International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056 p-ISSN: 2395-0072
Railway Track Surface Defect Classification Using InceptionV3 and Deep Learning Techniques Rikesh Srivastava1, Abhishek Mishra2, Sachin kumar singh3 r1Student, Dept Of Civil Engineering ,Iet Lucknow India 2
Assistant Professor, Dept Of Civil Engineering ,Iet Lucknow India Assistant Professor, Dept Of Civil Engineering ,Iet Lucknow India ---------------------------------------------------------------------***--------------------------------------------------------------------Abstract- Railway infrastructure forms a critical failures has escalated, posing significant risks to public component of national transportation systems, requiring safety, operational continuity, and asset longevity. Early continuous monitoring to ensure safety and operational detection and timely maintenance of surface-level defects efficiency. Traditional inspection methods are manual, such as cracks, corrosion, debris, and fastener time-consuming, and often fail to detect subtle or degradation are crucial for preventing accidents and emerging surface defects. This study proposes a deep minimizing service disruptions. However, conventional learning-based framework for automated railway surface inspection practices—primarily reliant on manual visual defect classification using the InceptionV3 Convolutional surveys—are inherently labor-intensive, inconsistent, and Neural Network (CNN) architecture. A custom dataset unable to scale across expansive or inaccessible railway comprising 6,500 high-resolution railway track images environments. Moreover, human-based assessments are was curated and labeled into five classes: crack, corrosion, susceptible to errors due to fatigue, subjective judgment, and variable field conditions. debris, fastener failure, and healthy. Transfer learning was applied to fine-tune the InceptionV3 model for domain-specific features, supported by extensive data In recent years, deep learning and computer vision augmentation techniques to enhance generalizability technologies have emerged as powerful tools in the field across varying field conditions. The final model achieved a of structural health monitoring (SHM), offering the potential to automate and optimize defect detection test accuracy of 91.42% with high precision and F1-scores processes with high accuracy and repeatability. Among across most classes, particularly excelling in detecting various convolutional neural network (CNN) corrosion and fastener anomalies. The model also architectures, InceptionV3 stands out due to its efficiency demonstrated robustness to lighting, texture, and surface in multi-scale feature extraction and its proven inconsistencies. Its lightweight architecture makes it effectiveness across diverse image classification tasks. suitable for deployment on edge-computing platforms The architecture's ability to learn complex patterns from such as drones or embedded IoT systems, facilitating real- high-resolution images makes it particularly well-suited time, on-site defect detection. Overall, the proposed for classifying heterogeneous and often visually system offers a scalable, AI-driven solution to enhance ambiguous railway surface defects. structural health monitoring, contributing toward predictive maintenance strategies and the modernization This research presents a comprehensive CNN-based of railway infrastructure in line with Industry 4.0 framework for automated defect classification on railway objectives tracks, utilizing a curated dataset of 6,500 high-resolution images annotated across five classes: crack, corrosion, Key Words: Railway Defects, InceptionV3, Structural debris, fastener failure, and healthy. By leveraging Health Monitoring, Deep Learning, Crack Detection, transfer learning from a pre-trained InceptionV3 model Automated Inspection and applying extensive data augmentation techniques, the framework is designed to achieve robust performance 1. INTRODUCTION across variable lighting, texture, and noise conditions. The final model is not only accurate—achieving a test Ensuring the structural integrity and operational accuracy of 91.42%—but also lightweight, enabling reliability of railway infrastructure is paramount to the deployment on embedded platforms such as drones and safety and efficiency of modern transportation systems. IoT devices for real-time, on-site monitoring. This study With the growing scale of railway networks and contributes to the evolution of AI-driven predictive increasing traffic demand, the frequency of track-related maintenance systems in railway infrastructure, aligning 3
Impact Factor value: 8.315
ISO 9001:2008 Certified Journal