Automatic updating and verification of road maps using high-resolution remote sensing images based on advanced machine learning methods
- Publication Type:
- Thesis
- Issue Date:
- 2022
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
One of the significant objects among urban features is road network. Automatic updating road network from high-resolution remote sensing imagery (HRSI) is a major application in the field of remote sensing and geospatial information systems (GIS), which has a significant role in various purposes such as GIS maps updating, urban cover change detection, emergency tasks, navigation, etc. Nowadays, obtaining accurate information of road networks using various supervised and unsupervised classification approaches from HRSI is a challenging task. Recently, deep learning (DL) techniques have obtained efficient performance in the field of remote sensing images processing. Therefore, this thesis investigated the state-of-the-art deep convolutional neural networks (DCNNs) for automatic updating and verification of road network from high-resolution remote sensing images.
Firstly, in objective 1, I solve the issues of conventional ML methods by implementing robust DCNN approaches for road surface segmentation from different HRSI. The presented networks are implemented to the various remote sensing datasets for road surface segmentation and compared with other state-of-the-art deep learning-based networks, which the results prove the superiority of the proposed networks in the road segmentation task.
Secondly, in objective 2, I propose a shape and connectivity-preserving road identification deep learning-based architecture called SC-RoadDeepNet to overcome the discontinuous results and road shape and connectivity quality of most of the existing road extraction techniques. The proposed model comprises a new measure based on the intersection of segmentation masks and their (morphological) skeleton called connectivity-preserving centerline Dice (CP_clDice) that aids the model in maintaining road connectivity. The qualitative and quantitative assessments demonstrate that the proposed SC-RoadDeepNet can produce high-resolution results, particularly in the area of road network completeness.
Thirdly, in objective 3, I present a new automatic deep learning-based network named road vectorization network (RoadVecNet), which comprises interlinked UNet networks to simultaneously perform road segmentation and road vectorization with achieving important information such as width/length and location of the road network. Particularly, RoadVecNet contains two UNet networks. The first network can obtain more coherent road segmentation maps and the second network is linked to the first network to vectorize road networks. Classification results indicate that the RoadVecNet outperforms the state-of-the-art deep learning-based networks. In short, the proposed methods and the outcomes (high quality and accurate road network data) of the study has high potential in environmental applications such as land use change detection in urban areas, and emergency tasks and also commercial value in navigation and road maps updating.
Please use this identifier to cite or link to this item: