6 and YOLOv9 were integrated, achieving higher reliability in defect identification. Additionally, Flask-based front-end interface with secure user authentication was developed, providing an accessible platform for real-time defect monitoring and analysis. The proposed framework not only improves detection accuracy but also ensures scalability, usability, and security, making it suitable for deployment in practical railway inspection scenarios.