International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 08 | Aug 2025
p-ISSN: 2395-0072
www.irjet.net
DETECTION AND CLASSIFICATION OF SURFACE DAMAGE IN CONCRETE USING EFFICIENTNET-B1 Priyansh Verma¹, Prof. Abhishek Mishra², Prof. Sachin Kumar Singh³ ¹M.Tech Scholar, Department of Civil Engineering, I.E.T Lucknow, India ²Assistant Professor, Department of Civil Engineering, I.E.T Lucknow, India ³Assistant Professor, Department of Civil Engineering, I.E.T Lucknow, India ---------------------------------------------------------------------***--------------------------------------------------------------------goal is to create a robust, mobile-compatible inspection tool Abstract - Concrete structures often experience surfacefor real-time structural damage assessment.
level deterioration due to environmental, mechanical, or chemical exposure. Manual inspection methods for identifying cracks, spalling, and corrosion are laborintensive and subjective. This paper proposes an automated image-based classification system using EfficientNet-B1, a convolutional neural network model optimized for high performance and low computational cost. A curated dataset of 5032 labeled images across four categories—Healthy, Crack, Spalling, and Rust—was used. Transfer learning was applied, and the model was trained using augmented data in Jupyter Notebook. The trained model achieved an accuracy of 95.4% on the test set, with high precision and recall for all damage classes. These results indicate that EfficientNet-B1 is suitable for real-time concrete surface assessment applications in retrofitting and structural health monitoring.
2. LITERATURE REVIEW Concrete surface damage detection using deep learning has gained significant attention in recent years due to the success of Convolutional Neural Networks (CNNs) in image classification. Several studies have explored different models and architectures for automating damage identification. Zhang et al. [1] used a basic CNN architecture for binary classification of concrete cracks and reported an accuracy of 89.2%. However, their model did not generalize well across other types of surface damage such as spalling or rust. Li et al. [2] applied MobileNetV2 to classify cracks in lightweight environments, achieving an accuracy of 91.5%. Their model was optimized for mobile deployment but lacked robustness in detecting other defects under varying lighting and surface conditions.
Key Words:
Concrete Damage, Crack Detection, Spalling, Rust, EfficientNet-B1, Deep Learning, Structural Health Monitoring, Transfer Learning
1.INTRODUCTION
Reddy et al. [3] used VGG16 for detecting cracks in concrete and achieved high classification accuracy, but the training process was computationally expensive, and overfitting was observed on small datasets.
Concrete is the most extensively used construction material worldwide, favored for its strength, durability, and adaptability. However, surface-level deterioration such as cracking, spalling, and corrosion is inevitable over time. These damages are early indicators of more significant structural distress. Manual inspection remains the primary assessment method but is prone to subjectivity and inefficiency, especially in large-scale infrastructures.
Tan and Le [4] proposed the EfficientNet family, which uses compound scaling to balance network width, depth, and resolution. EfficientNet-B1 was shown to provide a favorable trade-off between accuracy and computational efficiency. It outperformed deeper models like ResNet-50 while being lightweight enough for real-time applications.
Recent advancements in computer vision and deep learning offer scalable and objective solutions. Convolutional Neural Networks (CNNs) have demonstrated promising performance in detecting and classifying structural defects. However, many conventional CNNs require large datasets and extensive training time, which limits their practical implementation.
Singh et al. [5] applied transfer learning on EfficientNet-B0 for crack detection in concrete and achieved an accuracy of 94.2%. However, their dataset only included binary classes (crack and no-crack), limiting its applicability for real-world retrofitting scenarios. The present study expands on this prior work by implementing EfficientNet-B1 for multi-class classification, covering cracks, spalling, rust, and healthy concrete surfaces. It addresses the need for a practical and scalable tool for comprehensive surface condition monitoring.
This study presents an approach using EfficientNet-B1, a lightweight and scalable CNN model, for the automated classification of concrete surface damage into four categories. The model is trained using transfer learning on a dataset of 5032 labeled images and evaluated using standard metrics like accuracy, precision, recall, and F1-score. The
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 159