Surface Krack Detection Using Machine Learning and Deep Learning

Abstract

The Surface crack detection plays a crucial role in ensuring the structural integrity and safety of materials across various industries, including construction, manufacturing, and transportation. This study presents an automated approach for detecting surface cracks using advanced image processing and machine learning techniques. High-resolution images of material surfaces are analyzed using edge detection, thresholding, and morphological operations to identify potential crack regions. Furthermore, convolutional neural networks (CNNs) are employed to enhance detection accuracy and distinguish between cracks and other surface anomalies. The proposed method significantly reduces inspection time and human error, offering a reliable and scalable solution for real-time surface defect monitoring.

Introduction

1. Introduction

Manual inspection for detecting surface cracks in infrastructure is labor-intensive, error-prone, and inefficient. To overcome these limitations, automated crack detection systems using deep learning (DL) and image processing have been developed. These systems aim to accurately detect and classify cracks in various materials such as concrete, metal, and wood, even under diverse surface textures and lighting conditions.

2. Project Objective

The main goal is to create a robust, automated crack detection model that:

• Accurately identifies and classifies cracks.

• Works across various materials and environments.

• Uses deep learning models to improve efficiency and minimize manual labor.

• Integrates with infrastructure maintenance systems to enhance safety and decision-making.

3. Literature Review

Deep learning techniques such as CNNs, ResNet50, and hybrid models (CNN + VGG) are used for feature extraction and classification. Key challenges in existing research include:

• Lack of generalization across real-world conditions.

• Scalability issues for large-scale deployment.

• Poor real-time performance in the field.

• Insufficient environmental robustness (e.g., to weather or dust).

• Limited workflow integration and lack of uniform evaluation metrics.

4. Problem Statement

Manual analysis of surface data (or reviews) is inconsistent and slow. There's a need for intelligent systems that can automatically detect surface cracks using ML/DL with high accuracy.

5. Dataset Description

• Data Type: High-resolution images with and without cracks.

• Classes: "Cracked" and "Non-Cracked."

• Annotations: May include bounding boxes or segmentation masks.

6. Methodologies

• CNN (Convolutional Neural Network): Extracts hierarchical features from images using convolutional and pooling layers.

• ResNet50: Deep residual network used for spatial feature extraction and crack classification.

7. Preprocessing Techniques

• Image Acquisition: Captures high-quality images using cameras or drones.

• Image Enhancement: Techniques like Histogram Equalization, CLAHE, and Gamma Correction to improve image quality.

• Noise Removal: Filters such as Gaussian Blur and Median Filtering to remove artifacts and preserve edges.

8. Feature Extraction

The system analyzes input images to detect cracks using trained DL models. It acts as a standalone or integrated tool for real-time monitoring or post-processing.

9. Machine Learning Models Used

• SVM: Good for edge and texture-based features.

• Random Forest: Handles noise well.

• KNN: Suitable for small datasets.

• Logistic Regression: Simple baseline classifier.

10. Deep Learning Models Used

• Architectures: VGG16/VGG19, ResNet50, InceptionNet, MobileNet.

• Focus: Classify image regions as “cracked” or “non-cracked” using advanced neural networks.

11. Results and Evaluation

• Best Model: ResNet + GRU + LSTM.

• Accuracy: 99.89% on test data.

• Precision & Recall: Both 1.00 for cracked and non-cracked classes.

• F1 Score: 1.00 across both classes.

• Confusion Matrix:

  • Class 0: 899 correct, 1 false negative.

  • Class 1: 899 correct, 1 false positive.

12. Evaluation Metrics

• Accuracy: Measures total correct predictions.

• Classification Report: Includes precision, recall, F1-score, and support.

• Confirms that the model performs exceptionally well across both classes.

Conclusion

In this study, we addressed the critical need for an automated solution for surface crack detection in infrastructure maintenance. Leveraging deep learning techniques, particularly Convolutional Neural Networks (CNNs), ResNet, and MobileNet, we aimed to accurately identify and classify cracks in various materials like concrete, metal, and wood. Our approach integrated advanced image processing methods to handle diverse surface textures and lighting conditions effectively. Through comparative analysis, we found that CNN and MobileNet outperformed ResNet in terms of crack detection accuracy. This comparison underscores The significance of choosing suitable deep learning frameworks for particular tasks. The success of our endeavour holds the promise of significantly reducing maintenance costs, preventing structural failures, and enhancing the safety and integrity of critical infrastructure. By enabling proactive maintenance strategies, our robust deep learning models pave the way for long-term sustainability in vital infrastructure systems. The scope for future of automated surface detection of crack using DL and AI is bright with opportunities for innovation. DL-based models trained on large datasets promise unprecedented accuracy across various materials and environments, leveraging CNNs. Integration of AI techniques like RL and active learning enhances efficiency, while fusion with other sensing modalities like LiDAR and infrared imaging improves reliability. Edge computing solutions enable real-time inference, ensuring privacy and data security. Ongoing research in explainable AI addresses transparency, bridging the gap between AI models and human operators. In summary, advancements in model accuracy, efficiency, multi-modal fusion, edge computing, and explain ability will drive the development of next-generation crack detection systems, enhancing infrastructure safety and sustainability.

References

[1] Surface crack detection using deep learning with shallow CNN architecture for enhanced computation B Kim, N Yuvaraj, KR Sri Preethaa… - Neural Computing and …, 2021 – Springer [2] Machine learning for crack detection: Review and model performance comparison YA Hsieh, YJ Tsai - Journal of Computing in Civil Engineering, 2020 - ascelibrary.org [3] Automated vision-based crack detection on concrete surfaces using deep learning RS Rajadurai, ST Kang - Applied Sciences, 2021 - mdpi.com [4] Automated bridge surface crack detection and segmentation using computer vision-based deep learning model J Zhang, S Qian, C Tan - Engineering Applications of Artificial Intelligence, 2022 – Elsevier [5] Development of deep learning model for the recognition of cracks on concrete surfaces TT Le, VH Nguyen, MV Le - Applied computational intelligence and soft …, 2021 - hindawi.com [6] Automatic crack classification and segmentation on masonry surfaces using convolutional neural networks and transfer learning D Dais, IE Bal, E Smyrou, V Sarhosis - Automation in Construction, 2021 – Elsevier [7] A comprehensive review of deep learning-based crack detection approaches Y Hamishebahar, H Guan, S So, J Jo - Applied Sciences, 2022 - mdpi.com [8] Structural crack detection using deep convolutional neural networks R Ali, JH Chuah, MSA Talip, N Mokhtar… - Automation in …, 2022 – Elsevier [9] Computer vision framework for crack detection of civil infrastructure—A review, D Ai, G Jiang, SK Lam, P He, C Li - Engineering Applications of Artificial …, 2023 [10] Automatic pixel-level crack detection for civil infrastructure using Unet++ and deep transfer learning Q Yang, X Ji - IEEE Sensors Journal, 2021 - ieeexplore.ieee.org

Copyright

Copyright © 2025 A Ramya, Mr. M Navalan . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.