Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Deep Convolutional Neural Networks for Automated Road Damage Detection
The well-maintained roads are crucial for the safety and convenience of both drivers and pedestrians. The challenges faced by civil engineers are effectively addressing pavement damage which is of utmost importance. To handle this issue cost-effectively, various methods have been proposed. In this study, we present a deep CNN, ResNet50, ResNet101, and ResNet152 model to classify the pavement damages them accurately. This methods are trained on RDD 2022 (road damage detection) dataset which consists of 47,420 images captured from 6 different countries like India, the Czech Republic, Japan, Norway, China, and United States having 55,000 annotations of road damage. To address the issue of imbalanced data distribution across different classes, data augmentation techniques like contrast transformation, Gaussian blur, and brightness adjustment as preprocessing steps done prior to training. Experimental results exhibit the F1-score value of 0.4773, 0.5123, 0.6435, 0.6954 value for deep CNN, ResNet50, ResNet101, and ResNet152 models, respectively. The results obtained demonstrate that ResNet152 gave better results. A successful implementation of this study can enable the immediate repair of pavement and better civil infrastructure monitoring.
Deep Convolutional Neural Networks for Automated Road Damage Detection
The well-maintained roads are crucial for the safety and convenience of both drivers and pedestrians. The challenges faced by civil engineers are effectively addressing pavement damage which is of utmost importance. To handle this issue cost-effectively, various methods have been proposed. In this study, we present a deep CNN, ResNet50, ResNet101, and ResNet152 model to classify the pavement damages them accurately. This methods are trained on RDD 2022 (road damage detection) dataset which consists of 47,420 images captured from 6 different countries like India, the Czech Republic, Japan, Norway, China, and United States having 55,000 annotations of road damage. To address the issue of imbalanced data distribution across different classes, data augmentation techniques like contrast transformation, Gaussian blur, and brightness adjustment as preprocessing steps done prior to training. Experimental results exhibit the F1-score value of 0.4773, 0.5123, 0.6435, 0.6954 value for deep CNN, ResNet50, ResNet101, and ResNet152 models, respectively. The results obtained demonstrate that ResNet152 gave better results. A successful implementation of this study can enable the immediate repair of pavement and better civil infrastructure monitoring.
Deep Convolutional Neural Networks for Automated Road Damage Detection
Smart Innovation, Systems and Technologies
Peng, Sheng-Lung (Herausgeber:in) / Mondal, Ayan (Herausgeber:in) / Kagita, Venkateswara Rao (Herausgeber:in) / Sarkar, Joy Lal (Herausgeber:in) / Rakshitha, R. (Autor:in) / Srinath, S. (Autor:in) / Kumar, N. Vinay (Autor:in) / Rashmi, S. (Autor:in) / Poornima, B. V. (Autor:in)
International Conference on Advanced Communications and Machine Intelligence ; 2023 ; Warangal, India
Proceedings of International Conference on Advanced Communications and Machine Intelligence ; Kapitel: 12 ; 155-165
06.12.2024
11 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Deep Learning‐Based Crack Damage Detection Using Convolutional Neural Networks
| Online Contents | 2017
Convolutional neural networks for automated damage recognition and damage type identification
| Wiley | 2018