A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Lightweight Deep Convolutional Neural Network for Pavement Crack Recognition with Explainability Analysis
Pavement crack detection is an integral part of maintaining pavements. Identifying the pavement areas that require maintenance through manual processes is a laborious and time-consuming task. Latest researches have employed machine learning and deep learning to devise automatic and semi-automatic solutions for pavement distress detection. In this research, we propose a lightweight deep convolutional neural network (DCNN) model for crack recognition in pavement images, with 20 times fewer parameters than ResNet34 model. The model was tested on three public datasets that include Concrete Crack Images for Classification (CCIC) dataset, Sharif University of Technology Crack (SUT-Crack) dataset, and bridge crack detection dataset. The model obtained an average F1-score of 99.98%, 98.75% and 100% on the above-mentioned datasets respectively. We also tested the model on our own dataset that has pavement images with finer cracks which are harder to detect. Our dataset consists of 5,506 images in total. The model obtained 89.13% average F1-score. We explored the transfer learning technique on the proposed model as well. We fine-tuned the model (pre-trained on CCIC dataset) on our dataset which improved the results by 6.34% and increased the average F1-score to 95.47%. The model shows good generalizability with just a little over a million trainable parameters. Model explainability analysis using Captum AI library confirmed that the model is focusing on the cracked sections of the pavement images to classify the images as cracked. Having relevant areas of the image contribute more toward the prediction increases the confidence in the model.
Lightweight Deep Convolutional Neural Network for Pavement Crack Recognition with Explainability Analysis
Pavement crack detection is an integral part of maintaining pavements. Identifying the pavement areas that require maintenance through manual processes is a laborious and time-consuming task. Latest researches have employed machine learning and deep learning to devise automatic and semi-automatic solutions for pavement distress detection. In this research, we propose a lightweight deep convolutional neural network (DCNN) model for crack recognition in pavement images, with 20 times fewer parameters than ResNet34 model. The model was tested on three public datasets that include Concrete Crack Images for Classification (CCIC) dataset, Sharif University of Technology Crack (SUT-Crack) dataset, and bridge crack detection dataset. The model obtained an average F1-score of 99.98%, 98.75% and 100% on the above-mentioned datasets respectively. We also tested the model on our own dataset that has pavement images with finer cracks which are harder to detect. Our dataset consists of 5,506 images in total. The model obtained 89.13% average F1-score. We explored the transfer learning technique on the proposed model as well. We fine-tuned the model (pre-trained on CCIC dataset) on our dataset which improved the results by 6.34% and increased the average F1-score to 95.47%. The model shows good generalizability with just a little over a million trainable parameters. Model explainability analysis using Captum AI library confirmed that the model is focusing on the cracked sections of the pavement images to classify the images as cracked. Having relevant areas of the image contribute more toward the prediction increases the confidence in the model.
Lightweight Deep Convolutional Neural Network for Pavement Crack Recognition with Explainability Analysis
Lect. Notes in Networks, Syst.
Abdelgawad, Ahmed (editor) / Jamil, Akhtar (editor) / Hameed, Alaa Ali (editor) / Gulfam, Muhammad (author) / Seals, Cheryl D. (author) / Vargas-Nordcbeck, Adriana (author) / Dozier, Gerry V. (author)
International Conference on Intelligent Systems, Blockchain, and Communication Technologies ; 2024 ; Sharm El-Sheikh, Egypt
2025-03-05
15 pages
Article/Chapter (Book)
Electronic Resource
English
Recognition of asphalt pavement crack length using deep convolutional neural networks
| Taylor & Francis Verlag | 2018
Automatic classification of pavement crack using deep convolutional neural network
| Taylor & Francis Verlag | 2020
Vision-Based Crack Detection of Asphalt Pavement Using Deep Convolutional Neural Network
| Springer Verlag | 2021