Publication: Embedded deep convolutional neural network for uav-based roads inspection
| dc.contributor.author | Lee, Li Ping | |
| dc.date.accessioned | 2024-02-19T04:00:40Z | |
| dc.date.available | 2024-02-19T04:00:40Z | |
| dc.date.issued | 2020-08-01 | |
| dc.description.abstract | Roads are the conduit of life’s activities, however, a variety of factors cause the roads to deteriorate gradually such as crack and get risk in terms of road maintenance and road traffic safety. The most common road inspection is done by a manual process by humans. Nevertheless, it can be time consuming and laborious. Various attempts have been made to automate road inspection such as by using image processing method. However, this method is not applicable for all road condition. Besides that, sensing technologies are too expensive. Hence, this project proposed a deep Convolutional Neural Network (CNN) to perform the road inspection task and deploy it on Raspberry Pi. This project is divided into 5 stages which are data collection, model training, model validation, model testing and deployment of trained model into Raspberry Pi, a lightweight and low-cost embedded computing board with high reliability. A total of 669 images are extracted from the video captured by using a drone. Then, a deep CNN is employed to carry out the training and testing processes on three different CNN architectures; Architecture 1, Architecture 2 and Architecture 3. Based on the results, Architecture 3 CNN model with 7 layers has the best performance for road crack detection with accuracy of 95%, precision of 93%, recall of 94.6% and F1-score of 93.8%. After that, the model is deployed onto Raspberry Pi. It is proved that the model can perform the road crack detection at the speed of 3 frames per second with a good accuracy of 94.3%, precision of 96.9%, recall of 93.4% and F1-score of 95.1%. Overall, Architecture 3 designed in this project proved its great potential to be deployed for crack road inspection. | |
| dc.identifier.uri | https://erepo.usm.my/handle/123456789/18383 | |
| dc.language.iso | en | |
| dc.title | Embedded deep convolutional neural network for uav-based roads inspection | |
| dc.type | Resource Types::text::report | |
| dspace.entity.type | Publication | |
| oairecerif.author.affiliation | Universiti Sains Malaysia |